Medical Imaging

MEDICAL IMAGING
Edited by Okechukwu Felix Erondu
Medical Imaging
Edited by Okechukwu Felix Erondu

Published by InTech
Janeza Trdine 9, 51000 Rijeka, Croatia

Copyright 2011 InTech
All chapters are Open Access distributed under the Creative Commons Attribution 3.0
license, which allows users to download, copy and build upon published articles even for
commercial purposes, as long as the author and publisher are properly credited, which
ensures maximum dissemination and a wider impact of our publications. After this work
has been published by InTech, authors have the right to republish it, in whole or part, in
any publication of which they are the author, and to make other personal use of the
work. Any republication, referencing or personal use of the work must explicitly identify
the original source.

As for readers, this license allows users to download, copy and build upon published
chapters even for commercial purposes, as long as the author and publisher are properly
credited, which ensures maximum dissemination and a wider impact of our publications.

Notice
Statements and opinions expressed in the chapters are these of the individual contributors
and not necessarily those of the editors or publisher. No responsibility is accepted for the
accuracy of information contained in the published chapters. The publisher assumes no
responsibility for any damage or injury to persons or property arising out of the use of any
materials, instructions, methods or ideas contained in the book.

Publishing Process Manager Marija Radja
Technical Editor Teodora Smiljanic
Cover Designer InTech Design Team
Image Copyright Kirsty Pargeter, 2011. DepositPhotos

First published December, 2011
Printed in Croatia

A free online edition of this book is available at www.intechopen.com
Additional hard copies can be obtained from orders@intechweb.org

Medical Imaging, Edited by Okechukwu Felix Erondu
p. cm.
ISBN 978-953-307-774-1

free online editions of InTech
Books and Journals can be found at
www.intechopen.com

Contents
Preface IX
Part 1 Basic Concepts in Medical Imaging 1
Chapter 1 Identification of Structures in Medical Images 3
Marina de S Rebelo, Srgio Shiguemi Furuie,
Lincoln de Assis Moura Jr, Eduardo Tavares Costa
and Marco Antonio Gutierrez
Chapter 2 Detection of Abnormalities in a Biological Tissue by
Diffuse Optical Tomography: A Computational Study 27
H. Trabelsi, M. Gantri and E. Sediki
Chapter 3 3D Ultrasound Image Segmentation:
Interactive Texture-Based Approaches 43
Julien Olivier and Ludovic Paulhac
Chapter 4 Radiation Protection in Medical Imaging 67
Horaiu Colosi, Dan Colosi, Vlad Murean and Marius Roman
Part 2 Image Processing Techniques 87
Chapter 5 Current Trends in Archiving and Transmission
of Medical Images 89
Lus S. Ribeiro, Carlos Costa and Jos Lus Oliveira
Chapter 6 Large-Scale User Facility Imaging and Scattering Techniques
to Facilitate Basic Medical Research 107
Stephen D. Miller, Jean-Christophe Bilheux, Shaun S. Gleason,
Trent L. Nichols, Philip R. Bingham and Mark L. Green
Chapter 7 Tools to Improve the Patients Processes
at Imaging Centers 135
Liliana Neriz and Francisco J. Ramis
VI Contents

Chapter 8 High to Microwave Frequencies Imaging Techniques 147
George A. Kyriacou, Ilias N. Aitidis, Dimitrios G. Drogoudis
and John N. Sahalos
Chapter 9 A Mutual Information-Based Image Quality Metric
for Medical Imaging Systems 195
Du-Yih Tsai, Eri Matsuyama and Yongbum Lee
Part 3 Applications in Clinical Settings 213
Chapter 10 Diagnostic Imaging in Oral and Maxillofacial Pathology 215
Hasan Ayberk Altug and Aydin Ozkan
Chapter 11 Fast MRI Methods for the Clinical Evaluation
of Skeletal Disorders 239
Renato Toffanin, Giuseppe Guglielmi and Maria A. Cova
Chapter 12 Determination of Cardiac Ejection Fraction by
Electrical Impedance Tomography 253
Franciane C. Peters, Luis Paulo da S. Barra
and Rodrigo W. dos Santos
Chapter 13 Assessment of Human Skin Microcirculation and Its
Endothelial Function Using Laser Doppler Flowmetry 271
Helena Lenasi
Chapter 14 Determination of Limb Hemodynamics During Rhythmical
Muscle Contractions Assessed by Doppler Ultrasound 297
Takuya Osada and Gran Rdegran
Part 4 Advances in Medical Imaging Techniques 309
Chapter 15 Current State of the PET/CT Hybrid Technique
and Clinical Indications 311
Patricia Carreo-Morn and Mara de las Nieves Gmez Len
Chapter 16 Advances in Medical Imaging
Applied to Bone Metastases 339
ngel Gonzlez-Sistal, Alicia Baltasar Snchez,
Michel Herranz Carnero and lvaro Ruibal Morell
Chapter 17 A Hierarchical Framework for Mitosis Detection in
Time-Lapse Phase Contrast Microscopy Image Sequences
of Stem Cell Populations 355
Anan Liu, Kang Li and Tong Hao
Chapter 18 Safety of Interactive Image-Guided Surgery 373
Alain Beaulieu
Contents VII

Preface
Medical imaging was primarily seen as the use of non-invasive techniques to create
internal images of the human body for clinical or medical purposes. This rapidly
evolving field of medicine originated in the first decade of the 19
th
century, when
Wilhelm Rntgen, a professor of physics at Wrzburg University in Germany,
discovered electromagnetic radiation. After the World War Two, the development of
computer technology has triggered an amazing revolution in medical imaging
techniques.
What we know about and do with medical imaging has changed tremendously over
the past decade, beginning with the basics, following with the breakthroughs, and
moving on to the abstract.
There is a continuous drive not only to improve the diagnostic yield of medical
imaging techniques for clinical use, but also the management of the huge amount of
digital information available to medical imaging departments. Today there is an
increasing interest to harness medical imaging processes to improve research in the
biophysical domain.
Medical imaging has experienced tremendous progress during the past decade.
Continuous innovations have been fueled by the need to improve diagnostic yield and
achieve fast turnaround through robust information management, which should
ultimately improve patient outcome. The rapid changes in digital technology have also
created opportunities for the development of high-tech equipment employed in
medical imaging practice.
Incidentally, these changes tend to revolve around key modalities such as
radiography, computed tomography, scintigraphy, magnetic resonance imaging and
ultrasound. In many instances, one modality tends to augment the limitations
associated with others. This book approaches the concept of medical imaging from a
wider perspective not considered by many authors in this area. It consists of four
sections and eighteen chapters and topics.
The first section deals with basic concepts such as identification of structures in
medical images, 3D image segmentation, image analysis, texture analysis, radiation
protection and simulations.
X Preface

Section two concentrates on the image processing techniques, quality metrics as well
as tools required to optimize patient processes.
Section three deals with applications of medical techniques in clinical settings, while
the last section explores the advances made in this area including PET/CT hybrid
technique, optical imaging thermography, mitosis detection, phase contrast
microscopy and use in image-guided surgery.
It is hoped that this book bridges the gap between the core clinical books reserved for
medical diagnosis and the wide ranging possibilities in the field of biological research.
It is a must-read for all practitioners and researchers interested in medical imaging.
This book is an attempt to present the old, understand the current and appreciate the
great advancements into the future of medical imaging. It will be an invaluable
resource to those interested in history, clinical applications and unlimited horizons
presented by medical imaging science.
My acknowledgement goes to all the authors, who have made scholarly contributions
to the book from diverse backgrounds. I am also grateful and indebted to the
publishing process managers Bojana Zelenika and Marija Radja for carefully directing
the editorial process and for their patience even when I couldnt meet the scheduled
timeline.

Felix Okechukwu Erondu Ph.D
Adjunct Senior Lecturer in Radiography and Medical Imaging
University of Nigeria
Nigeria

Part 1
Basic Concepts in Medical Imaging

1
Identification of Structures in Medical Images
1

Marina de S Rebelo
1
, Srgio Shiguemi Furuie
2
, Lincoln de Assis Moura Jr
2
,
Eduardo Tavares Costa
3,4
and Marco Antonio Gutierrez
1

1
Heart Institute (InCor) - University of So Paulo Medical School
2
Polytechnic School University of So Paulo
3
School of Electrical and Computer Engineering- University of Campinas
4
Center for Biomedical Engineering - University of Campinas
Brazil
1. Introduction
The development of automatic systems for medical image processing, which can effectively act
as an agent to aid medical diagnosis, is a goal that has been pursued by researchers since the
first works on the field of medical image processing in the 80s. The automation of image
analysis tasks can produce very interesting results such as less time spent by specialists,
decrease of intra-and inter-observer differences, second opinions to non-specialists and in
educational systems. Any automatic system deployed for analysis and visualization of images
involves the identification of objects and often the relationships among them. The automatic
identification of structures is a research area of image processing that still has great challenges.
In tasks that involve primarily activities of calculation, the computer's processing power is
incomparably higher than the humans. However, in recognition and analysis tasks, the human
brain possesses a strong ability that is not obtained by any computational system. For a given
scene, humans possess a unique ability to distinguish the objects that are significant and,
among them, those that represent the focus of interest for a particular situation.
In the routine of medical image analysis, doctors often deal with images in which the visual
perception is not sufficient for identification of some particular structure under study. They
need, therefore, conceptual hypotheses so as to perform this task. At each new image
analyzed a doctor uses a huge amount of knowledge accumulated over the years of clinical
practice. This is a typical feedback and a case base reasoning system, as each new image
analyzed gives the doctor more knowledge about a specific problem. A system of medical
aid for identification and analysis of structures should also be able to incorporate and use
knowledge about the problem domain.
2. Identification of structures in digital images
2.1 Problem statement
Automatic recognition and identification of structures are fields of Computer Vision. The
recognition of structures is a difficult task because many factors influence its computation.

1
Portions adapted, with permission, from Computers in Biology and Medicine 37 (2007) 1183 1193

Medical Imaging

4
These can include restrictions on permissible forms, the semantics of the context of the scene
and the information present in the image itself (Suetens et al, 1992). After recognizing the set
of objects in the scene, the next step in the interpretation process is the identification of one
or more objects of particular interest in the recognized set. There is a distinction between the
concepts of recognition and identification of structures. There are many definitions in the
literature for recognition (or discovery) of structures. We use here a simple definition, given
by Suetens et al (1992):
The recognition (or detection) of structures is the task of finding and labeling parts of a
digital image that correspond to objects in the scene.
Once the recognition step has been performed, the next step is to identify one or all
instances of a particular structure. The definition of object identification is not precise and,
depending on the application, may be mistaken for recognition. In this text, identification is
defined by the authors as:
The identification of a particular set of structures is to set apart from the recognized
structures the one or ones that correspond to the object under study.
The term segmentation is sometimes used by researchers as a synonym to identification as
defined above. The canonical definition of segmentation is the subdivision of the image into
its components (Gonzales & Woods, 2000; Jain, 1989). Strictly speaking, segmentation
corresponds only to the first step of the recognition process.
2.2 Automatic identification
The steps involved in the automatic identification of structures are usually divided into
different levels. Some authors have proposed a division into two levels: low and high
(Ballard & Brown, 1982; Sonka et al, 1998) while others use three processing levels: low,
medium and high (Gonzales & Woods, 2000; Niessen, 1997).
The low processing level deals with functions similar to automated human visual reactions
that normally require little or no knowledge about the content of the image (Gonzalez &
Woods, 2000; Sonka, 1998). It is mainly or almost entirely based on intrinsic image data,
without inclusion of a priori information. It includes activities such as the basic process of
image formation, filtering for noise reduction and correction of acquisition artifacts. In
addition, there are processes to decrease the huge amount of data present in the original
image to obtain parametric images describing relevant information. Examples of relevant
parameters are local contrast, texture, curvature and movement. At the intermediate
processing level the tasks are performed by extraction and characterization of components
in the images resulting from the low level process. This stage includes activities such as
identification, representation and description of the image information through models. For
authors who do not use this classification, these functions are usually defined as low-level
processing (Ballard & Brown, 1982; Sonka, 1998). The authors that include the intermediate
level point out that the difference to the low level processing is that, based on some
restrictions, it is possible to select - or highlight - relevant information to a particular
application and formalize this information using models that include some information a
priori about the application domain, thus making it easier the subsequent high-level tasks
(Niessen, 1997). Moreover, the result of this processing step generates no more pictures, but
representation of objects in the picture. The high processing level is based on knowledge,


5
goals and strategies on how to achieve those goals more efficiently. Using knowledge about
the image structure, image data is systematically evaluated. Hypotheses are confirmed or
rejected by the data. Routines for processing high-level information should be organized to
produce rapid searches. The difference between high-level and intermediate level
processing is the explicit use of knowledge about the context of the image under analysis.
3. Automatic identification in medical images
In medical applications, the ultimate goal of automatic image analysis is to obtain not only
qualitative results, but also quantitative data for assessing objectively the changes in relation to
normal standards. To obtain such results it is necessary first to identify the object of interest
and then carefully analyze it. Among the applications for which identification of structures is
an important step are: visualization and surgical planning, planning for radiotherapy
procedures, identification of tumors, malformations and dysfunctions, classification of organs,
tissues and cells, and retrieval of similar cases in image content based databases.
In the case of image volumes, or 3D image data, automatic identification involves a large
number of operations starting from the original acquired data: they must be processed so
that object slices are formed by applying filters and reconstruction algorithms. Once the
slices have been obtained, further transformations in the image are needed in order to make
the data more suitable for identification. Such transformations may include processes as
edge detection algorithms and structure recognition. Then high level procedures involving
the use of a priori knowledge to determine the searched structure are used and, specific
analysis of the found structures can be performed
The results of methods for identification of structures in medical images can be affected by
image artifacts. Images contain noise that can alter the intensity of a pixel such that its
classification becomes uncertain. They also have finite pixel size and, for this reason, are
subject to partial volume averaging effect where individual pixel volumes contain a mixture
of tissue classes so that the intensity of a pixel in the image may not be consistent with any
one structure. The intensity level of a single tissue class can vary over the extent of the
image, and internal materials (for instance surgical clips) may cause distortion in the imaged
organ (Rani et al, 2011; Withey & Koles, 2007). Some of these effects are depicted in Figure 1.

Fig. 1. Examples of effects that can affect structure identification can be seen in the three
images: (A) kidney in Ultrasound (US), (B) lung in Nuclear Medicine (NM) and (C) brain in
Computed Tomography (CT). The noise associated with each image modality has to be dealt
with in different ways. In image (C) the presence of dental metal pieces degraded the
reconstructed slice of the brain (images acquired at the Imaging Department - Heart
Institute HCFMUSP- with patient anonymization).

Medical Imaging

6
All these factors may affect the identification of organs due to poorly defined boundaries,
blur or weak edges homogeneity. Moreover, the identification of organs in medical imaging
presents some unique challenges as the creation of efficient models for representation of
biological structures that are not easily described by mathematical models. All these factors,
added to the variability in tissue distribution among individuals in the human population,
means that some degree of uncertainty must be attached to all segmentation results. Figure 2
exemplifies the difficulty in identifying brain structures in Computerized Tomography (CT)
and Magnetic Resonance (MR) images.

Fig. 2. Example challenging identification tasks in brain images. The areas inside the red
circles are the areas to be identified. Due to the aforementioned problems, the identification
of some tissues in medical images can be difficult even in modalities with high resolution as
CT (left image) and MR (middle and right) (images acquired at the Imaging Department -
Heart Institute HCFMUSP- with patient anonymization).
3.1 Use of a priori information for identification in medical images
In the context of medical images, the concept of identification, as defined above, involves
the discrimination against a particular organ or the discrimination between normality and
abnormality for a particular organ, based on the attributes of the image. Although trained
clinicians are able to find and classify body structures precisely, there is not a technique
that can reproduce precisely their ability in computational environments. The use of
simple methods that rely only on information contained in the image data, in general,
does not produce satisfactory results. In other words, the intrinsic information of the
image is not enough to characterize the structures present in real objects (Suetens et al,
1992). Effective procedures for identifying structures should incorporate knowledge of the
model or the context.
There are several studies in the literature aiming the characterization of the knowledge
about a particular organ in a medical image modality to aid the automatic processing of the
image. Typically, the methods use information about shape, size, texture and position of a
given structure. A major difficulty in using a priori information about the forms of the
organs is to characterize them geometrically. Unlike the rigid objects from other applications
of computer vision, most human organs have few calculable and stable invariants. Apart
from that, the features to be detected in images may be small and subtle, sometimes
characterized only by tiny changes in gray level. In many medical applications there is a
significant amount of known information about the human anatomy. However, there is a
major difficulty involving the creation of models to represent biological structures,
comprising very complex forms that are not described easily. Furthermore, for a given
population these structures usually show variations both in form and size.


7
3.2 A review of applications in medical imaging
Techniques specifically designed for computerized detection of abnormalities in chest
radiographs were among the first works in the area of identification of structures in medical
images (Duncan & Ayache, 2000). The field has evolved continuously since then and the
amount of knowledge has grown constantly with a huge number of methods described in
the literature, as pointed out in many survey papers (Bandyopadhyay, 2011; Duncan &
Ayache, 2000; Kirbas & Quek, 2002; Lesage et al, 2009; Ma et al, 2008; Oliver et al, 2010;
PetitJean & Dacher, 2011; Witney & Koles,2007).
The assumptions and requirements taken by identification methods vary substantially for
different body organs and imaging modalities. Therefore, the choice of the method will be
dependent on the particular application, regarding the organ or system under study, as well
as the imaging modality (Bandyopadhyay, 2011). There have been described general
methods that intended to be more comprehensive, and therefore can be applied to a number
of applications. However, their performance is worse than that of methods designed for
specific applications, as the latter generally use a priori information available about the
problem under study (Bandyopadhyay, 2011).
In the next paragraphs we will present some methods proposed in recent years. A general
overview of the methods described for medical image identification in the last decades is
presented in the first section. The next sections are dedicated to clinical applications with
emphasis on the organ under study, presenting the well-known specific challenges for
identifying the particular organ and exploring some recently published methods. This text
avoids performance comparisons due to fact that existing applications are so different in
their purposes and practical applications, that direct comparisons do not make sense.
3.2.1 Evolution of the methods along the last decades
There is a number of papers discussing the evolution of the algorithms for extraction or
identification of medical structures. It is out of the scope of this chapter to present and
compare all studies and methodologies for classifying the existing methods. Instead, we
present a description of the field evolution based on two surveys, presented by Ma et al
(2008) and Witney & Koles (2007), which we believe presents the evolution of the field in a
simple and informative way. For the other reviews, readers are referred to the literature
(Duncan & Ayache, 2002; Petitjean et al, 2011). Withey and Koles (2007) classified the
medical image identification approaches in literature into three generations, while Ma et al
(2008) classified the methods in three categories that can be fitted to the three generations
model of Witney and Koles.
The first generation techniques are those almost based on intrinsic image information,
without any a priori information. They include:
- Threshold: algorithms belonging to this type assume that the searched structures are
related to clearly quantifiable features as image intensity and gradient.
- Region growing: from an initial seed located in the image, adjacent pixels are checked
against a set of predefined homogeneity criteria. Pixels that meet the criteria are
included in the region.
- Edge tracing: after the application of an edge detection algorithm, the edge pixels with
adjacent neighbor connectivity are followed sequentially and collected into a list to
represent an object boundary.

Medical Imaging

8
The second generation is composed of algorithms using image models, optimization
methods, and uncertainty models, and includes:
- Pattern recognition and clustering: as certain structures in medical images can be
treated as patterns, segmentation algorithms that combine pattern recognition
techniques have been proposed to extract the searched structures.
- Deformable models: deformable models use curve evolution to perform segmentation.
A moving equation should be defined to drive the initial curves to the right structure
boundaries.
- Graph search: image pixels are used to form nodes in a graph and the nodes are
interconnected to neighbors, mapping the corresponding pixel associations in the
image. Costs are assigned for each interconnection. Algorithms from combinatorial
optimization are used to obtain minimum-cost solutions.
- Multiresolution methods: multiresolution or multiscale analysis refers to the use of
scale reduction to group pixels into image objects. A stack of images is formed by
recursively reducing the scale of the original image by blurring followed by down
sampling. The central idea in this category is that there may be scales that are more
suited for processing some image features.
The third generation is characterized by algorithms that are capable of incorporating
knowledge, such as:
- Shape and appearance models: the active shape model (ASM) was inspired by
deformable models with the added intention of limiting the extent of the model
deformation. A statistical representation of an object is formed by identifying a set of
landmark points on an object boundary and analyzing the variation of each across a set
of training images.
- Atlas-based: it is a generic technique for automatic delineation of structures in
volumetric images, which starts by registering an anatomical image from an atlas with a
target image to be segmented. A critical underlying assumption is that it is possible to
find a deformation that aligns the atlas with the target image in such a way that label
propagation lines up the objects of interest.
- Rule-based: image primitives are usually derived from first-generation and second-
generation algorithms and then interpreted using anatomical and image knowledge
applied as a set of rules.
3.2.2 Brain applications
Environmental factors, age and disease can affect neuroanatomical structures during the
process of ageing. These structures are also affected by genetic factors in patients with
degenerative diseases (Ashburner et al, 2003). Besides providing anatomical information,
medical imaging procedures are able to provide extremely relevant information about brain
physiology, which can be used to understand physical and psychological clinical conditions.
In order to capture the extraordinary morphological variability of the human brain, a
number of automatic methods for identification and analysis of structures have been
developed for different imaging modalities (Ashburner et al, 2003; Bandyopadhyay, 2011;
Bresser et al, 2011; Ishii et al, 2009; Lopes et al, 2008; Ribbens et al, 2010; Rousset et al, 2007;
Shiee et al, 2010; Tu & Bai, 2010; Yi et al, 2009; Zhang et al, 2011).


9
Most automatic systems for analysis of brain images have three processing steps: (i) brain
tissue identification; (ii) registration: the voxels of interest are matched to a template or to an
earlier scan from the same individual; (iii) statistical comparison of different groups of
patients or same patient in different times.
In their review of the methods to assess brain structures, Ashburner et al (2003) discuss the
strengths and limitations of algorithms for identification and registration of brain images,
together with evidence of their usefulness at the clinical and research level. A more recent
review of current methods used for computer automated identification in brain anatomical
images is presented by Bandyopadhyay (2011). Most methods for extraction of brain
structures and tissue segmentation use some kind of a priori information. Some authors,
however, try to minimize the need of a priori knowledge. As an example, Scherrer et al
(2009) extracted tissue and sub-cortical structures in MR images by considering a local
approach to cope with intensity non-uniformity and a multi-agent based implementation.
3.2.3 Cardiovascular applications
Cardiovascular diseases are the leading cause of death in developed countries (Allender
et al, 2008). Imaging techniques are nowadays among the most valuable tools in helping
diagnosis, treatment and follow-up of cardiovascular pathologies, as they provide
qualitative and quantitative information of the heart, which would be impossible to
obtain otherwise (Cordero-Grande et al, 2011). Diagnosis and treatment follow-up of these
pathologies can rely on numerous cardiac imaging modalities, which include echography,
CT (computerized tomography), digital subtractive angiography, coronary angiography
and cardiac MRI.
Heart. Quantitative analysis of the heart structure and function can be performed by the
extraction of cardiac descriptors from medical images. These descriptors can provide
information about global and local function. This includes estimating left ventricle volume,
ejection fraction, cardiac output, myocardial mass, myocardial morphology, myocardial
uptake rates, tissue viability, blood flow, local motion and deformation (wall thickening).
The determination of most of these descriptors involves an appropriate 3D identification of
cardiac structures being the cardiac chambers, cardiac muscle or cardiac valves - as a
prerequisite (Alattar et al, 2010; Cordero-Grande et al, 2011). The imaging modality of choice
will depend on the problem under study and the complexity of the identification of the
cardiac structures will depend on the modality chosen.
Automatic isolation of the heart structures in MR images is a challenging task because of the
inherent noise associated with cine MRI, caused by factors such as patient movement,
cardiac dynamics, and complex intensity characteristics (Lynch, 2008). Even in noiseless
cases, MR images may present blurred edges due to the partial volume effect and can be
affected by motion artifacts. In addition, low contrast between the myocardium and
surrounding tissues complicates the identification of the cardiac structures. As an example,
the papillary muscles, the trabeculae, or the liver have pixel intensity values which are
virtually indistinguishable from those of the myocardium (Cordero-Grande et al, 2011). In
the literature there is a huge number of recent works on heart identification in MR images
(Alattar et al, 2010; Cordero-Grande et al, 2011; Lekadir et al, 2011; OBrien et al, 2011;
ODonnell et al, 2006; Rouchdy et al, 2007; Schaerer et al, 2010; Zhao et al, 2009). A

Medical Imaging

10
comprehensive survey focusing on extracting the heart cavities in short axis view from
cardiac MR images has been published recently by Petitjean & Dacher (2011).
The analysis of cardiac conditions using CT images needs a highly consistent identification
of the involved surfaces. To measure blood volume, segmentation should consistently
follow the cavity border. Wall thickness measurements should be based on a convex hull
around the cavity that excludes papillary muscles from the myocardium. For the
epicardium, it is important to avoid local confusions with the nearby pericardium or lung
transition (Peters et al, 2010). A number of authors have described methods to isolate the
cardiac walls and chambers in CT images (Ecabert et al, 2008; ODonnell et al, 2006; Peters et
al, 2010; Zheng et al, 2010).
Ultrasound imaging is arguably the hardest medical imaging modality upon which to perform
organ identification (Noble, 2010). The main problem in detecting, for example, heart
boundaries in ultrasound images is related to the high level of multiplicative noise (mainly
speckle noise), low contrast among structures, artifacts such as shadowing from the lungs and,
attenuation. All these factors can hinder the automatic analysis of the images (Antunes et al,
2010). A recent work by Mora et al (2009) combined supervised neural networks with different
image processing techniques to identify heart cavities. Antunes et al (2010) applied a geometric
deformable model to identify the four heart chambers of newborns.
Vascular system. Vascular identification is a particular challenging problem. The network
trees of vessels are complex structures, with high variability in size and curvature. The general
tubular organ geometry can be perturbed by stents, calcifications, aneurysms, and stenosis,
besides being embedded in complex anatomical scenes, surrounded by other organs. Lesage et
al (2009) published a review on vessel lumen segmentation techniques focused on 3D contrast-
enhanced imaging modalities (Magnetic Resonance Angiography - MRA and Computed
Tomography Angiography - CTA). A former survey on vessel extraction techniques was
presented by Kirbas and Kerk (2002). Destrempes et al (2009) describe a method for extracting
carotid artery in ultrasound images. A segmentation and visualization of the left coronary
heart arteries in CT images was presented by Rahman et al (2010). The arteries extraction
methods tend to use a strong cylinder assumption. Qian et al (2009) presented a method in
which they try to avoid this common strong prior knowledge. The analysis of temporal change
in vessel images was addressed by Zhao et al (2009) who described a method to identify
subjects with connective tissue disorder in 4D cardiovascular MRI.
3.2.4 Liver applications
Precise measurement of shape and composition of liver is the basis for diagnosis, surgery
planning and therapy control of liver pathologies such as cirrhosis, liver cancer, and
fulminant hepatic failure (Campadelli et al, 2009). CT images are preferred by the doctors
for performing these tasks. Liver identification in CT scans poses problems due to the low
contrast and blurred edges that characterize CT images. In their survey of liver extraction in
CT scans, Campadelli et al (2009) identified the main applications of the technique as: (1)
automatic detection of liver cancer from other liver diseases; (ii) measurement of liver
volume - an important index in cases of living donor liver transplantation; (iii) 3D volume
rendering of abdominal organs for surgical planning and radiation treatment programs.
Another review of the methods for liver identification in CT images was published by
Heimann et al (2009). Recent works in this area have also been described by Rusk et al


11
(2009), Masuda et al (2010) and Pu et al (2009). Linguraru et al (2009) proposed a method to
isolate both the spleen and liver in contrast-enhanced CT images by first aligning the images
with models from an atlas and then improving the results by an active contour technique.
3.2.5 Lung applications
Pulmonary nodules are potential manifestations of lung cancer, and their detection and
inspection are essential for screening and diagnosis of the disease (Kubota et al, 2011). Once
the nodule has been detected, monitoring of its size, density, edge-smoothness and growth
rate provide information about what treatment, if any, is appropriate (Murphy et al, 2009).
An extensive review of computer aided diagnosis in chest radiography was published by
van Ginneken et al (2001). Automated detection of nodules has been described by Murphy
et al (2009) and Kubota et al (2011). Bouma et al (2009) proposed a system for the automatic
detection of pulmonary embolism in contrast-enhanced CT images and Pu et al (2011)
proposed an automated scheme to extract the airway tree depicted on CT images.
3.2.6 Other applications
Mammography. In mammography, the use of computer-aided diagnosis (CAD) systems is
intended to assist radiologists in the automatic detection and classification of
mammographic abnormalities. There is a large number of different types of mammographic
abnormalities. In the majority of cases the abnormalities are either micro-calcifications or
masses (Oliver et al, 2010). Mammography CAD systems usually start by a preprocessing
step to extract the breast tissue, background tissue and pectoral muscle (Cheng et al, 2003;
Samulski & Karssemeijer, 2010; Timp et al, 2007). Oliver et al (2011) published an extensive
review on the theme in which they divide the automatic methods for detection of
abnormalities in three groups: (i) Region-based methods: region growing and related
methods, watershed methods, split and merge methods; (ii) Contour-based methods; (iii)
Clustering methods.
Thyroid. Diseases of the thyroid gland are among the most frequent endocrine disorders.
Ultrasound has become the most important technique for thyroid gland imaging and
computerized systems have been described to aid doctors in the task of thyroid image
analysis. Hegeds (2004) and Kollorz et al (2008) presented methods to automatically isolate
the thyroid gland in US images, using geodesic active contour level set based approach.
Kidney. In renal applications Oguro et al (2011) described a system to aid CT-guided
percutaneous cryoablation of renal tumors. Xie et al (2005) used a priori information about
texture and size to segment kidney in ultrasound images. The described method provides the
ability to deal with objects with incomplete boundaries by taking into account their texture.
Meniscal tears. The meniscus is an important part of the knee supporting mechanism. The
knee joint can be severely damaged by a variety of causes, such as arthritis or knee injury.
This can cause pain and inability to walk. In some cases, replacing parts of the joint is the
appropriate course of action (Andra et al, 2008). Meniscal tears are a knee injury that is
common in both young athletes and the aging population. It requires accurate diagnosis
and, if necessary, surgical intervention. Ramakrishna et al (2009) described a novel CAD
diagnostic system for automatic detection of meniscal tears in the knee in MR images. Andra
et al (2008) described a system for pre-operative surgical planning based on structural

Medical Imaging

12
simulation of the bone-prosthesis system. The model of the bone was constructed based on
information from CT datasets.
Radiotherapy planning. In the process of radiotherapy planning the most critical issue is the
identification of the gross tumor volume and its relationship with surrounding tissues in order
to decide the appropriate irradiation angles to minimize non-tumor tissue damage. Zaidi & El
Naqa (2010) published a survey on methods for tumor identification in PET images.
3.2.7 General methods
Kohlberger et al (2009) described a method for general organ segmentation with level sets that
incorporates local intensity statistics and local curvature by means of a point-based tracking
mechanism. Campadelli et al (2010) described a method for extracting abdominal organs in CT
images, using a gray-level based segmentation framework and an a priori anatomical
knowledge that can be adapted to segment different abdominal organs. Noble (2010) made a
review on segmentation techniques in ultrasound images as well as ultrasound tissue
characterization for general applications. Crum et al (2004) reviewed non-rigid image
registration, with emphasis in cardiac and brain applications. Van Rixoort et al (2010)
described a framework for multi-atlas segmentation, in which the most appropriate atlases for
a target image are automatically selected. The framework was applied to segment heart from
non ECG gated volumetric chest CT scans and caudate nucleus from MR brain images.
4. Multiscale methods for identification, some research and examples
The fact that real objects appear different depending on the scale of observation has
important consequences to their description. The concept of scale is very important when
analyzing image data by means of automatic methods (Koenderink, 1984). An object in a
scene can be described in several forms, such as: its edges, its geometrical properties, its
density or color, or its surface area. Ideally, each object has a set of features that set it
aside from the other structures in the scene. Any method for the interpretation of image
content must involve a phase for the detection of those features. Such detection is
performed by the application of some operator to the image data. The application raises
some questions about the type and size of the operators: the result of the operation is
strongly dependent on these choices. The use of multiple scales is a way to overcome the
problem of determining the size of the operator (Yuille & Poggio, 1986; Witkin, 1983). For
optimal detection in a single scale, the size of the structure to be found should be known a
priori. Several multiscale methods have been described in the literature for detection and
classification of tissues and organs (Amira et al, 2008; Bernard & Friboulet, 2009; Gooya et
al, 2008; Ugarriza et al, 2009; Yoon et al, 2008; Yu et al, 2008). In the following section we
present an example of a representation for identification of organ structures that uses
prior information in a multiscale technique.
4.1 Example application
In this example we propose a representation for the identification of medical structures by
using a multiresolution approach, the scale-space. We evaluate the use of a data
representation that allows the inclusion of a priori knowledge about the structures in several
scales and we also develop the idea of an optimal scale to perform the processing.


13
4.1.1 Methodology
Linear scale-space. The linear scale-space concept was introduced in the Western literature by
Witkin (1983) and Koenderink (1984). Weickert (1997) showed that the concept had already
been introduced in Japan approximately 20 years before. One of the main requirements of the
scale-space formalism is that structures at coarser scales in the multiscale representation
should constitute simplifications of corresponding structures at finer scales. This requirement
has been stated in different forms by several authors (Babaud et al 1986; Koenderink, 1984;
Lindeberg, 1994; Yuille & Poggio, 1987). Adding the requisites of linearity and invariance to
spatial displacement, those authors have shown that the Gaussian ( ) ; ( o x g

) is the only filter
that leads to monotonic destruction of detail under consecutive blurring. A blurred version of
the original image ) (x E

is obtained by its convolution with the Gaussian filter, for a specific o
(Equation 1). In Equation 1, ) ; ( o x H

is the intensity of element after convolution. The width o
determines the scale of the output image in each image dimension. In an isotropic condition, o
assumes the same value for all image dimensions. The stack of images, as a function of
increasing scale parameter o, is known as linear scale-space.

x
H x E x g x
g x e
2
2
(2 )
2
( , ) ( ) ( , )
1
( ; )
2
| |
|
|
o
\ .
o = - o
o =
to
(1)
In order to keep the property of scale invariance, the sampling of the scale parameter, o,
must be logarithmic (Keonderink, 1984). In the implementation of the Gaussian filter bank, a
new parameter t is usually introduced and it varies linearly throughout scales and relates to
o as:
t
o
. n
n
e = , where n is a natural number (N) consisting of zero and all positive integers,
that represents the level in the scale space; o is the parameter related to the resolution, or
inner scale, of each level and t determines the degree of spacing between the levels of the
scale space. Linear scale space can be seen as a generic representation of image data that is
common to different types of processing tasks. The main issues at this point are: how to
describe properly the hierarchy throughout scales and how to extract important structures
in each level of scale space and create a new data representation that can be suitable for high
level processing routines.
Multiscale representation - scale space primal sketch. Primal sketch is an image structure
first described by Marr (2010) based on the concept of gray level blobs (GLBlob), which are
regions of the image that are lighter or darker than the surrounding background. Relating
these regions to image structures results from the fact they have an intensity, or range of
intensities, that may single them out among the other structures. Qualitatively, the concept
of gray level blob for bi-dimensional images can be explained using the watershed operation
from Mathematical Morphology (Lindeberg, 1994): the image function is seen as a flooded
geographic area. As the water level decreases maxima points appear. At a certain point or
water level - two different peaks become connected. This point is called saddle. The difference
between the image intensity at the maximum and at the saddle defines the contrast of each
blob. The gray level of the saddle point is called blob base level. The support region of the blob
is defined as the region containing the points with intensity values higher than the blob base

Medical Imaging

14
level and can be reached from a local maximum without crossing points with intensity values
lower than the blob base level. For 2D images, the gray level blob is defined as the
tridimensional volume defined by the surface of the gray level and the base level.
A more formal definition of the blob concept is: the elliptical region associated to an extreme
that is delimited by the above mentioned extreme and a saddle point. Differential geometry
theory states that any image characteristic that has a geometrical meaning can be
represented by a suitable differential operator. Florack et al (1992) presented an operator for
the detection of elliptic regions or blobs - in the images, the umbilicity operator. In 3D, it is
given by Equation 2:
U
x y x z y z x y x z y z
2 2 2 2 2 2 2 2 2
2 2 2
2 2 2 2 2 2
- - -
| | | | | |
c c c c c c c c c
= + + | | |
| | |
c c c c c c c c c c c c
\ . \ . \ .
(2)
Regions with positive values of the umbilicity define the gray level blob regions. To
discriminate light from dark blobs the laplacian operator is used (Equation 3):

U and L darkblob
L
U and L light blob
x y z
2 2 2
2 2 2
0 0
,
0 0
> > c c c
= + +

> <
c c c

(3)
Dark and light blobs are not only prominent in the original image, but also tend to
maintain this prominence throughout the scales (Koenderink, 1984). That characteristic
makes them a suitable choice for the elements of an image representation containing scale
information. The Scale Space Primal Sketch (SSPS), first described by Lindeberg (1994), is
an extension of the Primal Sketch and describes the characteristics of the blobs and their
relations across scale space. Blobs in different scales are linked and form a new structure,
the scale space blobs (SSBlob).
Figure 3 presents an example of a tree structure that maps blob relations in scale space.
Figure 3.c depicts 5 levels of a scale space built for that image and their corresponding
GLBlobs. Figure 3b presents the tree mapping of the relevant events occurring when
following the blobs up in the scale space: creation, destruction and merging of blobs. These
events are called bifurcations. A SSBlob is the object produced by the union of all GLBlobs
that exist in the range between two bifurcations and is represented as a node in the tree.
Each SSBlob is associated with a maximum scale (appearing scale) and a minimum scale
(disappearing scale). The difference, given as number of scales, between the appearance and
disappearance scales is called the life time. The information content of the tree nodes
comprises the number of scales a structure survives, its appearance and disappearance
scales, contrast with background, position at each survival scale, structure volume at each
scale and total volume throughout scales. The linking between nodes reflects the behavior of
the blobs during the smoothing process: creation, destruction and merging. This tree model
is called Scale Space Primal Sketch.
Stable blobs in the SSPS must represent significant structures in the image (Koenderink,
1984, Lindeberg, 1994). Based on the stability of the SSBlobs significance measurement rates
can be defined, allowing the classification of structures. The first significance measure
proposed by Lindeberg (1994) was solely based on intrinsic characteristics of the image: blob
volume and blob life time. The assumptions for this measurement were general and
assumed there was not any a priori knowledge about the structures present in the image.


15
1
2 3
4 5
n n n n n n n
merging
merging
merging
destruction
life
time
disappearing
scale
appearing
scale
original
level 1
3
(b)
(c)
level 5
level 3
(a)
level 4
level 2
1
2 3
3 4
5
3 4
5
3 4
5

Fig. 3. (a) an image that comprises three regions with higher counts than their background.
The lower image of (a) shows the image represented as isosurface; (b) SSPS built for the
scale space shown in (c); (c) Six levels of the scale space of image (a) and their corresponding
GLBlobs.
As an example, the general significance measure can be applied to extract significant
structures in thoracic magnetic resonance images, such as cardiac cavities, arteries and
veins, shown in Figure 4 left.

Fig. 4. (left): slice of a T1 spin echo MRI thoracic sequence, the arrows point to the five target
structures. (right) Dark blobs from 9 levels of the scale space of the image shown in (a).
Values of o are: o
1
= 1.0 (original image), o
2
= 1.19; o
3
= 1.41; o
4
= 1.68; o
5
= 2.00; o
6
= 2.37; o
7
= 2.82; o
8
= 3.35; o
9
= 3.97; o
10
= 5.61. (Original image acquired at the Imaging Department -
Heart Institute HCFMUSP- with patient anonymization).

Medical Imaging

16
The thin arrows in Figure 4 right point to the 15 most significant dark blobs extracted using
the general significance rate and the scale where the blob is more distinguished. Figure 4
right shows that almost all target structures are present amongst the set of significant blobs
(exception is the atrium). Other anatomic structures are also present in the set, like the lungs
(two lower arrows in scale 8). The general approach is not able, therefore, to set aside a
particular structure from others present in the image. Furthermore, there are some
structures that do not have any anatomic meaning, such as disconnected parts of organs in
lower scales (lower arrows at scales 5 and 6) and some spurious structures originated from
noise in upper scales, identified by the thicker arrows (Rebelo, 2000). In order to overcome
the problems found with the application of the general approach, an important feature of
medical images can be embedded in the image representation: the knowledge about the
general characteristics and expected location of the target structure.
Including a priori information on SSPS. The SSPS is used as an image structure that serves
as a guide to subsequent high level processing step where a priori knowledge about the
desired feature is modeled and included in the representation. The image structure is first
used to build patterns of known objects present in the image. A particular object can then be
identified in new images by matching with symbolic features described in the pattern. Any
kind of information can be embedded in the SSPS structure. The question that arises
immediately regards defining the kind of information that is useful to include in the
representation. The answer to this question is highly dependent on the problem domain,
including the modality of the image under analysis, the kind of structure to be identified
and the representation chosen for the biological structures. To validate the proposed method
we have built a prototype application, which is described in detail in the next section.
4.1.2 Prototype application
In a previous work we proposed a prototype, in which geometrical information of the target
structure and its relation with other structures present on the scene were included in SSPS
(Rebelo, 2007). The set of parameters chosen for building the pattern was: (i) SSBlob volume,
or the sum of the support regions of all GLBlobs; (ii) support region of the GLBlob; (iii)
contextual information in the form of relative position of the target structure and the others
present in the image, specifically distances and angles. For an image containing N GLBlobs
at a given scale level the calculation of the set of parameters has a complexity of O(N
3
). In
addition, the calculation has to be performed for the images in all levels of the scale space. A
score is determined for each structure present in the image based on the closeness of its
feature values to the pattern.
The matching procedure presents, therefore, a problem related to the huge amount of
GLBlobs at the lower levels of scale. This can lead to an explosion in the number of
parameters to be processed, especially the number of angles, greatly increasing processing
time to detect the node with the highest score. In order to decrease the possible
combinations for matching, we have developed the following strategy: (1) decrease the
number of nodes for the contextual matching phase by performing a selection of the
possible nodes, using a global parameter. The prototype uses SSBlob volume, since the
previous experiments with thoracic images showed its potential for recognition of
relevant structures in medical images. A set of candidate nodes is created by eliminating
the nodes with SSBlob volume too distant from the pattern; (2) choosing a suitable scale in
the scale space for the contextual matching.


17
The concept of suitable scale for processing is based on the idea that the smoothness degree
for which an object vanishes in the scale space is related to the object size. This issue links
directly to the subject of scale selection, a task that is not trivial and that has been studied by
some authors (Lindeberg, 1994; Majer, 2000; Pauly et al, 2003). These authors articles
considered the case when image analysis is performed without any a priori knowledge about
the structures present in the scene. In the present work the suitable scale is determined by
constructing the SSPS representation for the pattern images. The analysis of the survival
scales for the structure under study and all other structures representing real entities in the
image allows the determination of the best scale for the geometrical matching. The following
steps are performed: (1) inspect all scales at which the target structure can be detected as a
single blob; (2) the levels of the scale space where all blobs representing real structures in the
image are present are considered informative and (3) the upper informative scale is
considered the most suitable for the contextual matching.
The reasons behind that choice are: (1) the occurrence of spurious structures in lower scales
is higher. This fact would highly increase the number of useless calculation of distances and
angles; (2) although structures are smoother in higher scales, the determination of angles
and distances is based on the geometrical centers of each structure. These points should not
change their position under the smoothness process. There is a problem with this strategy
when there is not a single level in which all blobs representing real structures exist as a
unique structure. This is the case when a small and low contrast structure is present and its
corresponding blobs appear only in low scale levels, when bigger objects do not constitute a
single blob anymore due to noise. In such cases, the suitable scale is chosen as the higher
scale where the blob representing the target structure appears.
4.1.3 Schema of the prototype application
The identification of structures modeled in the prototype comprises the following steps: (1)
creation of the scale space by using the linear Gaussian filter; (2) detection of GLBlobs
through the application of the umbilicity operator; (3) creation of the Scale Space Primal
Sketch; (4) comparison of the SSBlob volumes from all nodes with the pattern. Elimination
of nodes with values outside a defined range; (5) determination of the distance and angles
for the remaining SSBlob in the suitable scale; (6) contextual matching between each
candidate node and the features pattern: support region, angles and distances and (7) the
node with the best matching is identified as the target structure.
Similarity criteria. In the simplification of the SSBlob tree processing phase, the candidate
nodes are defined as those having SSBlob volume values lying in a range defined as a
function of the pattern. That range was initially set to 20% around the pattern. The
suitability of this choice are analyzed in the experiments. The similarity criterion for
contextual matching consists in minimizing the Euclidean distance between each of the
feature values obtained for each node of the test image and the pattern.
Pattern creation. The pattern is obtained by applying the proposed method to a set of
data. The following parameters of the blob representing the target structure are stored in
the pattern: SSBlob volume, the suitable scale, the support region of the GLBlob in this
scale, all distances and angles between the target structure and the others present in this
scale, the spatial resolution of the original image, the value of t used for the construction
of the scale space.

Medical Imaging

18
4.1.4 Experiments
The Mathematical Cardiac Torso (MCAT) phantom is an anthropomorphic phantom,
developed at the University of North Carolina that models size, shape and configurations of
the major thoracic structures and organs by using of mathematical formulae (Tsui, 1993). A
set of experiments was performed with two-dimensional images obtained from
manipulations - adding and mirroring - of MCAT images. The manipulation aimed at
creating frames with more organs, to make the matching step more complex. The goal was
to analyze the ability of the method to identify organs at different levels of noise and
contrast. The results for each case are described in the following sections.
Behavior for different noise levels. Five images were generated from the MCAT frames
without noise. Three different levels of additive Gaussian noise with zero mean were then
added to each image. For computing the noise level, we used the signal-to-noise ratio (SNR)
described by Rangayan (2005).

signal
noise
SNR dB
10
20* log
o | |
=
|
|
o
\ .
(4)
In equation 4,
signal
o is the standard deviation of the intensity values of the original image
(not corrupted with noise);
noise
o is the standard deviation of the intensity values of the
noise image added to the original to generate the final corrupted image. The signal-to-noise
ratio of the three different levels were 8dB (high noise level), 19 dB (medium level) and 33
dB (low level).
The five groups of images were classified into two distinct sets, according to their geometry.
Figure 5 shows two images representatives of these sets: an image from Set 1 in first line and
Set 2 in second line. In each set, one structure was chosen to be identified: structure 4 in Set1
and structure 5 in Set 2. The pattern was built based on the original images without noise. In
Figure 5 the original images (not corrupted by noise) are shown in column A. The noisy
images are presented in column B (low level of noise), C (medium level of noise) and D
(high level of noise).

Fig. 5. Two images representative of the two Sets (image 3 from Set 1 and image 4 from Set
2). Column (A) shows the original images (without noise), column (B) the low noise images,
column (C) medium noise images and column (D) high noise images.


19
After the construction of the scale space, using o = 0.19, the GLBlobs were detected in 12
levels of scale. Figure 6 depicts the GLBlobs detected for the images of Set 1 and Set 2,
presented in Figure 5. In Figure 6 it is possible to observe that the number of blobs in lower
scales increases for higher noise images. This effect tends to disapear for upper scale levels.
The application of the proposed method led to the successful identification of the chosen
structures in most experiments, exception made to two with the highest noise level.

Fig. 6. Set of GLBlobs determined in 12 levels of scale for two groups of images shown in
Figure 5. The set from Set 1 (original without noise, low level noise, mid level noise and high
level noise) is shown on the left and the set from Set 2 is shown on the right.
Behavior for different contrast levels. The contrast between a structure and its background
is one of the factors affecting the survival of the structure in the scale space, or, in other
words, it affects the shape of the SSPS. In order to test the application of the method at
several levels of contrast a simpler version of image 3 from Set 1, shown in Figure 5 (first
line), was used. From the original image, two sets of images were created to assess how
contrast influences SSPS shape. The first set was created by changing the counts of the

Medical Imaging

20
thoracic region. The second set was created by changing both the counts of the thoracic
region and the counts of the background, in such a way that the contrast between thorax
and background remained constant. The contrast C(s,b) between a target structure S and its
local background B, was calculated by Equation 5:

S B
B
C

=

. (5)
S
is the mean value of intensity inside the structure;
B
is the mean value of intensity in its
background B. For each contrast manipulated image a set of noisy images with three
different noise levels was created. For this experiment, noise was quantified by the contrast
to noise ratio, defined in relation to the target structure S by Equation 6 (Rangayan, 2005):

S B
B
CNR

=

o
(6)
where o
B
is the standard deviation of intensity in the background.
The structure number 2 (Figure 5 upper A) was chosen as the target. An interesting
observation was that when the contrast between thorax and background remained at the
same value, the blobs of the images with low contrast between the internal structures and
the thorax had a behavior not very different from the one observed in the images with high
contrast. However, all internal structures survived in a lower number of levels as the
contrast decreased.
The results of identification were satisfactory. In 87% of the experiments the target structure
was identified correctly. The pattern was built based on the original image with the lowest
contrast between the internal structures and the thorax.
5. Conclusion
In this chapter we discussed the challenges, methods and clinical examples of medical image
identification. We pointed out the importance of this task as a previous step for medical
image applications such as visualization and the quantitative analysis of organs and
systems. A classification for the methods described in the literature and a brief review of
their clinical applications were presented.
It was shown that identification of structures in medical images is a very difficult task due to
the complexity of the anatomical structures and their changing with different physiological
states. A simplification of the task can be achieved by dividing it in two phases: first,
identification of the desired structures that includes a rough segmentation; then a post-
processing phase for segmentation refinement.
In this text we presented a method for the first phase that uses an image representation
based on a scale space approach that embeds a priori knowledge in it. One of the main goals
of this proposal was to develop a powerful representation for images that would be suitable
for application of high level processing routines, such as the matching used in the prototype
implemented. A study with MCAT phantom images was performed in order to assess the
method under controlled conditions reproducing anatomical structures. The results we have
obtained were quite encouraging. However, further intensive investigation with real patient


21
data should be carried out in order to assess the full clinical usefulness of the method. The
presented method is general and can be classified as a third generation method, according to
the classification presented in the review, as it is a second generation multiresolution
representation capable of incorporating knowledge to be used in a specific application.
Although the representation is generic, the information chosen for definition of the pattern
will depend on particular applications, and can be easily inserted in the data structure.
The review presented in this chapter was intended to give the reader a general scenario of
medical applications. We focused on clinical applications and the issues for identification of
particular organs and not on detailed application and modality-dependent aspects. Methods
and applications that have appeared in the recent literature were briefly described. The main
goal was to to give the readers a glimpse of this broad and extremely active research field.
In spite of the huge number of existing methods, the problem of identification in medical
images is still challenging, with no general and unique solution and remains an open and
promising area in the field of image processing. The growing research in the clinical areas
makes the number of objects of interest in medical images to grow constantly, which
generates a continuous demand for automatic solutions to aid clinicians. On the other side,
fast advances in radiological imaging systems result in increasingly higher volume (3D)
images to be processed. Processing of these images in radiological diagnostic systems
requires accurate and fast algorithms. Intense use of apparel techniques must be studied,
and the use of new computing platforms, such as computational grids, must be explored.
The goal of automated systems is not to replace the experts, whose experience and
knowledge are essential in any automatic systems for identification or analysis. Instead, the
development of automated systems to assist the clinicians is intended to reduce their
workload in the step of analyzing patient data, saving them time to other valuable tasks as
strategies and decision making.
6. Acknowledments
This work is supported by National Council for Scientific and Technological Development,
The National Institute for Science and Technology - Medicine Assisted by Scientific
Computing (INCT-MAAC) and Zerbini Foundation.
7. References
Alattar, M. A., Osman, N.F. & Fahmy, A. S. (2010). Myocardial Segmentation Using
Constrained Multi-Seeded Region Growing. Lecture Notes in Computer Science -
Image Analysis and Recognition, vol. 6112/2010, pp. 89-98.
Allender, S., Scarborough, P., Peto, V., Rayner, M., Leal, J., Luengo-Fernandez, R. & Gray, A.
(2008). European cardiovascular disease statistics: 2008 edition, British Heart
Foundation, London.
Amira, A., Chandrasekarana, S., Montgomery, D.W. & Uzunb, I.S. (2008) A segmentation
concept for positron emission tomography imaging using multiresolution analysis.
Neurocomputing , vol.71 (2008), pp. 1954 1965.
Andra, H., Battiato, S., Bilotta, G., Farinella, G.M., Impoco, G., Orlik, J., Russo, G. & Zemitis,
A. (2008). Structural Simulation of a Bone-Prosthesis System of the Knee Joint.
Sensors, vol 8 (2008), pp. 5897-5926.

Medical Imaging

22
Antunes, S.G., Silva, J.S. & Santos, J.B. (2010) A Level Set Segmentation Method of the Four
Heart Cavities in Pediatric Ultrasound Images. Lecture Notes in Computer Science,
vol. 6112/2010 (2010), pp.99-107.
Ashburner, J., Csernansky, J.G., Davatzikos, C., Fox, N.C., Frisoni, G.B. and Thompson, P.M.
(2003). Computer-assisted imaging to assess brain structure in healthy and diseased
brains. Lancet Neurology, vol. 2 2003, pp. 7988.
Babaud, J., Witkin, A.P., Baudin, M. & Duda, R.O. (1986). Uniqueness of the Gaussian kernel
for scale space filtering, IEEE PAMI 8, (1986), pp. 26-33.
Ballard, D.H., Brown, C.M. (1982). Computer Vision. Prentice-Hall, Inc. Englewood Cliffs
Bandyopadhyay, S.K. (2011). A Survey on Brain Image Segmentation Methods. Journal of
Global Research in Computer Science, vol 2, no 2 (2011), pp. 4-7.
Bernard, O. & Friboulet, D. (2009) Fast medical image segmentation through an
approximation of narrow-band B-spline level-set and multiresolution. IEEE
International Symposium on Biomedical Imaging: From Nano to Macro, pp. 45 48.
Bouma, H., Sonnemans, J.J., Vilanova, A., Gerritsen, F.A. (2009) Automatic Detection of
Pulmonary Embolism in CTA Images. IEEE Trans on Med Imag, vol. 28, no 8 (2009),
pp. 1223 1230.
Bresser, J., Portegies, M.P., Leemans, A., Biessels, G.J., Kappelle, L.J. & Viergever, M.A.
(2011). A comparison of MR based segmentation methods for measuring brain
atrophy progression. NeuroImage, vol 54 (2011), pp. 760768.
Campadelli, P., Casiraghi, E. & Esposito, A. (2009) Liver segmentation from computed
tomography scans: A survey and a new algorithm. Artificial Intelligence in Medicine,
vol. 45 (2009), pp.185196.
Campadelli, P., Casiraghi, E. & Pratissoli, S. (2010) A segmentation framework for abdominal
organs from CT scans. Artificial Intelligence in Medicine, vol. 50 (2010), pp. 311.
Cheng, H.D., Cai, X. , Chen, X., Hu, L. & Lou, X. (2003). Computer-aided detection and
classification of microcalcifications in mammograms: a survey, Pattern Recognition,
vol. 36 (2003), pp. 2967-2991.
Cordero-Grande, L., Vegas-Snchez-Ferrero. G., Casaseca-de-la-Higuera, P., San-Romn-
Calvar. J.A., Revilla-Orodea, A., Martn-Fernndez, M. & Alberola-Lpez, C. (2011)
Unsupervised 4D myocardium segmentation with a Markov Random Field based
deformable model. Medical Image Analysis, vol. 15 (2011), pp. 283301.
Crum, W.R., Phil, D., Hartkens, T. & Hill, D.L.G. (2004) Non-rigid image registration: theory
and practice. The British Journal of Radiology, vol. 77 (2004),pp. 140153.
Destrempes, F., Meunier, J., Giroux, M.F., Soulez, G., Cloutier, G. (2009) Segmentation in
Ultrasonic B-Mode Images of Healthy Carotid Arteries Using Mixtures of
Nakagami Distributions and Stochastic Optimization. IEEE Trans on Med Imag, vol
28, no 2 (2009), pp. 215-229.
Duncan, J.S., Ayache, N (2000). Medical Image Analysis: Progress over Two Decades and the
Challenges Ahead, IEEE Trans on Pat Anal and Mach Intell, vol 22, no 1 (2000), pp.
85-106.
Ecabert, O., Peters, J., Schramm, H., Lorenz, C., von Berg, J., Walker, M.J., Vembar, M.,
Olszewski, M.E., Subramanyan, K., Lavi, G. & Weese, J. (2008) Automatic Model-
Based Segmentation of the Heart in CT Images. IEEE Trans on Med Imag, vol. 27, no
9 (2008), pp. 1189-1201.
Florack, L.M.J., Romeny, B.M.T.H., Koenderink, J.J. & Viergever, M.A.. (1992). Scale and the
differential structure of images, Image and Vision Computing, vol 10 (1992), pp. 376-388.
Gonzales, R., Woods, R.E. (2000). Digital Image Processing. Editora Edgard Blcher Ltda. So
Paulo


23
Gooya, A., Liao, H., Matsumiya, K., Masamune, K., Masutani, Y. & Dohi, T. (2008). A
Variational Method for Geometric Regularization of Vascular Segmentation in
Medical Images. IEEE Trans on Imag Proc, vol. 17, no 8 (2008), pp. 1295-1312.
Hegeds, L. (2004) Thyroid ultrasound as a screening tool for thyroid disease. Thyroid, vol.
14, no 11 (2004), pp. 879880.
Heimann, T., van Ginneken, B., Styner, M.A. et al (2009). Comparison and Evaluation of
Methods for Liver Segmentation from CT Datasets, IEEE Trans on Med Imag, vol 28,
no 8 (2009), pp. 1251-1265.
Ishii, K., Kanda, T., Uemura, T., Miyamoto, N., Yoshikawa, T., Shimada, K., Ohkawa, S. &
Minoshima, S. (2009) Computer-assisted diagnostic system for neurodegenerative
dementia using brain SPECT and 3D-SSP. European Journal of Nuclear Medicine and
Molecular Imaging, vol 36 (2009), pp. 831840.
Jain, A.K. (1989). Fundamentals of Digital Image Processing, Prentice Hall Englewood Cliffs.
Kirbas, C. & Quek, K. (2002). A Review of Vessel Extraction Techniques and Algorithms,
ACM Computing Surveys, vol-36 (2002), pp. 81-121.
Koenderink, J.J. (1984). The structure of images. Biological Cybernetics, vol. 50 (1984), pp. 363-
370.
Kohlberger, T., Uzunba, M.G, Alvino, C. & Kadir, T. (2009) Organ Segmentation with Level
Sets Using Local Shape and Appearance Priors. MICCAI 2009, Part II, LNCS, vol.
5762, pp. 3442.
Kollorz, E.N.K., Hahn, D.A., Linke, R., Goecke, T.W., Hornegger, J. & Kuwert, T. (2008)
Quantification of Thyroid Volume Using 3-D Ultrasound Imaging. IEEE Trans on
Med Imag, vol 27, no 4 (2008), pp. 457-466.
Kubota, T., Jerebko, A.K., Dewan, M., Salganicoff, M. & Krishnan, A. (2011) Segmentation of
pulmonary nodules of various densities with morphological approaches and
convexity models. Medical Image Analysis, vol. 15 (2011), pp. 133154.
Lekadir, K., Keenan, N.G., Pennell, D.J. & Yang, G-Z. (2011) An Inter-Landmark Approach
to 4-D Shape Extraction and Interpretation: Application to Myocardial Motion
Assessment in MRI. IEEE Trans on Med Imag, vol 30, no 1 (2011), pp. 52-68.
Lesage, D., Angelini, E.D., Bloch, I. & Funka-Lea, G. (2009) A review of 3D vessel lumen
segmentation techniques: Models, features and extraction schemes. Medical Image
Analysis, vol. 13 (2009), pp. 819845.
Lindeberg, T. (1994). Scale-space theory: a basic tool for analyzing structures at different
scales, Journal of Applied Statistics, vol 21, (1994), p. 225-270.
Linguraru, M.G., Sandberg, J.K., Li, Z., Pura, J.A. & Summers, R.M. (2009) Atlas-Based
Automated Segmentation of Spleen and Liver Using Adaptive Enhancement
Estimation. MICCAI 2009, Part II, LNCS, vol. 5762/2009, pp. 10011008.
Lopes, R., Dubois, P., Makni, N., Szurhaj, W., Maouche, S. & Betrouni, N. (2008)
Classification of brain SPECT imaging using 3D local multifractal spectrum for
epilepsy detection. International Journal of CARS 2008, vol 3, pp. 341346.
Lynch, M., Ghita, O. & Whelan, P.F. (2008) Segmentation of the Left Ventricle of the Heart in
3-D+t MRI Data Using an Optimized Nonrigid Temporal Model. IEEE Trans on Med
Imag, vol. 27, no 2, pp. 195-203.
Ma, Z., Tavares, J.M.R.S. & Jorge, R.N. (2008) Segmentation of Structures in Medical Images:
Review and a New Computational Framework. Proceedings of International
Symposium on Computer Methods in Biomechanics and Biomedical Engineering, Porto,
Portugal, Feb-March 2008.
Majer, P.A. (2000) Statistical Approach to Feature Detection and Scale Selection in Images, PhD
Thesis, University of Gttingen, Germany, 122p.

Medical Imaging

24
Marr, D. (2010). Vision: A Computational Investigation into the Human Representation and
Processing of Visual Information. The 2010 Edition. ISBN-13: 978-0-262-51462-0 The
MIT Press. Cambridge, MA.
Masuda, Y., Foruzan, A.H., Tateyama, T. & Chen, Y.W. (2010) Automatic liver tumor
detection using EM/MPM algorithm and shape information. 2nd International
Conference on Software Engineering and Data Mining (SEDM) 2010, pp. 692695.
Mora, M., Leiva, J. & Olivares, M. (2009) Heart Cavity Segmentation in Ultrasound Images
Based on Supervised Neural Networks. Lecture Notes in Computer Science, vol.
5496/2009, pp. 58-68.
Murphy, K., van Ginneken, B., Schilham , A.M.R., de Hoop, B.J., Gietema, H.A. & Prokop,
M. (2009) A large-scale evaluation of automatic pulmonary nodule detection in
chest CT using local image features and k-nearest-neighbour classification. Medical
Image Analysis, vol. 13 (2009), pp. 757770.
Niessen, W. (1997) Medical Image Analysis. PhD Thesis. Utrecht University, The Netherlands.
Noble, J.A. (2010) Ultrasound image segmentation and tissue characterization. Proceedings of
the Institution of Mechanical Engineers, Part H: Journal of Engineering in Medicine 2010,
pp. 224-307.
OBrien, S.P., Ghita, O. & Whelan, P.F. (2011). A Novel Model-Based 3D Time Left
Ventricular Segmentation Technique. IEEE Trans on Med Imag, vol 30, no 2 (2011),
pp. 461-474.
ODonnell, T., Funka-Lea, G., Tek, H., Jolly, M.-P. & Rasch, M. (2006). Comprehensive
cardiovascular image analysis using MR and CT at Siemens Corporate Research.
International Journal of Computer Vision, vol. 70, no 2 (2006), pp. 165178.
Oguro, S., Tuncali, K., Elhawary, H., Morrison, P.R., Hata, N. & Silverman, S.G. (2011).
Image registration of pre-procedural MRI and intra-procedural CT images to aid
CT-guided percutaneous cryoablation of renal tumors. International Journal CARS,
vol. 6 (2011), pp. 111117.
Oliver, A., Freixenet, J., Mart, J., Prez, E., Pont, J., Denton, E.R.E. & Zwiggelaar, R. (2010). A
review of automatic mass detection and segmentation in mammographic images.
Medical Image Analysis, vol.14 (2010), pp. 87110.
Pauly, M., Keiser, R. & Gross, M. (2003) Multi-scale Feature Extraction on Point-sampled
Surfaces, EUROGRAPHICS 2003, pp. 22.
Peters, J., Lessick, J., Kneser, R., Wchter, I., Vembar, M., Ecabert, O. & Weese, J. (2010).
Accurate Segmentation of the Left Ventricle in Computed Tomography Images for
Local Wall Thickness Assessment. MICCAI 2010, Part I, LNCS, vol. 6361, pp. 400408.
Petitjean, C. & Dacher, J-N. (2011). A review of segmentation methods in short axis cardiac
MR images. Medical Image Analysis, vol. 15, no 2, pp. 169-184.
Pu, J., Fuhrman, C., Good, W.F., Sciurba, F.C. & Gur, D. (2011). A Differential Geometric
Approach to Automated Segmentation of Human Airway Tree. IEEE Trans on Med
Imag, vol. 30, no 2 (2011), pp. 266-278.
Pu, J., Leader, J.K., Zheng, B., Knollmann, F., Fuhrman, C., Sciurba, F.C. & Gur, D. (2009). A
Computational Geometry Approach to Automated Pulmonary Fissure Segmentation
in CT Examinations. IEEE Trans on Med Imag, vol 28, no 5 (2009), pp. 710-719.
Qian, X., Brennan, M.P., Dione, D.P., Dobrucki, W.L., Jackowski, M.P., Breuer, C.K., Sinusas,
A.J. & Papademetris, X. (2009). A non-parametric vessel detection method for
complex vascular structures. Medical Image Analysis, vol. 13 (2009), pp. 4961.
Rahman, M., Uddin, S. & Hasan, M. (2010). 3D Segmentation and Visualization of Left
Coronary Arteries of Heart Using CT Images. IJCA Special Issue on Computer Aided
Soft Computing Techniques for Imaging and Biomedical Applications, 2010, pp. 88-92.


25
Ramakrishna, B., Liu, W., Saiprasad, G., Safdar, N., Chang, C.-I., Siddiqui, K., Kim, W.,
Siegel, E., Chai,I.W., Chen, C., Lee, S-K. (2009). An Automatic Computer-Aided
Detection System for Meniscal Tears on Magnetic Resonance Images. IEEE Trans on
Med Imag, vol. 28, no 8 (2009), pp. 1308 - 1316. (verificar autores)
Rangayan, R.M. (2005). Biomedical Image Analysis, 1st edition Boca Raton Florida CRC
Press, 1272p.
Rani. U.N., Subbaiah, P.V. & Rao, D.V. (2011). Segmentation of Brain Tumors from Medical
Scan Images Using DRLSE Variational Levelsets. International Journal of Computer
Science and Security (IJCSS), vol. 1, no 3, 7p.
Rebelo, M.S., Furuie, S.S., Gutierrez, M.A., Costa, E.T., Moura, L.A. (2007). Multiscale
representation for automatic identification of structures in medical images.
Computers in Biology and Medicine, vol. 37, no 8, pp. 1183-1193
Rebelo, M.S., Gutierrez, M.A., Furuie, S.S. & Moura, L.. (2000). Extraction of cardiac
structures through the incorporation of a priori knowledge in a multi-scale
approach, Proc. Computers in Cardiology, vol 27 (2000), pp. 611-614.
Ribbens, A., Hermans, J., Maes, F., Vandermeulen, D. & Suetens, P., (2010). SPARC: Unified
Framework for Automatic Segmentation, Probabilistic Atlas Construction,
Registration and Clustering of Brain MR Images. IEEE International Symposium on
Biomedical Imaging: From Nano to Macro (2010), pp. 856 - 859.
Rouchdy, Y., Pousin, J., Schaerer, J. & Clarysse, P. (2007). A nonlinear elastic deformable
template for soft structure segmentation: application to the heart segmentation in
MRI. Inverse Problems, vol. 23 (2007), pp. 10171035. (checar autores)
Rousset, O., Rahmim, A., Alavi, A. & Zaidi, H. (2007). Partial Volume Correction Strategies
in PET. PET Clinics, vol. 2 (2007), pp. 235249.
Rusk, L., Bekes, G. & Fidrich, M. (2009). Automatic segmentation of the liver from multi-
and single-phase contrast-enhanced CT images. Medical Image Analysis, vol. 13
(2009), pp. 871882.
Samulski, M. & Karssemeijer, N. ( 2011). Optimizing case-based detection performance in a
multi-view CAD system for mammography. IEEE Trans on Med Imag, vol. 30, no 4
(2011), pp. 10011009.
Schaerer, J., Casta, C., Pousin, J. & Clarysse, P. (2010). A dynamic elastic model for
segmentation and tracking of the heart in MR image sequences. Medical Image
Analysis, vol. 14 (2010), pp. 738749.
Scherrer, B., Forbes, F., Garbay, C. & Dojat, M. (2009). Distributed Local MRF Models for
Tno and Structure Brain Segmentation. IEEE Trans on Med Imag, vol. 28, no 8 (2009),
pp. 12781295.
Shiee, N., Bazin, P-R., Ozturk, A., Reich, D.S., Calabresi, P.A. & Pham, D.L. (2010). A
topology-preserving approach to the segmentation of brain images with multiple
sclerosis lesions. NeuroImage, vol 49 (2010), pp. 15241535.
Sonka, M., Hlavac, V., Boyle, R. (1998). Image processing, analysis, and machine vision.
Brooks/Cole Publishing Company, Pacific Grove.
Suetens, P., Fua, P. & Hanson, A.J. (1992). Computational strategies for object recognition.
ACM Computing Surveys, vol. 24, no 1 (1992), pp. 5-55.
Timp, S., Varela, C. & Karssemeijer, N. (2007). Temporal Change Analysis for
Characterization of Mass Lesions in Mammography. IEEE Trans on Med Imag, vol.
26, no 7 (2007), pp. 945-953.
Tsui, B.M.W., Terry, J.A. & Gullberg, G.T. (1993). Evaluation of cardiac cone-beam SPECT
using observer performance experiments and ROC analysis, Investigative Radiology.
vol. 28, no 12 (1993), pp. 1101-12.

Medical Imaging

26
Tu, Z. & Bai, X. (2010). Auto-Context and Its Application to High-Level Vision Tasks and 3D
Brain Image Segmentation. IEEE Trans on Pat Anal and Mach Intel, vol. 3, no 10
(2010), pp. 1744-1757.
Ugarriza, L.G., Saber, E., Vantaram, S.R., Amuso, V., Shaw, M. & Bhaskar, R. (2009).
Automatic Image Segmentation by Dynamic Region Growth and Multiresolution
Merging. IEEE Trans on Imag Proc, vol. 18, no 10 (2009), pp. 2275-2288.
van Ginneken, B., Romeny, B.M.T.H. & Viergever, M.A. (2001). Computer aided diagnosis in
chest radiography: a survey. IEEE Trans on Med Imag, vol. 20, no 12 (2005), pp. 1228-
1241.
van Rikxoort, E.M., Isgum, I., Arzhaeva, Y., Staring, M., Klein, S., Viergever, M.A., Pluim, J.
P.W. & van Ginneken, B. (2010). Adaptive local multi-atlas segmentation:
Application to the heart and the caudate nucleus. Medical Image Analysis, vol. 14
(2010), pp. 3949.
Weickert, J. (1997). Recursive separable schemes for nonlinear diffusion filters, In: Scale-space
theory in computer vision 1252. Romeny, B.T.H., Florack, L., Koenderink, J.,
Viergever, M. pp. 260-271, Springer-Verlag, Berlin Heidelberg.
Withey, D.J. & Koles, Z.J. (2007). Medical Image Segmentation: Methods and Software.
Proceeding of Noninvasive Functional Source Imaging of the Brain and Heart.
International Conference on Functional Biomedical Imaging (2007), pp. 140-143.
Witkin, A.P. (1983). Scale space filtering. Proc. Int "l Joint Conference Artificial Intelligence, pp.
1019-1022, Karslruhe, Germany, Aug. 1983.
Xie, J.; Jiang, Y. & Tsui, Y. (2005). Segmentation of Kidney From Ultrasound Images Based
on Texture and Shape Priors. IEEE Trans on Med Imag, vol. 24, no 1 (2005), pp. 45-57.
Yi, Z.; Criminisi, A.; Shotton, J. & Blake, A. (2009). Discriminative, Semantic Segmentation of
Brain Tno in MR Images. Lecture Notes in Computer Science, Volume 5762 (2009), pp.
558-565.
Yoon, S.W.; Lee, C.; Kim, J.K. & Lee, M. (2008). Wavelet-based Multi-resolution Deformation
for Medical Endoscopic Image Segmentation. Journal of Medical Systems, vol. 32
(2008), pp. 207214.
Yu, G.; Lin, P.; Cai, S. (2008). Robust Vessel Segmentation Based on Multi-resolution Fuzzy
Clustering. Lecture Notes in Computer Science, vol. 5326/2008, pp. 338-345.
Yuille, A.L. , Poggio, T.A. (1986). Scaling theorems for zero-crossings, IEEE PAMI, vol. 8
(1986), pp. 1525.
Zaidi, H. & El Naqa, I. (2010). PET-guided delineation of radiation therapy treatment
volumes: a survey of image segmentation techniques. European Journal of Nuclear
Medicine and Molecular Imaging, vol. 37 (2010), pp. 21652187.
Zhang, N., Ruan, S., Lebonvallet, S., Liao, Q. & Zhu, Y. (2011). Kernel feature selection to
fuse multi-spectral MRI images for brain tumor segmentation. Computer Vision and
Image Understanding, vol. 115 (2011), pp. 256269.
Zhao, Z., Zhang, H., Wahle, A., Thomas, M.T., Stolpen, A.H., Scholz, T.D. & Sonka, M.
(2009). Congenital aortic disease: 4D magnetic resonance segmentation and
quantitative analysis. Medical Image Analysis, vol. 13 (2009), pp. 483493.
Zheng,Y., Vega-Higuera,F.V., Zhou,S.K. & Comaniciu, D. (2010). Fast and Automatic Heart
Isolation in 3D CT Volumes: Optimal Shape Initialization. Lecture Notes in Computer
Science, vol. 6357/2010, pp. 84-91
2
Detection of Abnormalities in a Biological
Tissue by Diffuse Optical Tomography:
A Computational Study
H. Trabelsi, M. Gantri and E. Sediki
University of Tunis El Manar, Facult des Sciences de Tunis
Unit de Rayonnement, Dpartement de Physique, Tunis
Tunisia
1. Introduction
Diffuse optical tomography (DOT) has a real potential as a new medical diagnostic tool. This
fact has stimulated considerable research interest in the last years. When it is possible, this
technique has some advantages not available in classical medical imaging methods. Among
the benefits of this technique, nonionizing radiation, low-cost instrumentation, mobility and
possible functional imaging are the most important. Also, DOT could be used in
combination with other imaging methods to provide high-resolution structural information.
The transparency of tissues is related to absorption and scattering of light and it reaches its
maximum in the near-infrared. Interaction of radiation from this range with biological
tissues is widely studied [1-5]. In particular, some works was interested to detection of
tumor-like inclusions within a biological tissue [6-8].
Most of models of light interaction with biological matter are based on radiative transfer
theory [9-13]. The most used parameters in modelling Laser radiation interaction with
biological tissue is absorption and scattering. However some other studies evoked a
significant variation of refractive index of abnormal biological tissues especially in the
near infrared range. More precisely, experimental results [14-17] showed that the tissue of
malignant tumors could manifest an increase of the refractive index which could attain
until 10 % of that of a normal tissue which encircles them. So, medical imaging by diffuse
optical tomography should take advantage from the emergence of a third contrast
parameter which is the refractive index. This drove to the appearance of a big number of
numerical and fundamental works in the field of radiative transfer in a varying refractive
index medium. Among them some papers are interested to varying refractive index
biological tissues [18-22].
In this chapter, infrared laser radiation interaction with a biological tissue including a
spatially variation of refractive index is studied through a time-dependent radiative
transfer model. Our general concern is to contribute at the usability of the radiative
transfer theory in a potential optical tomography setting in medical imaging. At this level,
studying the effect of refractive index on the transmitted light through a biological
rectangular layer is crucial. This should be a contribution to obtaining information about

Medical Imaging

28
abnormal tissue in a typical DOT scheme. However, it is important to note that in a
varying refractive index medium, the rays are not straight lines but curves. So even in a
rectangular geometry, the varying refractive index radiative transfer equation displays
angular derivative terms [23-25]. In this paper, these terms are treated with finite
differencing scheme. More precisely, we present a computational model that could be
suitable for basic optical tomography forward problem with spatially varying refractive
index. We investigate cases concerning optical tomography applications. We consider a
biological tissue-like domain submitted to a near infrared light source. Transmitted
intensity detected on the boundary is computed. The effect of the embedded non-
homogenous objects on the transmitted signal is studied. Particularly, a possible detection
of these objects is discussed. Also, the model is used to study infrared radiation in a
multilayered heterogeneous medium. The effect of optical parameters on the transmitted
signal is presented. In particular, the effect of refractive index variation is pointed out.
2. Physical model and numerical implementation
In this work, the radiative transfer in a human biological tissue is described by using a
varying refractive index RTE [26, 27]

2
0
( ) ( , , ) 1
. ( , , ) ( ( ) ( )) ( , , ) . ( , , )
( )
2
( . ) ( , , ) ( , , ) ( ) ( , ') ( , ', ) '
( )
a s
s
n r I r t
I r t r r I r t n I r t
c t n r
n I r t S r t r p I r t d
n r
t

O
c O
+ OV O + + O + V V O
c
OV O = O + O O O O
}
(1)
where I r t ( , , ) O
is the directional energetic radiance at the spatial position r
and n r ( )
is the
refractive index distribution. Equation (1) takes into account the fact that the rays are not
straight lines but curves. It involves terms that illustrate the expansion or the contraction of
the cross section of the tube of light rays in the medium. On the boundary, the radiance is
the sum of the external source contribution and the partly-reflected radiance due to the
refractive index mismatch at the boundary
( , , ) ( , , ) ( , , ), . 0 . .
b ref b b ref b
I r t S r t RI r t n and n n O = O + O O < O = O

,
where
b
r
is a position on the boundary and

b
n
is an outer normal unit vector. The

reflectivity R can be calculated for each direction using Fresnels relations. For a two-
dimensional problem and in Cartesian coordinate system of the x-y plane, the terms due to
the refractive index variation can be expressed as

1 2 1 1
. ( . ) (sin 2cos ) (cos 2sin )
n I n I
n I n I I I
n n n x n y

O
c c c c
V V OV = + +
c c c c
, (2)
where cos and sin are the Cartesian coordinates of the unit direction vector in the x-y
plane. In fact we assume that the radiance of out of plane directions is negligible. By using
notations cos = and sin q = , equation (2) displays the classical form of the angular
redistribution term commonly appearing when dealing with spherical and cylindrical
geometries with uniform refractive index [28, 29]
Tissue by Diffuse Optical Tomography: A Computational Study

29

n n
n I n I I I
n n x y n
2 2
2 2
2
1 2 1
. ( . ) [(1 ) ] [(1 ) ]
2
q
q
O

c c c c
V V OV = +
`
c c c c

)
(3)
The angular redistribution terms will be noted

x y
D I andD I
2 2
[(1 ) ] [(1 ) ] q
q
c c
= =
c c
,
so equation (3) becomes

2 2
2
1 2 1
. ( . )
2
x y
n n
n I n I D D
n n x y n
O

c c
V V OV = +
`
c c

)

In our numerical implementation, we use a rectangular domain which is divided into a set
of I x J elementary uniform volumes V with a uniform unitary depth. The angular
discretization is obtained through a discrete ordinate technique. This yields a set of M
discrete directions, { }
m
m M , 1.. = giving a set of angular discrete direction
cosines{ }
m m
m M ( , ), 1,... q = . An orientation depending on the incident ray direction is
adopted for each cell [12]. Calculations are done by using integration of equation (1) over an
elementary volume V for each discrete direction. This gives

m P
m m E m W m m N m S a s m p
x m y m m P s m mm m p
M
s m mm m p
m m m
I
n
y I I x I I x y I
c t
n n
D D x y S w p I
n x n y
w p I
,
, , , , ,
, , , ,
' ' ',
' 1, '
( ) ( ) ( )
1 1
(
)
q
= =
c
+ A + A + A A +
c
c c
+ + = A A +
c c
+

(4)
where
x m
D
,
and
y m
D
,
are the discrete angular derivative terms at the angular ordinate
m

and
m
respectively and w
m
is a weighting factor. The discrete term of Henyey-Greenstein
phase function is written as

mm
m m m m
g
p
g g
2
'
2 3/2
' '
1
4 (1 2 ( )) t q q
=
+ +

If the direction cosines are positive, the directional radiances are known on the faces W and
S and they are unknown on the faces E and N of the (i,j)-cell and also in the centre P.
Therefore, we need two complementary relations to eliminate
m N ,
+ and
m E ,
+ , this can be
obtained by using interpolation formula

m P m E m W
m P m N m S
I I I
I I I
, , ,
, , ,
(1 )
(1 )
o o
o o
= +
= +
,
where is an interpolation parameter. Using these relations, equation (3) becomes

Medical Imaging

30

m P m m
m P m W m P m S a s m p
x y m P s m mm m p
M
s m mm m p
m m m
I y x n
I I I I x y I
c t
n n
D D x y S w p I
x y n
w p I
,
, , , , ,
2 2
, ,
2
' ' ',
' 1, '
( ) ( ) ( )
1
(
2
)
q

o o
= =
c A A
+ + + A A +
c

c c
+ + = A A +
`
c c

)
+

(5)
For time discretization, an implicit three level second order time differencing scheme is used:

n n n
m P m P m P m P
I I I I
n n
t t
1 1
, , , ,
max
3 4
; 1,...
2
+
c +
= =
c A

where t A being the discrete time step.
Theoretically, if we know the solution in the (i,j)-cell, we can do calculus over the cells (i+1,j)
and (i,j+1) using the boundary conditions and the following relations

m w m E
m S m N
I i j I i j i I
I i j I i j i J
, ,
, ,
( 1, ) ( , ); 1,... 1
( , 1) ( , ); 1,... 1
+ = =
+ = =

If the direction cosines are both positive, we get the following equation

n n n
m i j m i j m i j
n n n n m m
m i j m i j m i j m i j
n
a s m i j x y
M
n n
m i j s m mm m i j s m mm m i j
m m m
I I I
y x n
I I I I
c t
n n
x y I D D
x y n
x y S w p I w p I
1 1
, , , , , ,
, , , 1, , , , , 1
2 2
, ,
2
, , , , ' ' ', ,
' 1, '
3 4
( ) ( )
2
1
( )
2
( )
q
o o

+

= =
+
A A
+ + +
A

c c
A A + + +
`
c c

)
= A A + +

3. Numerical treatment of angular derivative terms
The numerical treatment of the angular redistribution terms can be done by using the
classical angular differencing scheme [23, 28].

m m i j m m i j
x m i j
m
A I A I
D
w
1/2 1/2, , 1/2 1/2, ,
, , ,
.
+ +
=

m m i j m m i j
y m i j
m
B I B I
D
w
1/2 1/2, , 1/2 1/2, ,
, , ,
+ +
=
where
) I I (
2
1
I
j , i , m j , i , 1 m j , i , 2 / 1 m
+ =
+ +
and

31

k k k
m i j m i j m i j
I I I
1/2, , 1, , , ,
1
( )
2

= + with
M i j i j i j M i j
I I I I
1, , 1, , 0, , , ,
,
+
= = .
The coefficients A and B are given through the recurrent relations

m m m m
A A w
1/2 1/2
2
+ +
= ,

m m m m
B B w
1/2 1/2
2 q
+ +
= with A B
1/2 1/2
0 = = ,
and w
m
being a quadrature weight factor. In this paper, we use a constant weight quadrature
and an equally spaced distribution of angular ordinates ( , ) q . To solve the above equation
we use successive iterations in order to actualise the implicit internal source term in the
right member. So, this gives

( ) ( )
( ) ( )
( ) ( ) ( )
m i j
m i j m i j m i j m i j
x m i j y m i j
k
i j
n m m
a i j s i j
k k
i j i j
n n n n m m
k k
i j i j i j i j
i j
m i
n
y x
I x y
c t
n n
y x
I I I I
c t c t
y n n D x n n D
n
x y S
, ,
, , , , , 1, , , 1
, , , , , ,
1
1
,
1
, , , ,
1 1
, ,
1 1 1
2 2 2 2
, 1, , , 1
2
,
,
3
2
2
2
1
2
q

o o
q
o o

+
+
+ +
+ +

(
A A
= + + + A A +
(
A

A A
+ +
A A
+ A + A
+A A
( ) ( )
m i j m i j
M
k k
n n
j s m mm s i j m mm
m m m
w p I w p I
, , 1 ', , 1
1 1
, , , ' '
' 1, '

+ +
= =
(
(
(
(
(
(
(
(
| |
(
+ + |
| (
\ .

The iteration process is repeated until a convergence criterion is attempted. To improve
convergence speed, we use a successive over relaxation method. So the updated value
( )
m i j
k
n
I
, ,
1 +
is a linear combination of the iterated value
( )
m i j
k
n
I
, ,
and the previously computed
value:

( ) ( ) ( )
m i j m i j m i j
k k k
n n n
updated
I I I
, , , , , ,
1 1
(1 )
+ +
= +
where is a relaxation parameter whose value is usually between 1 and 2. The solution is
obtained when the relative discrepancy value:

( ) ( )
( )
m i j m i j
m i j
k k
n n
k
n
I I
I
, , , ,
, ,
1
c
+
=
is smaller than a tolerance value. In that case the result is noted
m i j
I
, ,
. In all our calculus, we
have taken 10
-8
as a tolerance value. As initial condition, we take a field of null radiances.
Also, all our calculations are done in the case of interpolation diamond scheme (=0.5).
If the direction cosines are not both positive, the precedent equations are valid provided that
the orientation WESN of cells is done according to the direction of propagation. In all our

Medical Imaging

32
investigations, the injected power source is assumed to be equivalent to a forward
collimated monochromatic intensity placed at a source point on the middle of the bottom
side of the boundary. Results shown below are obtained by using a near infrared collimated
source with a uniform equivalent intensity value of 50 mW.cm
-1
. To present our results, we
use the detected fluence rate which is given in a (i
d
,j
d
)-detector point on the boundary as

d d d d
M
d m i j m m i j
m
R w I
, , , ,
1
(1 )
=
u =
, with

d d
d d
d d d d
d d d d
air
m
i j
m i j
i j m air t i j t air m
i j m air t i j t air m
n
if
n
R
n n n n
else
n n n n
,
, ,
, ,
2 2
, ,
1 arcsin
cos cos cos cos
1
[( ) ( ) ], .
2 cos cos cos cos

=
| |
|
|
\ .
+ +

where

d d
air m
t
i j
n
n
,
sin
arcsin

| |
| =
|
\ .
and
air
n 1 ~ .
Also, we make use of a normalized detected fluence rate defined as

d
N
D
d d
d
D w
1
1 / '
=
u
u =
u
,
where D is the number of the detector points on one side of the boundary and w
d
is a
weighting factor from the generalized trapezoidal integration rule. In all calculations, we
have used 28 detector points on each side. Also, all calculus is carried out by using 16
uniformly distributed discrete directions and a space grid of 121x121cells.
4. Results and discussion
4.1 First test: A continuous varying refractive index medium
As a first investigation, we study near infrared radiation transport in a rectangular medium
exposed to a continuous collimated source which is placed on the bottom side of the
boundary. Figure 1 shows the considered medium, it is assumed to be 3cmx3cm sized with
varying refractive index. Within the medium, we consider a y-axis linear varying refractive
index between the extreme values 1.33 and 1.58. It is an increasing- decreasing variation of
refractive index RI. To show the effect of the refractive index on detected fluence rate, we
have used a weakly absorbing and forward scattering background medium whose optical
parameters are shown in figure1. Figure 2 shows the medium response detected on the right
side of the boundary in the case of linear and parabolic variation respectively. The detected
fluence rate curve presents distinguish increasing and decreasing regions with a peak of
transmission when refractive index is close to the minimum value 1.33. Transmission on the
boundary is relatively augmented when the refractive index is the closest to the air index.

33

Fig. 1. A test-medium with background properties and detector points

Fig. 2. Response of a continuous varying refractive index medium (a) in the case of a linear
variation and (b) in the case of a parabolic variation.
4.2 Second investigation: Detection of heterogeneous object
In this investigation, we consider a 3cmx3cm sized background medium containing a
circular heterogeneous object. The area of the object is 0.5 cm
2
. We limit the analysis to the
effect of heterogeneity by increase of the refractive index only. The background refractive
index is taken 1.33. The other optical properties are the same as in the precedent
investigation for the background and the object. The geometry of the medium and the
position of the heterogeneous object are shown in figure 3. The detected signal on different
air, n=1

a
s
cm
cm
g
1
1
0.5
25
0.8
=
=
=

Source
x-axis
Detector points
y-axis
y-varying RI
(a) (b)

Medical Imaging

34
sides of the medium is shown for an increase of refractive index by 5% and 10%
respectively. There is no significant effect of the object presence on the detected signal on the
left side because it is far from the detectors on that side. However there is a visible effect of
the object on the right side and on the top side where the detectors are relatively near by the
object. In such cases, there is an obvious effect of the heterogeneity on the detected signal
especially when the refractive index is increased by 10%. The response shows a distinct
distortion of the fluence rate curve. It is occurring on the targets where the object centre is
laying exactly. This distortion allows a principled decision on the existence of such object in
the medium. These findings highlight the potential of refractive index as a possible
detection parameter of a tumour in a surrounding safe tissue.

Fig. 3. Detection of heterogeneous objects by the effect of refractive index variation. (a)
Geometry of the considered medium and object, (b) Detected fluence rate on the left side,(c)
Detected fluence rate on the right side (d) Detected fluence rate on the top side.
air, n=1
Background
(RI: n)

Object
(RI: n +on)
2
2
y-axis
x-axis
(a)
(b)
(c)
(d)

35
4.3 Third investigation: Detection of inclusions in a biological tissue
In the course of this investigation, we simulate a post-mortem human prostate tissue-like
medium containing two inclusions. The geometry of the medium and the targets of the
inclusions are shown in figure 3. Each inclusion is a circle of area 0.5cm
2
. The background
medium properties at 850nm wavelength are taken from [30, 31]. Inclusion 1 is double
scattering only and inclusion 2 is double absorbing only. The refractive index of the two
inclusions is taken the same and it is noted n
2
. Detected fluence rate on the top and on the
lateral sides of the medium is computed and presented in figure 5 and figure 6. In particular,
we confront results from refractive matching case (n
1
=n
2
) and from refractive mismatching
case (n
2
>n
1
) on each side. According to figure 5, the introduction of refractive index with
scattering and absorbing parameters gives a distinguish effect on top side detected signal.
However this effect is more attenuated in lateral sides because of the strong forward
scattering within the medium. This is shown in figure 6.

Fig. 4. Heterogeneous inclusions in human prostate-like tissue: geometry of the medium and
inclusions.

Fig. 5. Effect of heterogeneous inclusions on detected fluence rate on the top side (a) in the
case of refractive matching (n
1
=n
2
=1.37) and (b) in the case of refractive mismatching
(n
1
=1.37 and n
2
=1.51).
air, n=1
=
= =
=
1
1
1
0.6
100 , 0.94
1.37
a
s
cm
cm g
n
Inclusion 2
2a, n2
Inclusion 1
2
s
, n
2
2 1
2

Medical Imaging

36

Fig. 6. Detected fluence rate on lateral sides in both cases refractive matching and
mismatching cases: (a) Response on the right side (effect of inclusion 1) and (b) Response on
the left side ( effect of inclusion 2)
4.4 Fourth investigation: Effect of refractive mismatching in multilayered tissue
As a fourth investigation, we consider a multilayered biological tissue-like medium. It
contains three superposed layers of skin, fat and muscle. Figure 7 shows the considered
medium. It is 3cmx3cm-sized medium with a fixed layer of skin (2mm) and with a varying
fat-layer thickness of 3mm, 6mm and 9mm respectively. Optical properties of different
tissues at 850nm-wavelength and refractive indices of the layers are taken from [30, 31].
Calculus is carried out as in the just previous section.

Fig. 7. A three-layered biological tissue-like medium with optical properties of skin, fat and
muscle at 850 nm-wavelength.
Figure 8 shows results concerning transmitted fluence rate on the right side of the boundary
in the case of a continuous wave source. We can see that in most detector points and for the
three fat-layer thicknesses transmitted fluence rate profiles are almost superposed in the
case of refractive matching with evidence of detected intensity increase when the fat

= = =
1 1
1
0.12 , 87.5 , 0.8
,
a s
cm cm g
Skin n

= = =
1 1
2
0.01 , 40 , 0.8
,
a s
cm cm g
Fat n

= = =
1 1
3
0.15 , 44.7 , 0.8
,
a s
cm cm g
Muscle n


37
thickness increases. When taking different refractive indices, a distinct distortion in the
curves at every interface is observed especially at the fat-muscle interface. In all cases much
more intensity is transmitted from the muscle layer.

Fig. 8. Response of the three layered medium: (a) for uniform refractive index and (b) for
different refractive indices.
4.5 Fifth investigation: Detection with a short pulse
In this investigation, we study the effect of abnormalities laying in a homogenous background
medium as in the just precedent paragraph but we use a short-pulsed source. It is a 100fs-laser
pulse injected in the medium through a point source on the middle of the bottom side of the
boundary. Figure 9 shows the considered medium and the geometry and the properties of the

Fig. 9. A skin-like tissue containing a heterogeneous cyst-like inclusion submitted to short
pulse and the detected signal.

Short pulse
Detector
S
u
1
' 0.00012cm
=
Q
u
1
0.0049cm
=
RI=n
1
Heterogeneous
s
u
1
' 13.6cm
=

Q
u
1
0.35cm
=

RI=n
2
Background skin

Medical Imaging

38
a circular cyst-like object laying just forward the source. The detected signal on the opposite
side of the source is shown in same figure. There is a distinct phase shift due to refractive
mismatching showing the presence of the cyst within the background tissue.
4.6 Sixth investigation: Spectroscopy on a rat-liver
In this investigation, we consider a rat liver tissue-like medium as it is shown in figure 10.
Optical properties of the tissue are taken from reference [30] from a set of several
wavelengths. The response illustrate transmission through the considered medium on the
top side. Only near infrared light displays a relative appreciable transmission. A maximum
of detected fluence rate is noted at 800nm-light. A little delay of the maximum detection of
radiation is observed when wavelength increases. Figure 11 shows fluence rate distribution
into the medium at different moments after the pulse. In all moments, symmetric
propagation of light is observed into the medium. Also, it can be observed that scattering
prevails in 488nm while more and more absorption is observed in 2100 nm. At 488nm, the
phenomenon of multiple scattering is dominant into the medium. The dispersion of the
pulse is accentuated in all sides. Light disappears after 250ps. At 2100nm, more light is
absorbed into the medium so a fraction of energy persists into the medium for more time.
Figure 12 shows typical movement of different photons in the medium at three different
wavelengths. Calculations in our studied medium are carried out by using Monte Carlo
simulations as it is described in reference. It can be observed that a typical 800nm-photon is
almost snake. For 488nm-photon, the mean scattering free path is very small. Multiple
scattering is dominant. The typical trajectory in this case is not ballistic and the photons are
almost diffusive. For 2100nm, the mean scattering free path is longer than the 488nm-photon
but the absorption mean free path is narrower. A typical photon in this wavelength is
almost absorbed into the medium.

Fig. 10. Geometry of rat-liver tissue and detected signal for different wavelengths.

A Rat liver tissue

2cm
Short pulse

39

Fig. 11. Internal light distribution in the rat-liver tissue-like medium for different instants
after the pulse.

Fig. 12. Typical trajectory of photons for different values of the wavelength-source as
obtained through Monte Carlo simulations.

Medical Imaging

40
5. Conclusion
In this chapter, we attempted to develop a computational way helping in detection of
abnormalities in a biological tissue. This should enable predictions of eventual tumor
existence when using a diffuse optical tomography scheme. The used model is based on
radiative transfer theory including a possible variation of refractive index. The model is
implemented to investigate some practical situations in DOT setting and near infrared
spectroscopy. Obtained results showed that variation of refractive index can yield useful
predictions about the target and the location of abnormal inclusions within the tissue.
6. Nomenclature
c Speed of light, (in vacuum), [cm/s]
DOT Diffuse Optical Tomography
D
x,y
Angular derivative terms
g anisotropy factor
I intensity, [W.cm
-2
sr
-1
]
n refractive index
p(,) phase function
r position vector, [cm]
R reflectivity coefficient,
r
b
position vector in the boundary
R I Refractive Index
S source term, [W.cm
-3
sr
-1
]
w
d
weighting factor from trapezoidal integration,
w
m
weight associated with discrete direction
x,y cartesian coordinates, [cm]
Greek symbols
interpolation parameter
fluence rate, [W.cm
-2
]
d
Detected fluence rate, [W.cm
-2
]
wavelength, [nm]
direction solid angle, [sr]
a
absorption coefficient, [cm
-1
]
s
scattering coefficient, [cm
-1
]
s
reduced scattering coefficient, [cm
-1
]
relaxation parameter, [-]
scattering angle , [rad]
, direction cosines, [-]
7. References
[1] Selser J C, Yeh Y, Baskin R J. A light scattering characterization of membrane vesicles.
Biophysical Journal; 16 (4): 337-356, (1976)
[2] Jacques S L, Prahl S A. Modeling optical and thermal distributions in tissue during laser
irradiation. Lasers in Surgery and Medicine; 6: 494-503, (1987)

41
[3] Arridge S R, Hebden J C. Optical imaging in medicine: II. Modeling and reconstruction.
Phys Med Biol; 42: 841853, (1997)
[4] Arridge S R, Schweiger M. A gradient-based optimisation scheme for optical
tomography. Opt Express; 2 (6): 213226, (1998)
[5] Hielscher A H, Bluestone AY, Abdoulaev AY, Klose A D, Lasker M, Stewart M, Netz U,
Beuthan J. Near-infrared diffuse optical tomography. Disease Markers; 18: 313337,
(2002)
[6] Mourant J R, Hielscher A H, Eick A A, Johnson T M, Freyer J P. Evidence of intrinsic
differences in the light scattering properties of tumoregenic and nontumorigenic
cells. Cancer (Cancer Cytopathology); 84 (6): 366-384, (1998)
[7] Ntziachristos V, Hielscher A H, Yodh A G, Chance B. Diffuse Optical Tomography of
Highly Heterogeneous Media. IEEE transactions on medical imaging; 20(6) :470-
478, (2001)
[8] Pu Y, Wang W.B , Das BB, Alfano RR.Time-resolved spectral wing emission kinetics and
optical imaging of human cancerous and normal prostate tissues. Optics
communications; 282:4308-4314, (2009)
[9] Klose, A, D, Netz,U, Beuthan,J and Hielscher, A, H, Optical tomography using the time-
independent equation of radiative transfer. Part I: Forward model, JQSRT, 72: 715-
736, (2002)
[10] Cai W, Xiaohui Ni, Gayen S K, and Alfano R R. Analytical cumulant solution of the
vector radiative transfer equation investigates backscattering of circularly polarized
light from turbid media. Phys. Rev. E (74) 056605, (2006)
[11] Gantri M, Trabelsi H, Bensalah R and Sediki E. Solution of a radiative transfer problem
in a biological tissue. An optical tomography model. AIP Conf. Proc; 935: 237-243,
(2007)
[12] Trabelsi H, Gantri M and Sediki E. Visible and Near Infrared Laser Radiation in a
Biological Tissue. A Forward Model for Medical Imaging by Optical Tomography.
Lasers in Med Sci; 25:41-53, (2010)
[13] Gantri M, Trabelsi H, Sediki E, Ben Salah R. Computational Laser Spectroscopy in a
Biological tissue. Journal of Biophysics, article ID 253763, 9pages, (2010)
[14] Li H, Xie S. Measurement method of the refractive index of biotissue by total internal
reflection. Applied Optics; 35(10):1793-1795, (1996)
[15] Das B B, Liu F , Alfano R R. Time-resolved fluorescence and photon migration studies
in biomedical and model random media. Rep. Prog. Phys; 60: 227-292, (1997)
[16] Ohmi M, Ohnishi Y, Yoden K, Haruan H. In vitro simultaneous measurement of refractive
index and thickness of biological tissue by the low coherence interferometry. IEEE
Transactions on Biomedical Engineering; 47(9):1266-1270, (2000)
[17] Lai J, Li Z, Wang C, He A. Experimental measurement of the refractive index of
biological tissues by total internal reflection. Applied Optics; 44(10):1845-1849,
(2005)
[18] Beuthan J, Minet O, Helfmann J, Herrig M, Muller G. The spatial variation of the
refractive index of biological cells. Phys Med Biol; 41:36982 (1996)
[19] Dehghani H, Brooksby B, Vishwanath, K, Pogue B W, Paulsen K D. The effects of internal
refractive index variation in near-infrared optical tomography: a finite element
modelling approach, Physics in Medicine and Biology; 48 (16):27132727, (2003)

Medical Imaging

42
[20] Khan T, Jiang H. A new diffusion approximation to the radiative transfer equation for
scattering media with spatially varying refractive indices. J. Opt. A: Pure App. Opt;
5:137-141 (2003)
[21] Shendeleva M L, Molly, J, A, Scaling property of the diffusion equation for light in a
turbid medium with varying refractive index, Journal of the Optical Society of
America A; 24 (9): 29022910, (2007)
[22] Trabelsi H, Gantri M, Sediki E, Ben Salah R. A Time-dependent Computational Model
for Radiation in a Spatially Varying Refractive Index Biological Medium.
Proceedings of Eurotherm 83, Computational Thermal Radiation in Participating
Media III; Lisbon, Portugal, (2009)
[23] Lemonnier D, LeDez V. Discrete ordinate solution of radiative transfer across a slab
with variable refractive index. JQSRT; 72(2/5): 195204, (2005)
[24] Liu, L, H, Benchmark numerical solutions for radiative heat transfer in two-dimensional
medium with graded index distribution, JQSRT; 102: 293 303, (2006)
[25] Bal G. Radiative transfer equations with varying refractive index: a mathematical
perspective, Journal of the Optical Society of America A; 23: 16391644 (2006)
[26] Ferwerda, H, A, Radiative transfer equation for scattering media with a spatially varying
refractive index, Journal of Optics A: Pure and Applied Optics; 1(3): L1L2, (1999)
[27] Tualle J M, Tinet E. Derivation of the radiative transfer equation for scattering media with
a spatially varying refractive index, Optics communications; 228:33-38, (2003)
[28] Modest M, F. Radiative heat transfer; ISBN 0-12-503-163-7, Academic Press, (2003)
[29] Trabelsi H , Sghaier T et Sifaoui M S. Theoretical Study of Radiation between two
Concentric Spheres Using a Modified Discrete Ordinates Method associated with
Legendre Transform, JQRST; 93:415-428, (2005)
[30] Cheong W F, Prahl S A, Welch A J. A Review of the Optical Properties of Biological
Tissues. IEEE J. Quantum Electronics; 26:2166-2185, (1990)
[31] Mobley J and Vo-Dinh T. optical properties of tissues, biomedical photonics handbook;
ISBN 0-8493-1116-0, CRC Press LLC (2003)
0
3D Ultrasound Image Segmentation: Interactive
Texture-Based Approaches
Julien Olivier
2,1
and Ludovic Paulhac
1
1
Universit Franois Rabelais Tours, Laboratoire Informatique (EA2101)
2
cole Nationale dIngnieurs du Val de Loire
France
1. Introduction
The recent breakthroughs in 3D medical imaging technologies open new promising
perspectives in the health domain. Signicant efforts have being carried out and the precision
of acquisition systems keeps on being improved. These devices are becoming widespread in
the hospitals and the constantly increasing queues for this kind of exams prove the interest
in these technologies. On the contrary, it seems that the evolution of 3D image analysis tools
does not generate the same interest. It may be because, for a long time, people believed that
considering 3Dimages as a succession of 2Dframes was a sufcient and efcient way to make
precise analysis. In our opinion, it is not the case and huge advancements can be obtained by
considering 3D images with their entire complexity. This approach needs the development
of specic models which can deal with multiple knowledge and are able to process huge
information quantities. As a consequence, this chapter will present two original and efcient
approaches of computer science for 3D image analysis and more particularly 3D ultrasound
images.
Ultrasound techniques present a certain number of advantages when compared to other
acquisition techniques like magnetic resonance imaging (MRI), X-ray computed tomography
(CT), etc. Ultrasounds are not ionizing, which means they are not invasive for the patients.
Moreover this technique is inexpensive and allows real time acquisitions. Nevertheless,
the interpretation of ultrasound images is very complex and specialists are usually needed
throughout the examination. In the same way, the segmentation (extraction of specic regions
in the image) of ultrasound images is a very difcult task as these medical images present
characteristic artifacts like shadows, speckle, attenuations, missing boundaries etc.
(Noble & Boukerroui (2006)) proposed a complete survey about native B-mode ultrasound
image segmentation in which they tried to point out what makes a good ultrasound
segmentation method. Among efcient techniques, the authors have identied methods
dealing with image features such as gray level distribution, intensity gradient, phase,
similarity measures and texture measures. As described in (Noble & Boukerroui (2006)),
texture analysis is very efcient for ultrasound classication and segmentation. The classical
method of Haralicks co-occurrence matrices (Haralick et al. (1973)) has often been used
and has obtained good performances in several applications (Basset et al. (1993); Valckx &
3
2 Will-be-set-by-IN-TECH
Thijssen (1997)). Despite the great number of methods, results are still imprecise and the
proposed solutions often focus on a given problematic, dedicated to one type of application.
Nevertheless, it could be interesting to develop generic tools, automatic or semi-automatic,
which would allow a more important usability and interactivity.
In this chapter, two original interactive systems for 3D skin ultrasound image segmentation
are presented. Section 2 describes the interests of using 3D ultrasound images for contents
analysis and presents classical methods to acquire this images. In section 3, we quickly
explain the different families of texture analysis and we provide a discussion about the interest
of using an interactive system in a man-aided application. In section 4, the rst approach
focuses on the development and the combination of efcient perceptual volumetric texture
attributes easily understandable by humans. In section 5, the second one is based on a
supervised segmentation model, interactively allowing the user to give useful information to
the algorithmbefore the segmentation. To conclude, we discuss about our work and introduce
the main prospects.
2. 3D ultrasound images
This section presents how to construct an acquisition of a 3D ultrasound image. We quickly
explain the interest of the three-dimensions and what it is possible to analyze in ultrasound
images of the skin.
2.1 Ultrasound image acquisition using a 3D probe
A 3D ultrasound image is acquired using a 3D probe that allows the user to obtain a volume
of echo (16 16 8 mm
3
) using the scanning of an acoustic beam. For the acquisition of 2D
ultrasound images, three main methods have been used (Grgoire et al. (2006)):
The linear scanning: gives rectangular images well adapted to supercial exploration but
the acquisition is slow.
The sectorial scanning: faster method but the obtained image is not rectangular and
presents an angulation.
The circular scanning: very simple method but the image shape is a disk.
By combining two 2D scanning modes, following two different axis, it is possible to obtain a
3D scanning. To construct a 3D acquisition, it is also possible to move manually a 2D probe
according to its perpendicular axis. In this case, the quality of the acquisition depends on the
operator dexterity. Then, it is preferable to use a real 3D probe, with small size and weight,
to obtain the best acquisition quality. This kind of probes has been used to acquire the 3D
ultrasound images in this paper.
2.2 Ultrasound images of the skin
Today, manufacturers propose echographic systems with a resolution ranging from 100 m
down to 30 m. This requires ultrasonic frequencies ranging from 20 MHz to 60 MHz.
The increase of ultrasonic frequencies allows a resolution improvement but the wave in
the media is attenuated, which limits the applications to supercial exploration. Resolution
44 Medical Imaging
3D Ultrasound Image Segmentation: Interactive Texture-Based Approaches 3
Fig. 1. Example of a three-dimensional image of the skin obtained with 20 MHz ultrasound
scanner (Atys Medical France)
provided with high frequencies ultrasound allows to observe the skin perfectly and especially
the dermis that has an average thickness ranging from 1 to 2 mm. It is also possible
to explore a part of the hypodermis. On the other hand, the resolution is insufcient
to explore the epidermis (Figure 1). Indeed, its thickness vary between 0.05 to 0.3 mm
which need ultrasound frequencies higher than 80 MHz. Sonography of the skin allows
tumor visualizations (cyst, nevus, melanoma, basal cell carcinoma (BCC) etc.), inammatory
pathologies, scars. The discrimination between the different lesions is not always obvious but
cutaneous sonography is an important help for detection and diagnosis. The possibility to
segment and characterize a lesion in three dimensions is very useful to establish therapeutic
strategies. The three-dimensional sonography of the skin is rarely used because of the
lack of three-dimensional image analysis tools but the recent evolution of three-dimensional
probes should allow the emergence of new techniques. With a three-dimensional acquisition,
it is possible to obtain features that are inaccessible with two dimensions. Moreover
three-dimensional sonography is well adapted to supervise the evolution of a structure or
a lesion notably using volume measures.
As it has been briey presented in section 1, Noble & Boukerroui (2006) established that
texture analysis was a very efcient way to process ultrasound images. Thus, the next section
present the principle of this kind of analysis.
3. Texture analysis
3.1 Denition
When dealing with complex images (such as medical images), relying only on rst order
statistics such as gray level intensity means, standard deviations or histograms is not sufcient
to carry out a precise analysis. Indeed, two objects with close intensity statistics and different
visual aspects can generally be found in this kind of images (see gure 2). Such objects require
45 3D Ultrasound Image Segmentation: Interactive Texture-Based Approaches
to use analysis of higher order such as texture analysis, which allow us to study not only the
intensity statistics of the pixels but also their spatial distribution in the image.
(a) Grey level intensity mean : 133.9 ; Standard deviation : 42.7
(b) Grey level intensity mean : 126.5 ; Standard deviation : 40.7
Fig. 2. First order analysis limitations: images a) and b) possess the same rst order statistics
(mean, standard deviation and gray level histogram) but their visual aspects are clearly
different. Thus a high order analysis becomes necessary
Before presenting the main texture features, let us try to present a denition for the notion of
texture.
Even if several denitions appear in the literature (see Tuceryan & Jain (1993)), if we stay in a
general point of view we can dene a textured zone in an image as a gray level distribution
respecting an ordered scheme and for which the determination of unique features is possible.
However, three main texture types are identied in the literature (Richards & Polit (1974)).
Deterministic textures: they are composed of one unique element, called a texton (see
Julesz (1975)), which is regularly repeated in the space according to specic orientation
and period. Even if this kind of textures can easily be characterized, it remains very rare in
natural images. That is why, most of the time, deterministic textures are articial ones.
46 Medical Imaging
Stochastic textures: they are composed of non-regular pattern. These textures can be
considered as bi-dimensional random elds.
Quasi-deterministic textures: they are composed of several patterns that are very close to
each other, but rarely identical. For this reason it remains difcult to isolate one unique
pattern. The natural textures often belong to this type of textures.
These three texture types are illustrated in gure (3).
(a) Deterministic textures
(b) Stochastic textures
(c) Quasi-deterministic textures
Fig. 3. The three main texture types
The next part of this chapter presents the most used texture features.
3.2 Texture features
The purpose of a texture feature is to describe a textured zone with, at least, one numerical
value. In an ideal scheme, two different textures will have two different values for one given
feature. Unfortunately, this rarely happens because, usually, the discriminating power of each
texture feature is strongly dependent to the kind of treated textures. In Richards &Polit (1974),
four principal texture feature families are identied: statistical methods, geometric methods,
model-based methods and, nally, frequency-based methods. Here, we will voluntary not
give details about the feature calculations but will only present their principles. If interested,
the reader will nd further details by following the reference of each approach.
The statistical methods represent the oldest approaches in the texture analysis eld. Their
principle is to study the pixels gray level spatial distribution in the image. They usually use
statistics of at least second order (i.e. the pixels are studied by pair instead of independently).
The features from this family represent the best choice for non-expert users because of their
implementation easiness. Among the statistical features, the following ones can be considered
as the most used: the Haralick coefcients of the coocurrence matrix (Haralick et al. (1973)),
the autocorrelation measure (Otsu & Kurita (1988)) and the Local Binary Pattern (LBP) (Ojala
et al. (1996)).
The approaches fromthe geometric method family are based on the principle of repetitiveness
present in most of the textures. Indeed, they try to identify and characterize the texture
primitive of a texture zone (the texton). As a consequence, this methods are very efcient on
deterministic textures but encounter more difculties when dealing with natural textures such
as stochastic or quasi-deterministic ones. The Vorono tessellation (Shamos & Hoey (1975))
and the structural methods (Voorhees & Poggio (1987); Zucker (1976)) can be considered as
the two principal approaches of this family.
The principle of the model-based methods is to build an articial texture model which
must be as similar to the studied texture as possible. The evolution of this articial model
characteristics is observed during its construction and once the model is close enough to the
studied texture, these characteristics are kept as the texture features. The Markov random
elds (Li (1995)) and the fractal approaches (Mandelbrot (1977)) represent the two principal
model-based methods.
The frequency-based methods consider that a texture can be characterized by analyzing the
repetition of one or several pattern according to various spatial frequencies. Thus, several
methods, initialy developed for 1Dsignals (like sound), have been adjusted to 2Dsignals such
as images. The most popular frequency-based methods are the Fourier analysis (Azencott
et al. (1997)), the Gabor lters (Turner (1986)) and the wavelet transform (Mallat (1989)).
3.3 Choosing the right texture feature
As it has been presented in the previous section, several texture features have been developed
over the past thirty years. Unfortunately, most of these methods do not have a general
applicability and cannot identify some classes of texture. For example, some of these
approaches are not able to describe the directionality properties. In comparison, the human
visual system can adapt to all types of textures even in the case of an unfavorable context.
As a consequence, when developing an texture-based image analysis application, it would
seems natural to think of including the maximum of texture features in order to obtain the
best accuracy. Unfortunately, this strategy does not represent a good choice as some of the
included features will alter the segmentation if they are not discriminating enough.
48 Medical Imaging
When dealing with ultrasound images, the various number of applications, each using
different texture features, conrm this observation. As shown in (Noble & Boukerroui
(2006)), the classical method of Haralicks co-occurrence matrices (Haralick et al. (1973))
has been widely used and has obtained great performances in several applications (Basset
et al. (1993); Valckx & Thijssen (1997)). However, the increasing use of three-dimensional
acquisition technologies in clinical practices requires three-dimensional segmentation or
analysis methods. Boukerroui et al. (2001) propose a multi-resolution segmentation of
three-dimensional ultrasound data (2D+T, 3D) using gray scale intensity, three-dimensional
Haralick texture features and three-dimensional tissue characterizing information obtained
from the local frequency spectra of the radio-frequency signals. The authors conclude that the
use of complementary and/or redundant texture features allows a more robust segmentation.
Sahiner et al. (2004) characterize breast masses on three-dimensional ultrasound images. They
developed 2D and 3D active contour models for an automated segmentation. Then, they
extracted three-dimensional texture and morphological features from the segmented mass.
In their study, classication results of malignant and benign breast masses are similar to
those experienced by breast radiologists. Zhan & Shen (2003; 2006) present a deformable
model to segment three-dimensional ultrasound images. They compute texture features
using two banks of two-dimensional Gabor lters located in the two orthogonal planes.
Indeed, the use of a three-dimensional Gabor lter bank should have increased the number of
lters and computation time. Nevertheless with two banks of two-dimensional Gabor lters,
information is lost in comparison with a bank of three-dimensional Gabor lters.
Thus, as no unique efcient texture feature can be identied, it remains very important to
choose the right texture features but this task is very difcult for a non-expert user because
no single mathematical model describing all the features exists. Moreover, the names of the
texture features rarely refer to something easily understandable.
This aspect of the texture analysis motivated us to develop interactive approaches which allow
a non expert user to carry out texture-based segmentations without having to choose some
precise texture features. This two approaches are presented in the following two sections.
4. Human Understandable Features (HUF)
In this rst approach, a set of 3D texture features, inspired by the human way to describe a
texture, is proposed. By using human understandable texture features, it is easier and possible
for an operator (specialist or not in image analysis) to select the relevant features to process an
image. Moreover in a given application it allows a better contents interpretation.
The set of characteristics has been chosen using analysis results from some previous
works (Amadasun & King (1989); Tamura et al. (1978)). The chosen characteristics are
the following: Granularity, which can be represented by the number of three-dimensional
patterns constituting the texture, shape information about these patterns (volume, compacity,
regularity), contrast and, nally, roughness of the image.
In sub-section 4.1, we present a multiresolution scheme for segmentation and we detail
the HUF computation. Sub-section 4.2 presents psychological experiments proving the
correspondence between our texture features and a human description of textures. Finally,
Fig. 4. Multiresolution segmentation
sub-section 4.3, shows segmentation results of 3D ultrasound images of the skin. More details
about this work are available in Paulhac, Makris, Gregoire & Ramel (2009).
4.1 A multiresolution scheme for texture segmentation
The proposed approach allows us to give an image description using texture features for
several resolutions by using the 3D discrete wavelet transform (Figure 4). Mallat (1989)
proposes a decomposition scheme using lters: a highpass lter, which allows to obtain
detail coefcients and a lowpass lter which gives approximation coefcients. Roughness
is computed using detail coefcients whereas the other proposed features are computed
using approximation coefcients. Before the segmentation process, the proposed features are
computed for each voxel and for several resolutions. In order to obtain a segmentation of the
start image, a step of feature upsampling is necessary for each resolution. Thus, a voxel of
the initial image is described by a vector containing 6n different features with n the number
of resolutions and 6 the number of proposed features. The user is then asked to choose the
features that seem the most relevant from his point of view. And it is because they are human
understandable that this choice will be easy. Finally, the K-means algorithm (Coleman &
Andrews (1979)) allows to generate a segmentation using the set of computed vectors.
4.1.1 A geometric approach for 3D texture analysis
We have chosen to describe the geometric structure of the textures with the help of the
three-dimensional connected components which can be viewed as the patch patterns in
the texture. To compute connected components, we propose a similar method to the one
presented in (Shoshany (2008)): a gray-level textured image is decomposed into a progressive
sequence of binary textures in order to study patterns and their evolution. In our approach, a
clustering algorithm applied on the voxels of the initial 3D image allows us to identify a set of
thresholds used to compute the sequence of binary versions of the image. Then, connected
50 Medical Imaging
(a) (b) (c) (d)
Fig. 5. a) One ultrasound image, b) The same ultrasound image after a binarization, [c-d]
Connected components extracted from different binary images
components are computed for each binary textures of the produced sequence. Figure 5
presents a binarized image with examples of connected components.
Connected components represent the basic objects inside binary textures. Their analysis can
provide important geometrical and volume information and it allows computation of features
like granularity which corresponds to the number of patterns per volume unit, the volume and
compacity of each connected components providing information about the shape of texture
patches. The regularity can also be estimated using the variance of these patterns. In our
case, the number of patterns corresponds to the number of connected components (nbCC) per
volume unit. For a given texture, if the number of connected component is important for the
resolution then, there is an important number of patterns and the granularity ( f
gran
) of the
texture is signicant.
Besides the number of connected components, we compute shape characteristics with the
average volume and the average compacity of connected components. Like the number of
connected components, the volume is an additional information to identify the neness of a
texture. The average volume of connected components is computed as follows:
f
vol
(x, y, z) = (
nbCC
i=1
V
i,
)/nbCC
(1)
where V
i,
is the volume of the i
th
connected component for the resolution . The considered
connected components are located in a cube of size N
3
centered at the coordinates (x, y, z).
Compacity of connected components gives information about pattern shape. A texture with
an important compacity is a texture with compact patterns. Otherwise this is a texture with
elongate shapes. This characteristic is invariant by translation, rotation but also to scale
changes (Zhang & Tan (2002)). It has been used to texture characterization in (Goyal et al.
(1995)). The compacity of a connected component can be computed as follows:
C
i,
=
S
3
2
i,
V
i,
(2)
where S
i,
is the surface and V
i,
is the volume of the i
th
connected component for the
resolution .
It is then possible to compute the average compacity f
comp
:
f
comp
(x, y, z) =
1
nbCC
nbCC
i=1
C
i,
(3)
We can also obtain information about the regularity of a texture from the study of the
connected components. Therefore, we decided to use the compacity variance. We have
seen that this characteristic is invariant by any transformation (Zhang & Tan (2002)). The
shape of patterns is the only element that affects the variance feature. A low variance of
the compacity indicates an important regularity of the connected components whatever their
spatial organization is.
f
reg
(x, y, z) = E(C
2
) (E(C
))
2
(4)
where E is the expected value.
4.1.2 Statistical and frequency based method for contrast and roughness measure
The surface of a rough texture presents a high number of asperities. In an image, roughness
can be described as a set of quick spatial transitions with varying amplitude. From a
frequential point of view, the image asperities in the spatial domain correspond to the presence
of high frequencies. Detail coefcients from the wavelet transform give a description of high
frequencies in an image among several directions. It is then possible to have an estimation of
the texture roughness for a specic resolution.
Finally, we propose to compute the roughness attribute as follows:
f
rgh
(x, y, z) =
M
=1
(
N
i=1
N
j=1
N
k=1
w
,
(i, j, k)
)/M (5)
where f
rgh
is the roughness at the resolution . w

,
(i, j, k) corresponds to the set of detail
coefcients in a cube of size N
3
centered at a voxel of a sub-band at the coordinates (x, y, z)
and M represents the number of detail coefcient sub-bands for a resolution.
Haralick et al. (1973) propose a measure to estimate contrast by computing the moment of
inertia from the main diagonal of the co-occurrence matrix. Nevertheless, the construction
of a co-occurrence matrix, only to obtain an estimation of the contrast, can be expensive in
computing time. Tamura et al. (1978) claim that four factors are supposed to inuence the
contrast difference between two textures. They are the dynamic range of gray-levels, the
polarization of the distribution of black and white on the gray-level histogram or ratio of
black and white areas, the sharpness of edges, and the period of repeating patterns. They
propose to approximate the contrast with a measure incorporating the two rst factors. We
use this approximation in our work. To obtain a measure of polarization the kurtosis
4
is
used. It allows a measurement of the disposition of probability mass around their center.
52 Medical Imaging
4,
=
4,
(6)
where
4
is the mean fourth moment and
2
the variance of gray-levels for the resolution .
In order to take into account the dynamic range of gray-levels, they combine the kurtosis with
the standard deviation as follows:
f
cont
(x, y, z) =
n
4,
(7)
where n is a positive value. In their paper, Tamura et al make comparisons between
psychological experiments and their operators before concluding that the value n = 1/4 gives
the best approximation. At last, the values of
and
n
4,
are computed in a cube of size N
3
centered at the coordinates (x, y, z).
4.2 Psychological experiments
These experiments have been carried out to prove that a strong correspondence exists between
the proposed features and the human vision. Thus, we propose to construct psychometric
prototypes and to compare them to our texture measures.
The set of textures presented in Figure 6 has been used in our experiments. These textures
have been constructed using methods presented in (Kopf et al. (2007); Paulhac, Makris &
Ramel (2009)) except for textures (j) and (l) that are subsets of ultrasound images. Each of
themis a volumetric texture (texture in the 3Ddomain) of size 128
3
with 256 gray-levels. They
have been printed using a printer HP Color LaserJet 3700 and presented to human subjects.
The group of human subjects was composed of 15 men and 11 women, and the majority of
them had no knowledge in image and texture analysis. We distributed to each one of them
a questionnaire containing the set of textures (Figure 6) and an explanation of the texture
features used in our model. For each feature, textures had to be classied in descending order,
i.e. from the most rough to the most smooth, from the most regular to the most irregular,
etc. Using these questionnaires, we constructed a ranking of these textures for each texture
attribute. For a given characteristic, a score has been assigned to a texture according to its
ranking. For example, The most rough takes the value +12 (for the roughness feature), the
second one +11, the last one +1, and this for all the questionnaires. The addition of the
questionnaire scores for each texture allowed us to obtain a nal ranking for a given texture
feature.
Using the proposed texture attributes, we also generate a feature ranking. A vector of 6
features has been computed for each texture in the questionnaire. Here only the rst resolution
has been considered ( = 1), because this is the one which corresponds the best to the
observation of textures by human subjects through questionnaires.
4.2.1 Comparison between human and computational ranking
To compare human and feature ranking, the degree of correspondences between them has
been determined. In this respect, we chose to use the well-known Spearmans coefcient of
(a) (b) (c) (d)
(e) (f) (g) (h)
(i) (j) (k) (l)
Fig. 6. Set of Solid textures used for psychological experiments
rank correlation:
r
s
= 1
6
n
3
n
n
i=1
d
2
i
(8)
where n is the number of individuals, and d
i
is the difference between the ranks assigned
to the i
th
object in the two measurements. The value of this coefcient varies between 1
and 1. Value 1 corresponds to the complete agreement of the two rankings whereas value 1
indicates complete disagreement. Table 1 presents coefcients of rank correlations between
human and feature ranking.
f
gran
f
comp
f
vol
f
reg
f
rgh
f
cont
0.83 0.9 0.61 0.82 0.75 0.65
Table 1. Coefcients of rank correlations between human and feature ranking
54 Medical Imaging
Region Description Texture attributes scoring
Nevus, Histiocytobroma,
Cyst, Melanoma, BCC
These lesions are present in the
dermis and have an average or
low echogenicity.
granularity:+, compacity:+++,
contrast:+, roughness:+,
regularity:++
Normal Dermis In this zone, there is a regular
echogenicity.
granularity:+++, compacity:+,
contrast:++, roughness:++,
regularity:+
Hypodermis This region of the skin
can contain more or less
echogenicity according to
zones.
granularity:++, compacity:++,
contrast:++, roughness:++,
regularity:+
Epidermis Resolution is to low to analyze
epidermis. Moreover it is
similar to the plastic membrane
in ultrasound images with a
high echogenicity.
granularity:+, compacity:+,
contrast:++, roughness:+,
regularity:++
Table 2. 3D interesting textures in skin images
These results show a huge correlation between human and feature ranking. For the volume,
the Spearmans coefcient indicates that there is a link between the two measurements with
a condence rate included between 95 and 98 percent. The compacity feature gives the best
result with a condence rate which tends toward 100 percent. The volume feature has the
smallest correlation results. It can be supposed that it is sometimes difcult for our human
subjects to visualize the volume of patterns because of the 3D.
4.3 Segmentation of 3D ultrasound images with the HUF system
As explained in Noble & Boukerroui (2006), the scatterer distribution and their
relative volumes to the wavelength of the incident ultrasound pulse produces different
three-dimensional texture patterns. During an ultrasound examination, echogenicity
represents the ability of a cellular tissue to create an echo. In an ultrasound image, echogenic
zones contain a large number of white three-dimensional patterns and this is an important
characteristic used by specialists to identify pathologies. From a texture analysis point of
view, echogenicity of a zone can be described mainly by granularity measures carried out on
these three-dimensional patterns. The interest of each of the proposed HUF features has been
validated by specialists in dermatology. Table 2 presents a synthesis of the correspondence
between the HUF features and the characteristics according to the regions in the skin. It was
not possible to predict the exact value of the HUF features and a score (low:+, medium:++,
high:+++) has been proposed for each texture attribute according to the regions in the skin.
4.3.1 Presentation of the segmentation software
The architecture of the proposed system is composed of three main modules. The rst one
computes texture features according to user-dened parameters. Then, HUF features are used
in the second module of segmentation. Finally, the segmented image can be exploited by
the 3D visualization and manipulation module. This module allows to visualize the initial
segmentation and to improve, in an interactive way, this rst result. It is also possible to
represent regions using a mesh or to compute volume information to help specialists in their
diagnostic.
Before running the segmentation, the user needs to select the features that seem the most
relevant to process a 3Dimage. Agraphic interface allows the user to dene his choice (Figure
7). The selected textural features, the parameters and the processed volumetric image are
then exploited by the module of feature computation. For each voxel, the selected textural
features are computed. Then, the segmentation module uses the set of computed feature
vectors to generate a rst segmentation with the K-means algorithm (Coleman & Andrews
(1979)). Finally, using the graphical interface, the user can improve the segmentation using
merging and splitting operations. Then, it becomes possible to merge two regions or to focus
on a particular region by running, once more, a segmentation (a Kmeans clustering) of this
region.
Fig. 7. Interface of the proposed software for segmentation of 3D ultrasound images.
4.3.2 Segmentation results
Our software has been provided to several dermatologists in order to help them to collect
3D information about pathologies. Figure 8 presents segmentation results for different skin
ultrasound images. Using a segmentation, many extractions are possible: lesions, tendons,
skin layers etc. Figure 8[a-b] shows two naevi, gure 8[c-d] shows two Histiocytobroma,
gures 8[a-d] contain at the left the original three-dimensional ultrasound image, at the center
an image of classied voxels and on the right a mesh of the lesion built using the segmented
image. With these results, it becomes possible to perform measures like volume and depth,
56 Medical Imaging
(a) Nevus, K=2 (b) Nevus, K=2
(c) Histiocytobroma, K=3 (d) Histiocytobroma, K=4
(e) Layers of the skin
Fig. 8. Segmentation of three-dimensional ultrasound images of the skin
in order to help specialists in their diagnostic, to track pathologies evolution or to carry out
more precise extractions etc. With a clustering, all the voxels of an image are classied and it
is possible to make a visualization of the different skin layers(gure 8(e)). For a specialist, this
visualization is interesting because skin layers evolve according to the patient age and it is
interesting to supervise the healing of a burned skin. This visualization is also of high interest
for the evaluation of cosmetic product effects. To evaluate our results, the segmented images
and their corresponding meshes have been presented to specialists in ultrasound images. The
evaluation is only a qualitative one for the moment because producing ground truth for 3D
images is a very hard task. Indeed, it is necessary to produce an expert segmentation for
each two-dimensional cut (for example in z-axis direction) of a three-dimensional ultrasound
image. With these set of two-dimensional ground truths it could be possible to construct
a three-dimensional one but the gathering of the two-dimensional images could generate
holes as well as precision problems. Thus, it stays difcult or even impossible to obtain a
ne resolution for the evaluation.
5. Texture-based supervised active contours driven by a classier for 3D
ultrasound image segmentation
In this section, we present our second interactive texture-based model. The heart of the system
is a 3Dactive contour, but for the user to easily deal with texture features, it has been chosen to
guide the active contour segmentation with a supervised binary classier. First, the principle
of active contour segmentation will be presented. In a second part, we will describe the
principle of supervised binary segmentation. Then, the complete segmentation process will
be presented and nally, the segmentation software will be described.
5.1 Active contour segmentation
Initially developed by Kass et al. (1988), active contours are powerful segmentation
approaches widely used in the segmentation eld. An active contour is dened as a
parametrized curve C mapping a parameter s to a pixel x(s) in the image . The curve is
initialized in and evolves, under some constraints, according to its normal and tangential
directions until it stops on the boundary of the object to be segmented. (see gure 9)
Fig. 9. Example of active contour segmentation
As the curve evolves in time, C(s, t) represents the family of curves obtained during the
evolution. An energy is attached to the model and dened to control its evolution as:
E(C(s, t)) =
f (s, t)ds , (9)

f being a force composed of geometric criteria and external terms computed from the image
(data terms). By minimizing (9), the model evolves until the nal curve is placed on a position
corresponding to a local minimum of E. Initially, the energy attached to the model was
obtained by integrating f only along the curve (Caselles et al. (1997); Xu & Prince (1998)).
These models are called boundary-based but their application is restricted to objects whose
boundaries can be dened by gradients and are not very efcient on complex images like
textures. Later, region-based models emerged by integrating f inside the curve or over the
entire domain of (Paragios & Deriche (2002); Zhu & Yuille (1996)).
Originally, Kass et al chose to represent the curve explicitly (the curve is sampled and its
evolution is guided by several control points) but this prevents the model from handling
topology changes without additional implementation. Yet, an implicit representation called
the level set implementation emerged from the work of Osher and Sethian (Osher & Sethian
(1988)). The main advantage of level sets is to allow the curve to handle topology changes
automatically, as it can actually split or merge with others naturally. In order to implicitly
represent a curve C, it is necessary to dene its evolution with a partial differential equation
58 Medical Imaging
(PDE or motion equation) such as:
C(s, t)
t
= FN , (10)
with F the speed function of the model and N its inward normal vector.
PDE of models implicitly represented with level sets are usually deduced from the energy of
the curve using Euler-Lagrange equations (Caselles et al. (1997); Zhu & Yuille (1996)), even
though the rst models were geometric ones, directly dened by their PDE (Caselles et al.
(1993); Malladi et al. (1995)).
As the goal of the presented application was to ensure the most precise segmentation on 3D
ultrasound images, the developed model is implicitly represented and uses Haralick texture
features (Haralick et al. (1973)). Moreover, in order to let our model choose the best texture
features, we introduce a supervised binary classier in the motion equation of the active
contour. The next section presents the principle of supervised binary classication. Details
about the introduction of the classier in the motion equation can be found in (Olivier et al.
(2008)).
5.2 Supervised binary classication
The principle of supervised binary classication is to observe a learning dataset X composed
of several samples such as X = {x
1
, ..., x
n
}. For each sample x
i
R
m
, i {1, ..., n}, its m
features as well as the class to which its belongs are known and used to dene a classication
rule

l = f (x) allowing the classier to automatically determine the class

l of each sample
x / R
m
.
In other words, given a set of already classied samples, a supervised classier is able to
determine the class of every unknown sample according to its features, automatically.
In the presented model, it has been chosen to give the classication task to a articial neuron
network (Rosenblatt (1958)).
The next section present the complete segmentation process.
5.3 Interactive segmentation process
As it as been presented in the previous section, using a supervised classier requires the
denition of a learning dataset composed of classied samples. As a consequence, our model
evolves in two step. First, during an interactive step, an expert segmentation is carried out
on one single slice of the 3D ultrasound image. The local Haralick texture features of some
pixels from this segmentation are used to create the samples of a learning dataset, allowing
the classier to achieve its learning task. In a second step, a 3D active contour driven by this
classier is launched on the 3Dimage to carry out the desired segmentation. Figure 10 depicts
the complete segmentation process. The extraction of the learning dataset is detailed in the
following section.
Segmentation Step
One slice
from the
3D image 3D mage
Classifier
Learning
Segmentation using
the Active Contour
Learning
Dataset
Extraction
nteractive Step
3D Active Contour
Driven by
Supervised
Binary Classifier
Expert
Segmentation
Segmented
image
3D Model
Fig. 10. The complete segmentation process: the expert segmentation is used to determine
the learning dataset and to carry out the classiers learning task. The segmentation is then
launched on the 3D image.
5.3.1 Extraction of the learning dataset
During this step, one single slice
is extracted from the complete 3D utrasound image.

Then, the user is asked to carry out an expert segmentation C
. This segmentation must be

composed of two ideal regions C
in
and C
out
. C
in
will contain pixels from the object to be
segmented while C
out
will be composed of pixels belonging to a narrow band surrounding
C
in
. Intuitively, C
in
should be taken as the interior of C
and C
out
as all the remaining image,
giving
= C
in
C
out
. But in real images the region outside the segmented object is rarely
homogeneous and sometimes even contains regions with textures very close to those of the
inside region. Moreover, even if this phenomenon does not appear in
, it can be present in
other slices of the 3D image. This may involve a high variability in the textures from C
out
in
the 3D image, making it hard for the classier to correctly carry out its task. Thus, in order
to maximize the distance between textures from C
in
and C
out
(giving a higher discriminant
power to the Haralick texture features and to keep texture properties as constant as possible,
we chose to dene C
in
and C
out
as respectively the interior of C
and a narrow band giving

homogeneous information about the boundary region of C
.
Once the two regions determined, m Haralick features are computed for each of their pixels.
Thus, a learning dataset X composed of n = n
1
+ n
2
samples, n
1
belonging to C
in
and n
2
belonging to C
out
, is determined. i [1, n], a sample s(i) will be described as:
s(i) = { k
1
(i), k
2
(i)...k
m
(i), l(i)} ,
l(i) =
1 if s(i) belongs to C
in
1 if s(i) belongs to C
out
,
(11)
60 Medical Imaging
with k
1
(i), ..., k
m
(i) the m Haralick coefcients of the i
th
extracted pixel and l(i) a label
representing the region to which it belongs. Figure 11 shows an example of a manual
segmentation carried out during the interactive step.
(a) (b) (c) (d)
Fig. 11. Example of expert segmentation. a): denition of C
, b) denition of the narrow

band, c) denition of C
in
, d) denition of C
out
.
5.3.2 3D segmentation
Once the learning dataset is extracted, the automatic segmentation of the image is carried out.
The learned classier is used to guide the 3Dactive contour in the 3Dimage. The segmentation
is stopped as the model reaches the boundaries of the object to be segmented. Once the
segmentation is carried out, the nal result is obtained by transforming the segmentation into
a 3D mesh. Such a representation allows the user to easily understand the geometry of the
object.
5.4 Segmentation software
5.4.1 Interface
Fig. 12. The segmentation software interface.
The gure 12 presents the complete segmentation interface. The left part (blue square) is the
visualization area. It allows the user to see the complete 3D ultrasound image. The red square
in the right shows the segmentation command buttons while the green square surrounds the
interactive part dealing with the manual segmentation and the supervised classication.
5.4.2 Interactive step
Once the 3D image is open, the user select one slice from it and carry out the manual
segmentation (see g 13.a). Then, the learning step of the articial neural network is launched
by pressing the "Learn" button. This step will allow the software to extract Haralick texture
features and to determine the best ones.
5.4.3 Segmentation step
Now that the texture features have been determined, the 3D segmentation can be launched.
When pressing the "Launch segmentation" button, the user will see the 3D active contour
grow in the image until it stops on the boundaries of the object to be segmented (see gure
13.b). It can be noticed that the average segmentation duration is about half a minute on a
common computer. The nal step of the segmentation is the 3D reconstruction of the model.
(a) (b)
(c)
Fig. 13. a) Interactive part of the software. b) 3D segmentation. c) 3D mesh
62 Medical Imaging
The segmentation can be turned into a 3D mesh by pressing the "Create mesh" button (see
gure 13.c). Figures 14.a and 14.b shows other segmentation results. As it has been presented
in section 4.3.2, numerically validating a 3D segmentation method is not easy as 3D ground
truths are hard to obtain. As a consequence, all segmentation results have been presented to
dermatologists and visually validated.
(a)
(b)
Fig. 14. Other segmentation results using the software.
6. Conclusion
In this chapter, two texture-based approaches for 3D ultrasound image segmentation have
been presented. These methods have been developed for the user comfort and propose an
important usability and interactivity.
In the rst method a multiresolution scheme for volumetric texture segmentation have been
dened. It includes perceptual features inspired by a human way to describe textures. The
main interests of this approach are an easier way for an operator (specialist or not in image
analysis) to select the relevant texture features to process an image and a better contents
interpretation in a given application. We also presented psychological experiments that prove
the strong correspondence between the proposed features and human perception of textures.
In the second approach, an interactive step is used where the user is asked to carry out a
manual 2D segmentation on one single slice from the 3D image. Then, this segmentation is
used to carry out the learning task of a supervised binary classier. This last will determine
the best texture features and will guide the segmentation of a 3D active contour among the
3D image, automatically. Thus, the major innovation of this work is to allow any non-expert
user (in the image analysis domain) to achieve a texture-based segmentation. In some future
work, this method will include more texture features in order to increase its adaptability.
Even if the applications presented in this chapter were focuses on ultrasound images, the two
presented systems has also been successfully applied for the segmentation of MRI images as
well as images obtained by confocal microscopy.
7. References
Amadasun, M. & King, R. (1989). Texture features corresponding to textural properties, IEEE
Transactions on Systems, Man, and Cybernetics 19(5): 12641274.
Azencott, R., Wang, J.-P. & Younes, L. (1997). Texture classication using windowed Fourier
lters, IEEE Transactions on Pattern Analysis and Machine Intelligence 19(2): 148153.
Basset, O., Sun, Z., Mestas, J. & Gimenez, G. (1993). Texture analysis of ultrasonic images of
the prostate by means of co-occurrence matrices, Ultrasonic Imaging 15: 218237.
Boukerroui, D., Basset, O., Baskurt, A. & Gimenez, G. (2001). A multiparametric
and multiresolution segmentation algorithm of 3-d ultrasonic data, TUFFC 01:
Proceedings of the IEEE transactions on ultrasonics, ferroelectrics, and frequency control.
Caselles, V., Catte, F., Coll, T. & Dibos, F. (1993). A geometric model for active contours,
Numerishe Mathematik 66: 131.
Caselles, V., Kimmel, R. & Sapiro, G. (1997). Geodesic active contours, International Journal of
Computer Vision 22(1): 6179.
Coleman, G. & Andrews, H. (1979). Image segmentation by clustering, Proceedings of the IEEE,
pp. 773785.
Goyal, R., Goh, W., Mital, D. & Chan, K. (1995). Scale and rotation invariant texture analysis
based on structural property, IECON 95: Proceedings of the International Conference on
Industrial Electronics, Control, and Instrumentation.
Grgoire, J.-M., Serrire, S., Georgesco, G., Jamet, F., Bleuzen, A., Ossant, F., Levassort, F.,
Tranquart, F. & Patat, F. (2006). Techniques et applications de lchographie haute
rsolution non invasive, Journal de Radiologie 87: 19201936.
Haralick, R. M., Shanmugam, K. &Dinstein, I. (1973). Texture features for image classication,
IEEE Transactions on Systems, Man, and Cybernetics 3(6): 610621.
Julesz, B. (1975). Experiments in the visual perception of texture, Vol. 232, Scientic American.
Kass, M., Witkin, A. & Terzopoulos, D. (1988). Snakes: active contour models, International
Journal of Computer Vision 1(4): 321331.
Kopf, J., Fu, C.-W., Cohen-Or, D., Deussen, O., Lischinski, D. & Wong, T.-T. (2007).
Solid texture synthesis from 2d exemplars, SIGGRAPH 07: Proceedings of the 34th
International Conference on Computer Graphics and Interactive Techniques.
Li, S. Z. (1995). Markov Random Field Modeling in Computer Vision, Springer-Verlag New York,
Inc.
Malladi, R., Sethian, J. A. & Vemuri, B. C. (1995). Shape modeling with front propagation:
a level set approach, IEEE Transactions on Pattern Analysis and Machine Intelligence
17(2): 158175.
Mallat, S. G. (1989). A theory for multiresolution signal decomposition: the wavelet
representation, IEEE Transactions on Pattern Analysis and Machine Intelligence
11: 674693.
Mandelbrot, B. B. (1977). Fractals, Forms, Chance and Dimension, Freeman, San Fransisco.
Noble, J. A. & Boukerroui, D. (2006). Ultrasound image segmentation: A survey, IEEE
Transactions on Medical Imaging 25(8): 9871010.
Ojala, T., Pietikinen, M. & Harwood, D. (1996). A comparative study of texture measures
with classication based on feature distributions, Pattern Recognition 29(1): 5159.
64 Medical Imaging
Olivier, J., Rousselle, J.-J., Bon, R. & Cardot, H. (2008). Active contours driven by supervised
binary classiers for texture segmentation, the 4th International Symposium on Visual
Computing : ISVC 08, Las Vegas, Nevada.
Osher, S. & Sethian, J. A. (1988). Fronts propagation with curvature-dependent speed:
algorithms based on Hamilton-Jacobi formulations, Journal of Computational Physics
79: 1249.
Otsu, N. &Kurita, T. (1988). Anewscheme for practical exible and intelligent vision systems,
IAPR Workshop on Computer Vision, Tokyo, Japan, pp. 431435.
Paragios, N. &Deriche, R. (2002). Geodesic active regions and level set methods for supervised
texture segmentation, International Journal of Computer Vision 46(3): 223247.
Paulhac, L., Makris, P., Gregoire, J.-M. & Ramel, J.-Y. (2009). Human understandable
features for segmentation of solid texture, ISVC 09: Proceedings of the 5th International
Symposium on vision computing, pp. 379390.
Paulhac, L., Makris, P. & Ramel, J.-Y. (2009). A solid texture database for segmentation
and classication experiments, VISSAPP 09: Proceedings of the Fourth International
Conference on Computer Vision Theory and Applications.
Richards, W. & Polit, A. (1974). Texture matching, Kibernetic 16: 155162.
Rosenblatt, F. (1958). The perceptron: A theory of statistical separability in cognitive systems,
Technical report VG-1196-G-1 M. A. Cornell Aeronautical Laboratory.
Sahiner, B., Chan, H.-P., Roubidoux, M. A., Helvie, M. A., Hadjiiski, L. M., Ramachandran,
A., Paramagul, C., LeCarpentier, G. L., Nees, A. & Blane, C. (2004). Computerized
characterization of breast masses on three-dimensional ultrasound volumes, Medical
Physics 31: 744754.
Shamos, M. I. & Hoey, D. (1975). Closest-point problems, the 16
th
Annual Symposium on
Foundations of Computer Science pp. 131162.
Shoshany, M. (2008). An evolutionary patch pattern approach for texture discrimination,
Pattern Recognition 41: 23272336.
Tamura, H., Mori, S. & Yamawaki, T. (1978). Texture features corresponding to visual
perception, IEEE Transactions on Systems, Man, and Cybernetics 8(6): 460473.
Tuceryan, M. & Jain, A. K. (1993). Texture analysis, World Scientic Publishing Co., Inc., River
Edge, NJ, USA, pp. 235276.
Turner, M. R. (1986). Texture discrimination by Gabor functions, Biological Cybernetics
55: 443457.
Valckx, M. J. F. & Thijssen, J. M. (1997). Characterization of echographic image texture by
cooccurrence matrix parameters, Ultrasound in Medicine and Biology 23(4): 559571.
Voorhees, H. & Poggio, T. (1987). Detecting textons and texture boundaries in natural images,
proceedings of the International Conference on Computer Vision (ICCV87), London,
England, pp. 250258.
Xu, C. & Prince, J. (1998). Snakes, shapes, and gradient vector ow, IEEE Transactions on Image
Processing 7(3): 359369.
Zhan, Y. & Shen, D. (2003). Automated segmentation of 3d us prostate images using
statistical texture-based matching method, MICCAI 03: Proceedings of the International
Conference on Medical Image Computing and Computer-Assisted Intervention.
Zhan, Y. & Shen, D. (2006). Deformable segmentation of 3d ultrasound prostate image
using statistical texture matching method, IEEE Transactions on Medical Imaging
25(3): 256272.
Zhang, J. & Tan, T. (2002). Brief review of invariant texture analysis methods, Pattern
Recognition 35: 735747.
Zhu, S. & Yuille, A. (1996). Region competition: unifying snake/balloon, region growing,
and Bayes/MDL/energy for multi-band image segmentation, IEEE Transactions on
Pattern Analysis and Machine Intelligence 18(9): 884900.
Zucker, S. W. (1976). Toward a model of texture, Computer Graphics and Image Processing
5: 190202.
66 Medical Imaging
4
Radiation Protection in Medical Imaging
Horaiu Colosi
1
, Dan Colosi
2
, Vlad Murean
3
and Marius Roman
3

1
Iuliu Haieganu University of Medicine and Pharmacy, Cluj-Napoca
2
State University of New York, School of Dental Medicine, Stony Brook, New York
3
Technical University, Cluj-Napoca
1,3
Romania
2
USA
1. Introduction
Life on Earth has developed continuously in the presence of natural radiation. Radiation is a
form of energy transfer through space and matter. The ability of radiation to transfer energy
at the molecular level to living organisms is at the root of current efforts to understand and
limit the effects of radiation on humans.
The radiation spectrum includes forms of radiation that carry various energy levels.
Radiation forms, such as x-radiation, which carry sufficient energy to overcome binding
energies of electrons and to eject them from their atomic orbits, can cause the formation of
ions in the matter through which they travel, and are termed ionizing radiation. Ionizing
radiation may also dissociate molecular bonds, leading to the formation of free radicals.
1.1 Biologic effects of ionizing radiation
A great proportion of the effects of ionizing radiation on living matter are a result of the
interaction of these forms of radiation with water, a ubiquitous molecule that is present in
most tissues and organs. This interaction leads to the generation of highly reactive radical
species, mainly from the hydroxyl and hydroperoxyl categories, which, in turn, can interact
with cellular DNA, leading to the formation of strand breaks and base damage (Hall &
Giaccia 2005). To a lesser extent, radiation can also directly ionize or damage DNA and
other biologic molecules: RNA, proteins, lipids. Many occurrences of damage limited to one
DNA strand are rapidly detected by the surveillance mechanisms of the cell and the DNA
integrity is restored using the intact complementary DNA strand as template. However,
double-strand breaks, in which an intact DNA template is not available, are more difficult to
repair accurately (Mladenov & Iliakis 2011). As a result, radiation-induced DNA double
strand breaks may often lead to misrepaired DNA, resulting in point mutations, deletions or
chromosomal translocations. Radiation-induced DNA damage that escapes detection by cell
cycle checkpoints during cellular mitosis can result in an uncontrolled, accelerated rate of
mitotic division of the affected cell (Hall & Giaccia 2005), a cellular hallmark of cancer. The
association between exposure to radiation and cancer induction is well documented through
epidemiological studies (Ron 2003).

Medical Imaging 68
1.2 Radiation doses
Various quantities are used to describe the radiation dose to organisms and the resulting
risk. The absorbed dose refers to the radiation energy absorbed by tissues per unit of
mass. It is measured in grays (Gy), with 1 gray equivalent to 1 joule of radiation energy
absorbed per kilogram.
The effective dose accounts for the type of radiation and the biologic tissue that is exposed, by
applying a radiation weighting factor and a tissue-weighting factor to the absorbed dose. The
radiation-weighting factor of x-radiation is one. Tissue weighting factors have been assigned to
various tissues and organs based on their estimated susceptibility to radiation damage. The
effective dose, measured in Sieverts (Sv), is used to estimate risks from radiation exposure. For
exposures to limited parts of the body, as is often the case in medical imaging, the calculated
effective dose is equivalent to the whole body exposure that carries the same risk.

Effective radiation dose
(Sv)
Background Equivalent
(days)
Cosmic radiation 390 59
Radon and other radioactive gases 1260 190
Other terrestrial radiation 770 116
Total natural 2.420 1 year
Medical exposure 600 90
Total annual average dose 3020 455
Table 1. Sources of natural radiation and their average contribution to human exposure
(adapted from United Nations Scientific Committee on the Effects of Atomic Radiation
Report 2008)
Natural radiation contributes approximately 2.4 milliSieverts (mSv) annually to the average
human radiation exposure of 3.0 mSv (United Nations Scientific Committee on the Effects of
Atomic Radiation [UNSCEAR] Report 2008). Main natural sources of ionizing radiation are
cosmic radiation, mainly from the sun and other stars, and terrestrial radiation, originating
from naturally occurring radioactive elements (table 1).
Because the Earths atmosphere acts as a radiation absorber and attenuator, exposure to
cosmic radiation is greatly impacted by altitude. The exposure value to cosmic radiation at
sea level approximately doubles with each 2000 m of elevation. While the yearly exposure to
cosmic radiation at sea level is on average 0.24 mSv, at an altitude of 1600 m it is
approximately 0.5 mSv. Similarly, airplane travel contributes to radiation exposure in the
general population, with doses estimated to approximately 0.025 mSv for every 5 hours
flown at 12,000 m (White & Pharoah, 2009).
Terrestrial radiation includes that originating from radionuclides in the soil, such as
potassium 40, uranium 238 and thorium 232, some of which are taken up by ingestion.
Together, these sources result in an annual dose of approximately 0.8 mSv. However, the
main single source of natural radiation exposure is radon, a gas and uranium decay product,
which contributes roughly 50% of the total natural radiation exposure to the general
population (table 1). Radon may enter buildings through imperfectly sealed foundations

Radiation Protection in Medical Imaging 69
and tends to accumulate in basements, from where it may leak into adjacent living spaces.
Its radioactive decay products are inhaled, taking residence in the respiratory epithelium
and contributing to the occurrence of lung cancer. Radiation exposure from radon varies
widely with the geologic composition throughout the world.
Human activity has led to the development of new sources of radiation over the past decades.
Medical radiation is the greatest peaceful contributor to radiation exposure in the general
population, with an average annual effective dose of approximately 0.6 mSv (table 1).
2. Radiation exposure and current practices for radiation protection
2.1 Radiation exposure in diagnostic imaging
Medical use of radiation has increased continuously throughout the world, irrespective of
the economic level of various regions. While therapeutic use of radiation in the treatment of
cancer contributes a significant amount of radiation to a small number of individuals, the
use of diagnostic radiation imparts a smaller amount of radiation to a large, and constantly
growing, number of individuals.
The development of computed tomography (CT) over the past few decades has
revolutionized diagnostic imaging. Its use has increased exponentially, and it is estimated
that in the United States more than 62 million CT scans have been performed in 2006,
compared to less than 20 million in 1995, and just 3 million in 1980 (Brenner & Hall, 2007).
Due to the relatively higher radiation dose from a typical CT scan compared to that from
conventional imaging modalities (table 2), and due to its widespread and increasing use,
medical CT is currently estimated to contribute a significant amount of radiation exposure
to the general population.

Effective Dose
(Sv)
Background Equivalent
(days)
Head plain radiograph 27 100 4 15
Dental radiographs (set of 19-20) * 33 150 5 23
Chest plain radiographs (PA + lateral) 100 290 15 44
Mammography 230 400 35 123
Abdomen plain radiograph 370 810 56 123
Head computed tomography 900 2800 137 426
Abdomen and pelvis computed tomography 3100 15300 1.3 6.4 years
Table 2. Medical diagnostic imaging procedures and their contribution range to patients
radiation exposure (data: United Nations Scientific Committee on the Effects of Atomic
Radiation Report 2008, * White and Pharoah, 2009)
Due to the large number of CT scans being administered in the present, and the trend of this
number to increase in the future, radiation exposure from CT may gain significance as a
public health concern (Brenner and Hall, 2007).
As a result, current radiation safety efforts carry a significant focus on monitoring and
limiting medical radiation exposure from CT to the general population.

Medical Imaging 70
2.2 Radiation protection measures in diagnostic imaging
The main concern from medical imaging involving ionizing radiation is the risk of cancer
induction. Radiation-induced cancer may occur at any time, including after decades,
following the radiation exposure event, and, if it occurs, it is indistinguishable from
spontaneously induced cancer.
The progressive understanding of biologic effects of radiation led to the development of
guidelines for radiation protection by the international community. Evidence-based
guidelines are outlined in the 2007 Recommendations of the International Commission for
Radiological Protection (International Commission for Radiological Protection, 2007) and in
the recommendations of the National Council on Radiation Protection and Measurements
(NCRP) of the United States (Kase 2004).
The overarching goal of radiation protection in medical diagnostic imaging is to obtain a
maximum of information relevant to the diagnostic task with the least amount of
exposure to ionizing radiation (White & Pharoah, 2009). In practice, economic and
individual factors, such as availability or access to imaging modalities of choice must also
be taken into account.
The application in diagnostic imaging of this principle, also referred to As Low As
Reasonably Achievable (ALARA), involves the implementation of general protective
measures, including the following:
1. Imaging decision. Selection of the imaging examination to answer a specific diagnostic
question should take into consideration all modalities that are likely to provide the
sought information. When the diagnostic question can be answered through a study
that involves no ionizing radiation - such as magnetic resonance or ultrasound imaging
- the added radiation dose from the diagnostic procedure is reduced to zero. The
selection of the imaging modality involving exposure to ionizing radiation should
follow a justification process based on the patients individual clinical circumstances
and on scientific evidence. This process should suggest that the potential diagnostic
benefit of the examination outweighs the estimated risk carried by the specific radiation
exposure. Decision support software, which can assist the clinician in determining the
appropriateness of a given imaging study to the clinical situation, has recently become
available (Rosenthal et al., 2006). Preliminary studies have suggested that its utilization
can contribute to the decreased prescription of examinations that are deemed of low-
utility to the diagnostic task (Coakley et al., 2011).
2. Patient information: Providing easy-to-understand information to patients is an
effective measure for bringing radiation exposure into context and for dispelling
exaggerated fear of radiation in patients. Radiation doses from medical examinations
can be compared with the amount of background radiation to which we are all exposed
by living on Earth (see table 2). Risk comparison with common daily activities, such as
driving or air travel, may also be effective (Coakley et al., 2011). For instance, travelling
64,000 km by car is associated with a risk of death from a motor vehicle accident of
approximately 1:2000 in the United States (National Highway Traffic Safety
Administration Fatality Analysis Reporting System, 2011). This risk is roughly equal to
the risk of developing a fatal cancer from an effective radiation dose of 10 mSv, a dose
five times higher than that from a typical CT scan of the head.

3. Shielding of the examination room, to limit or eliminate radiation exposure to
neighbouring areas;
4. Shielding the patients anatomic areas that are not in the region of interest of the
examination, to reduce radiation exposure to sensitive organs, such as the thyroid,
gonads, eye lens;
5. Filtration of the x-ray beam, which removes low energy photons that would otherwise
be absorbed by the patient, and hence would not contribute to the radiographic image;
6. Optimal film-screen combinations that maximize photon-to-image conversion, while
achieving the image quality required for the diagnostic task;
7. Appropriate x-ray source to-skin distance, whenever this factor is adjustable;
8. Collimation of the exposed field strictly to the region of interest;
9. Optimal combination of exposure factors (kVp, mA, seconds) to achieve low radiation
doses, while maintaining diagnostic image quality;
10. Quality assurance of the imaging and film processing equipment, to ensure consistent
optimal quality of images with minimum exposure to patients and personnel.
11. Personal dose monitors and area dose monitors that assist in documenting individual
exposure and area radiation levels. Personal dose monitors used in clinical practice
include thermoluminescent dosimeters (TLD) and, more recently, optically stimulated
dosimeters (OSLD). TLDs use lithium fluoride or calcium sulphate crystals laced with
magnesium, titanium, copper or phosphorus impurities, which store the energy from
radiation to which they are subjected. When the crystals are heated, the stored energy is
released as luminescence proportional to the radiation exposure. While TLDs are small
and they allow measurements of radiation doses ranging from microGy to kGy,
accurate and reproducible readings are heavily technique-dependent (Kron 1994).
OSLDs use aluminium oxide laced with carbon impurities to store radiation energy,
which is released as luminescence proportional to the radiation exposure in response to
stimulation light. OSLDs have a linear dose-response relationship in the 1-300 cGy
range but can be calibrated for a significantly wider range of radiation exposures that
are relevant in medical imaging (Jursinic 2007, Yukihara, 2008).
Due to the exponential growth in its utilization and its higher radiation effective dose
compared to plain film radiographs, computed tomography (CT) is one of the most
significant contributors to medical diagnostic radiation exposure today. While many of the
general principles outlined above apply to CT as well, several measures specific to CT can
be implemented to reduce patients radiation risk:
1. Bismuth shields are used in the exposure path in front of sensitive organs, such as the
breast. They partially absorb the direct radiation beam to underlying anatomic
structures, without significantly impacting the image quality of adjacent structures.
Bismuth shields can reduce radiation dose to the adult female breast by up to 40%
(Yilmaz et al., 2007). However, when bismuth shields are used together with automatic
exposure modulation, the latter may compensate for the radiation attenuation caused
by the shield, offsetting the radiation savings benefits of the shield. To circumvent this
situation, bismuth shields may be applied after exposure of the scout view, and before
acquisition of the volumetric data (Coakley et al., 2011).
2. Optimization of CT imaging protocols involves tailoring imaging protocols to the
patient and the diagnostic task. The radiation dose can be modified by adjusting one or
several scan parameters:

Medical Imaging 72
i. The maximum tube electric potential (kVp): while the typical exposure setting used
in routine CT scanning is between 100 and 140 kVp, a lower kVP can reduce
radiation dose, albeit at the expense of increased image noise.
ii. Tube current (mA): a CT scan conducted with lower mA results in noisier images,
but radiation exposure is diminished by an amount proportional to the decreased
tube current.
Lower values can be selected for both parameters when scanning children or small
adults, or in cases in which noisier images are unlikely to have a significant impact on
the diagnostic task outcome (e.g. a study to evaluate renal calculi).
3. X-ray tube current modulation and automatic exposure control refer to systems in
which the user (the radiologist, through the radiological technologist) indicates the
desired level of image quality, based on which the CT software calculates the radiation
output required to match that image quality in all regions of the scanned volume
(McCollough et al., 2005). During acquisition of each sectional image, as the gantry
rotates around the patient, angular tube current modulation works to modulate x-ray
tube current to equalize the average photon flux reaching the detectors (x-, y-axis tube
current modulation). As the patient progresses through the gantry, longitudinal (z-axis)
tube current modulation acts to maintain constant image noise levels along the
longitudinal axis of the patient as body regions with different attenuation properties are
scanned (McCollough et al., 2005).
4. Enhanced CT reconstruction algorithms: most current CT scanners utilize
reconstruction algorithms based on the filtered back-projection (FBP) method. When
used in combination with low tube currents, this method tends to generate noisy
images. The implicit radiation exposure threshold, below which the diagnostic image
quality decreases to unacceptable levels, limits the potential for radiation savings with
FBP reconstruction algorithms. Alternative solutions include iterative reconstruction
algorithms, based on complex mathematical models that aim to correct the noise in
images acquired with low tube currents. These algorithms are computation-intensive
and lead to longer reconstruction times (Coakley et al., 2011). A recent, enhanced,
iterative reconstruction method, known as adaptive statistical iterative reconstruction
(ASIR), shows great promise for improved reconstruction speeds. Furthermore, by
attempting to identify and correct for photon flux fluctuations that are unlikely to be
caused by anatomic features, ASIR can reduce image noise, as well as certain types of
image artefacts (Marin et al., 2010, Miville et al., 2011, Renker et al., 2011). As a result,
images reconstructed using ASIR from raw data acquired with low tube current may
show similar or improved image quality compared to FBP reconstructions, while low
current settings allow radiation savings in the range of 32-65% for a variety of clinical
utilization scenarios (Cornfeld et al., 2011, Flicek et al., 2010, Hara et al., 2009). Together,
these characteristics of ASIR protocols may lead to CT studies of overall similar or
better diagnostic quality compared to FBP reconstruction methods, with a potential for
significantly reduced patient radiation exposure.
5. Calculation and reporting of radiation dose: Most CT scanners have the integrated
capability to calculate radiation exposure as a dose-length product (measured in
mGy/cm), which can be recorded in the patients clinical record. This facilitates
tracking CT radiation exposure to patients.

3. Modelling and simulation for radiation control of imaging devices
3.1 Preliminaries
Automated control of radiation field intensity generated during medical imaging and
radiation therapy, constitutes an efficient method to reduce unnecessary radiation exposure
for both patients and device operators (Sajin et al., 2003, McCollough et al., 2006).
This heading presents an original method for automated control of radiation field intensity.
Its implementation in medical imaging devices employing ionizing radiation can offer
important exposure savings to both patients and medical personnel.
The goal of this study has been to develop an original and flexible model for radiation field
control, in order to allow an exact administration of the minimum necessary radiation dose
at any desired tissue depth, even through non-homogeneous tissue layers that intensify or
attenuate the electromagnetic field generated by an imaging device. Therefore, we modelled
and simulated the spread of radiation through non-homogeneous media, as is the case when
using medical imaging devices to investigate human tissues.
A general model of radiation field spread using Cartesian coordinates (0p; 0q; 0r) can be
represented as seen in figure 1. Patterns of spread have been considered, in relation to three
spatial axes and in relation to time (t).

a. y on 0p axis b. y on 0q axis; c. y on 0r axis

Fig. 1. General model of radiation field intensity, y(t, p, q, r) in Cartesian coordinates
Radiation field intensity y(t, p, q, r) shown in figures 1b, 1c, and 1d, can be expressed as:

00 y 0T 0S
y(t,p,q,r)=y [t,s,u(t)]=K F (t)F (s)u(t)

(1)
The output variable y=y[t, s, u(t)] is the intensity of the radiation field, u(t) is the input
(command) signal applied for the radiation source, Ky is a weighting coefficient and (2)
stands for the usual form of compression of the Cartesian space variable on the axes of
coordinates (0p), (0q) and (0r), as illustrated in figure 2.

2 2 2 2 2 2
f f f
f
s = p +q +r ; s = p +q +r

(2)

Medical Imaging 74

Fig. 2. Cartesian coordinates system and Cartesian space variable (s)

2 2 2
0S 0P 0Q 0R
F (s)= F (p)+F (q)+F (r)

(3)
where:

abs (p) abs (p)
- -
P P 1 2
0P
1 2 2 1
1 2
P P
F (p)= +
P -P P -P

(4)

abs (q) abs (q)
- -
Q Q 1 2
0Q
1 2 2 1
1 2
Q Q
F (q)= +
Q -Q Q -Q

(5)

abs (r) abs (r)
- -
R R 1 2
0R
1 2 2 1
1 2
R R
F (r)= +
R -R R -R

(6)
The axes (0p) and (0q) define the horizontal plane, in which the electromagnetic field
generator is located. Field intensity expressed in (1), and axis (0r) correspond to the depth of
field propagation. The constants in (4), (5) and (6) can be approximated by expert
procedures, namely:

f
1
P P
p
P =
(1+ )
;
2 P 1
P = P , where:
P
=(46) and
P
>1 (7)

f
1
Q Q
q
Q =
(1+ )
;
2 Q 1
Q = Q , where:
Q
=(46) and
Q
>1 (8)

f
1
R R
r
R =
(1+ )
;
2 R 1
R = R , where:
R
=(46) and
R
>1

(9)

in which the final abscises
f f f
(p ), (q ), (r ) correspond to negligible values (for example 0.05).
For
0P 0Q
F (p), F (q) ,
0R
F (r) , the abscises
d d d
(p ), (q ), (r ) may show disturbances due to
discontinuities in tissue structure, estimated by:

abs(s) abs(s)
- -
S S 1 2
0S
1 2 2 1
1 2
S S
F (s)= +
S -S S -S

(10)
where

f
1
S S
s
S =
(1+ )
;
2 S 1
S = S

(11)

f f f
S
f f f
P Q R
p +q +r
=
p q r
+ +

;
2 2 2
S
1 1 1
P +Q +R
=
P +Q +R

(12)
Therefore,
0S
F (s) can also be written:

(1+ ) (1+ )
- abs(s) - abs(s)
s s
0S S
S
S S S S
f f
1
F (s)= -
1-
(
(

(13)
or

C p +q +r C p +q +r
0S 0 2
2 2 2 2 2 2
1 3
F (s)=C +C
| |
|
\ .

(14)
in which

0
S
1
C =
1-
;
S S
1
f
(1+ )
C =
s
;
2 S
C =- ;
3 1
S
1
C = C

(15)
The results of
0S
F (s) in (4), (10), (13) and (14) also represent spatial curves deformed by
multiple degrees of freedom, for example
S S
( ), ( ) and
f
(s ) , while the length constants
(P), (Q) and (R) may exhibit inertia or attenuation of radiation field propagation
through different tissue layers.
Function components
0P 0Q
F (p), F (q) or
0R
F (r) can be approximated by:

-K (v -v)
0vd vd
2
vd d
F (v)=A

(16)
in which variable (v) becomes (p), (q) or (r), according to case.
The disturbance index (d) refers to the homogeneity disturbance due to tissue heterogeneity
located on the abscissa
d
(v ) , while
vd
(A ) in (16) represents the amplitude of heterogeneity
illustrated in figure 3.a.
A more or less steep, respectively a more or less symmetrical slope can result from the
convenient choice of parameter
vd
(K ) . The deformation effect of function
0vd
F (v) on the
right abscissa
d
(v ) over the component
0v
F (v) can be seen in figure 3.b.

Medical Imaging 76

a. Amplitude b. Trajectory in Cartesian space
Fig. 3. Tissue heterogeneity may lead to homogeneity disturbances of field intensity
If, with respect to time, the spread of radiation field intensity y(t,p,q,r) is identical on all
three axes (0p), (0q) and (0r) depicted in figures 1a, 1b and 1c, then:

0Tp 0Tq 0Tr 0T
F (t)=F (t)=F (t)=F (t)

(17)
which may also be approximated as a customary form of step response, related to a control
signal u(t) applied to the radiation source.
Thus:

t t
- -
T T 1 2
0T
1 2 2 1
1 2
T T
F (t)=1- -
T -T T -T

(18)
The final values
f f f f
(p ,q ,r ,v ) in figures 1a, 1b, 1c, 2 and 3 show the depth of radiation field
penetration, which will depend on this control signal u(t) and on the more or less
homogeneous structure of the tissue. The length
f
(t ) of this transient phenomenon (figure 4)
reflects a radiation propagation inertia, identical on axes (p, q, r, s) for which
f
t>t tends to a
unitary asymptote.

Fig. 4. Transient phenomenon

For a slow to fast ascending evolution we considered
T T
46, >1 ~ , resulting in the
following time constants:

f
1 2 T 1
T T
t
T = ; T = T ;
(1+ )

1 2 2
i
2 1 1
T T T
t = ln
T -T T
| |
|
\ .

(19)
where (t
i
) is the moment corresponding to the inflexion
0 i
F (t ) .
If, with respect to time, the spread of radiation field intensity y(t,p,q,r) is not identical on all
three axes (0p), (0q) and (0r) depicted in figures 1a, 1b and 1c, then condition (17) is no
longer accomplished, resulting in:

0Tp 0Tq 0Tr
F (t) F (t) F (t) = =

(20)
In this case, formally identical to relation (19), each axis will exhibit:

pf
1p
Tq Tp
t
T =
(1+ )
;
2p Tp 1p
T = T

(21)

qf
1q
Tq Tq
t
T =
(1+ )
;
2q Tq 1q
T = T

(22)

rf
1r
Tr Tr
t
T =
(1+ )
;
2r Tr 1r
T = T

(23)
where (tf), (
T
) and (
T
) are specifically established for each axis.
In order to transform into Cartesian coordinates it can be shown that:

pf qf rf
1p 2p 1q 2q 1r 2r
t +t +t
=
T +T +T +T +T +T
;
2p 2q 2r
1p 1q 1r
T +T +T
=
T +T +T

(24)
finally resulting in:

f
1

t
T =
(1+ )
;
2 1
T = T

(25)
which are the equivalent time constants of the Cartesian space, assuming that inertial
propagations are different along the three axes (p, q, r).
Following this method, F
0T
(t) in equation (1) becomes:

t t
- -
T T 1 2
0T
1 2 2 1
1 2
T T
F (t)= +
T -T T -T

(26)
which together with
0S
F (s) from (10), allows the computation of the radiation field intensity
defined by (1) and depicted in figure (1).

Medical Imaging 78
3.2 Analogical modelling
Figure 5 presents the proposed control block diagram for adjusting radiation fields. Its
components have been modeled by algebraic and differential equation systems (27) and (28),
as well as by partial differential equation (29).

Fig. 5. Control block diagram of the proposed adjustment system

0 0 o
a =w -m ;
m 00 0 m 1
K y =m +T m

(27)

2 2
IR 0 PR 1 DR 2 1 R 2
K a +K a +K a =c +T c a +b ;
E 0 0 E 1
K c =u +T u

(28)

00 00 10 10 01 01 20 20 11 11 02 02 y 0.00 S0 1.00 S1 2.00 S2
a y +a y +a y +a y +a y +a y =K ( u + u + u )

(29)
The indexes in relations (27), (28) and the first parameter in relation (29), correspond to the
order of derivative with respect to time (t). The second parameter in (29) corresponds to the
order of derivative with respect to variable (s), defined in relation (2).
The radiation field intensity (
00
y ) defined by (1) is converted by the measurement
transducer M into electrical signal (
0
m ) found in (27). The control error (
0
a ) results from
(27), in which (
0
w ) is the reference signal.
In the above presented adjustment scheme, the R controller with proportional integrative
derivative (PID) behaviour defined in (28) emits a control signal (
0
c ) which adjusts the
electromagnetic field generator E, as defined in relation (28).
The radiation emitting unit, E, generates the execution signal (
0
u ) representing the incident
radiation field applied to various environments (e.g. air, tissues), together with a resultant
disturbance signal (
0
), resulting in:

S0 0 0
u =u u

(30)
In the above equation, (
S0
u ) is the resulted electromagnetic field and (
0
u ) corresponds to
the homogeneity disturbance F
ovd
(v) from (16), where
d
v=s , which may be defined in
Cartesian space as seen in figure 6 and expressed in (31).


2 2 2
f pf qf rf
t = t +t +t ;
2 2 2
f f f f
s = p +q +r ;
2 2 2
d d d d
s = p +q +r

(31)

Fig. 6. F
OT
and F
OS
including homogeneity disturbance at s
d
Thus, the complex phenomenon of radiation spread can be estimated and adjusted by
partial differential equation (29).
In the left side of this equation we introduced the approximated solution

00 y 00 S0
y [t,s,u(t)]=K F (t,s)u (t)

(32)
where
00 0T 0S
F (t,s)=F (t)F (s) and (
y
K ) is a weighting coefficient.
This resulted for the right side of equation (29) in:

00 00 10 10 01 01 20 20 11 11 02 02 0.00
a F +a F +a F +a F +a F +a F =

(33)

01.00 10 00 20 10 11 01
=a F +2a F +a F

(34)

2.00 20 00
=a F

(35)
in which partial derivatives (
0...
F ), (
11
F ) and (
...0
F ) are obtained from (32). Only signal (
0
u ) is
included in the block diagram. The homogeneity disturbance (
0
u ) is an independent
external signal.
From the equation system (27)(31), the matrix containing partial derivatives of the state
vector (M
dpx
) represented in (36) has been deduced for the entire adjustment system
depicted in figure 5.

Medical Imaging 80

m
0
0
m
1
0
c
0
0
c
1
0
u
0
0
u
1
0
y
00
y
01
y
02
... y
08
y
10
y
11
y
12
... y
18
M
dpx
=
x x
S
=
m
2
0
x
T
x
TS
c
2
0
u
2
0
y
20
y
21
y
22
... y
28
m
3
0
c
3
0
u
3
0
y
30
y
31
y
32
... y
38
...
m
6
0
c
6
0
u
6
0
y
60
y
61
y
62
... y
68
(36)

Because the signals m=m(t), c=c(t) and u=u(t) are functions only of time, all their partial
derivatives with respect to (s) are equal to zero. Details concerning analogical modelling of
systems through
( )
dpx
M may be found in T. Colosi et al., 2009.
3.3 Numerical simulation
Numerical simulation has been performed using
( )
dpx
M and Taylor series, which required
knowledge of initial conditions CI(t
0
,s), for
CI 0 k-1
x =x(t ,s)=x . Hence, by partial derivative with
respect to (s),
S,CI s 0 S,k-1
x =x (t ,s)=x , in which sequence (k-1) corresponds to the moment
k-1
t =(k-1)t , and sequence (k) corresponds to
k
t =kt , where ( t ) is a sufficiently small
integration step.
By means of ( )
CI
x and
( )
S,CI
x we computed the composing elements of ( )
T
x and ( )
TS
x ,
using specific operations based on symbolic derivatives by indices: ( )
T,CI T 0 T,k-1
x =x t ,s =x and
( )
TS,CI TS 0 TS,k-1
x =x t ,s =x .
After setting these initial conditions, we approximated through iterative calculus:

T 6
k k-1 T,k-1
T=1
t
x =x + x
T!
;
T 6
S,k S,k-1 TS,k-1
T=1
t
x =x + x
T!
(37)

For each integration step ( t ) we operated with a number of 20 Taylor series, of which 6 for
(
0k 1k 0k 1k 0k 1k
m ,m ,c ,c ,u ,u ) and 14 for (
00k 01k 06k
y ,y ,...,y ).
It can be observed that 8 elements (signals) included in Taylor series are from the state
vector component ( ) x while the other 12 are from the matrix component ( )
S
x , all of them
included in
( )
dpx
M .
The truncation error at each integration step ( t ) is proportional to
7
(t /7!) , for
(
00
m, c, u, y ), as well as to
6
(t /6!) , for (
10
y ).
3.4 Examples of simulated radiation control
In order to illustrate the radiation control that would result from an electronic
implementation of the above adjustment system, we present the simulated results regarding
field intensity and its automatic adjustment for three simulated tissue types:
1. homogenous tissue
2. heterogeneous tissue with an intensity-increasing homogeneity disturbance located
between body surface and the tissue depth targeted for imaging
3. heterogeneous tissue with both an intensity-decreasing and an intensity-increasing
homogeneity disturbance, located between body surface and the targeted tissue depth
For every case, the intensity and spread of electromagnetic radiation has been simulated
in open loop (without radiation field control) and in closed loop (with radiation field
control).
In all simulated cases the adjustment goal was to ensure a predefined, constant and uniform
intensity level of 5 units (e.g. 5 mGy) at a tissue depth of 50 units (e.g. 50 mm), regardless of
potential intensity disturbers (radiation absorbing or intensifying environments) localized
between body surface and the targeted tissue depth of 50 units.
Numerical simulation was performed using Matlab 7.5.0.
Since our goal has been to develop a general model, we did not impose certain scales or
units of measurement. The scales and units of measurement constitute flexible choices that
may be conveniently defined for every imaging technique for which our method would be
implemented. Given that initialization parameters need to be conveniently chosen and
experimentally calibrated for every targeted imaging technique (CT, cone-beam CT,
conventional radiography, etc.), we will not undergo a presentation regarding the initial
parameters used for these simulated examples. Details regarding the choice of these initial
parameters may be found in Roman et al., 2009, 2010 and Colosi et al., 2010.
Simulation results regarding field intensity have been plotted against time and tissue depth,
as seen in figures 7-12. Colour coding has been employed in order to highlight different
intensity levels of the electromagnetic radiation field.
For every simulated case (figures 7 and 8; figures 9 and 10; figures 11 and 12), significant
changes regarding electromagnetic field intensity at different tissue depth can be observed
between the unadjusted systems and the ones adjusted by the PID controller.

Medical Imaging 82

Fig. 7. Intensity of the electromagnetic radiation field, simulated in open loop (without
radiation field adjustment), for a homogenous tissue.

Fig. 8. Adjustment of the electromagnetic radiation field towards a constant intensity of 5
units at a tissue depth of 50 units, simulated in closed loop, for a homogenous tissue.

radiation field adjustment), for a heterogeneous tissue with an intensity-increasing
homogeneity disturbance at a depth of 15 units.


units at a tissue depth of 50 units, simulated in closed loop, for a heterogeneous tissue with
an intensity-increasing homogeneity disturbance at a depth of 15 units.

radiation field adjustment), for a heterogeneous tissue with an intensity-decreasing
homogeneity disturbance at a depth of 15 units and an intensity-increasing homogeneity
disturbance at a depth of 35 units.
Figures 7, 9 and 11 illustrate the fact that, due to radiation absorption from a dense
environment, an unadjusted systems can not achieve the targeted dose of 5 intensity units at
the desired depth of 50 units, unless a significantly higher field intensity is administered to
begin with.
On the other hand, when the systems are adjusted by the PID controller, the electromagnetic
field intensity at a tissue depth of 50 units is stabilized at the targeted value of 5 units, in all
simulated cases.
Thus, the controlled systems in figures 8, 10 and 12 ensure a minimization of the total
administered radiation dose necessary to achieve 5 intensity units at 50 units of tissue depth.
As may be inferred from the examples above, this general and highly flexible model offers a
foundation for very precise electromagnetic field adjustments. Experimental validation and
calibration for concrete imaging techniques, as well as an electronic implementation of the
model, remain to be performed in order to implement this model for practical use.

Medical Imaging 84

units at a tissue depth of 50 units, simulated in closed loop, for a heterogeneous tissue with
an intensity-decreasing homogeneity disturbance at a depth of 15 units and an intensity-
increasing homogeneity disturbance at a depth of 35 units.
4. Conclusion
The proposed radiation adjustment model offers a precise tool to assist in meeting the
overarching goal of radiation protection: to obtain a maximum of relevant imaging
information, with the least amount of exposure to ionizing radiation.
5. References
Brenner, D.J. & Hall, E.J. (2007). Computed Tomography An Increasing Source of
Radiation Exposure. New England Journal of Medicine, Vol. 357, No. 22, pp.2277-2284,
ISSN 0028-4793.
Coakley, F.V. et al. (2011). CT radiation dose: what can you do right now in your practice?
AJR. American Journal of Roentgenology, Vol. 196, No. 3, pp.619-625, ISSN 1546-3141.
Colosi, H. & Roman, N.M. (2010). Digital Simulation and Control of Electromagnetic Field
Effects in Heterogeneous Organic Tissues, Proceedings of 2010 IEEE International
Conference on Automation, Quality and Testing, Robotics, pp. 321-326, ISBN 978-1-
4244-6722-8, Cluj-Napoca, Romania, May 28-30, 2010
Colosi, T.; Ungurean, M.L.; Dulf, E.H. & Cordo, R.C. (2009). Introduction to Anlogical
Modeling and Numerical Simulation, with (Mpdx) and Taylor Series for Distributed
Parameters Processes, Ed. Galaxia Gutenberg, ISBN 978-973-141-192-7, Trgu-Lpu,
Romania
Cornfeld, D. et al. (2011). Impact of Adaptive Statistical Iterative Reconstruction (ASIR) on
Radiation Dose and Image Quality in Aortic Dissection Studies: A Qualitative and
Quantitative Analysis. Am. J. Roentgenol., Vol. 196, No. 3, pp.W336-340.
Flicek, K.T. et al. (2010). Reducing the radiation dose for CT colonography using adaptive
statistical iterative reconstruction: A pilot study. American Journal of Roentgenology,
Vol. 195, No. 1, pp.126-131, ISSN 1546-3141.
Hall, E.J. & Giaccia, A.J. (2005). Radiobiology for the Radiologist. (6th ed.), ISBN 978-
0781741514 Lippincott Williams & Wilkins Publishing, Philadelphia, PA.

Hara, A.K. et al. (2009). Iterative reconstruction technique for reducing body radiation dose
at CT: feasibility study. AJR. American Journal of Roentgenology, Vol. 193, No. 3,
pp.764-771, ISSN 1546-3141
International Commission on Radiological Protection, 2007. The 2007 Recommendations of
the International Commission on Radiological Protection. ICRP publication 103.
Annals of the ICRP, 37(2-4), pp.1-332.
Jursinic, P.A., 2007. Characterization of optically stimulated luminescent dosimeters, OSLDs,
for clinical dosimetric measurements. Medical Physics, 34(12), pp.4594-4604.
Kase, K.R., 2004. Radiation protection principles of NCRP. Health Physics, 87(3), pp.251-257.
Kron, T., 1994. Thermoluminescence dosimetry and its applications in medicine--Part 1:
Physics, materials and equipment. Australasian Physical & Engineering Sciences in
Medicine / Supported by the Australasian College of Physical Scientists in
Medicine and the Australasian Association of Physical Sciences in Medicine, 17(4),
pp.175-199.
Marin, D. et al. (2010). Low-Tube-Voltage, High-Tube-Current Multidetector Abdominal CT:
Improved Image Quality and Decreased Radiation Dose with Adaptive Statistical
Iterative Reconstruction AlgorithmInitial Clinical Experience1. Radiology, Vol.
254, No. 1, pp.145 -153
McCollough, C.H., Bruesewitz, M.R. & Kofler, J.M., Jr, (2006). CT dose reduction and dose
management tools: overview of available options. Radiographics: A Review
Publication of the Radiological Society of North America, Inc, Vol. 26, No. 2, pp.503-512,
ISSN 1527-1323
Miville, F.A. et al. (2011). Paediatric cardiac CT examinations: impact of the iterative
reconstruction method ASIR on image quality - preliminary findings. Pediatric
Radiology. Available at: http://www.ncbi.nlm.nih.gov/pubmed/21717165
[Accessed July 1, 2011]. ISSN 1432-1998
Mladenov, E. & Iliakis, G. (2011). Induction and repair of DNA double strand breaks: The
increasing spectrum of non-homologous end joining pathways. Mutation Research,
Vol. 711 No. 1-2, pp.61-72, ISSN 0027-5107
National Highway Traffic Safety Administration Fatality Analysis Reporting System,
Fatality Analysis and Reporting System Encyclopedia. National Highway Traffic
Safety Administration Fatality Analysis Reporting System. Available at: http://www-
fars.nhtsa.dot.gov/Main/index.aspx [Accessed July 1, 2011].
Renker, M. et al. (2011). Evaluation of Heavily Calcified Vessels with Coronary CT
Angiography: Comparison of Iterative and Filtered Back Projection Image
Reconstruction. Radiology. Available at:
http://www.ncbi.nlm.nih.gov/pubmed/21693660 [Accessed July 1, 2011], ISSN
1527-1315
Roman, N.M.; Colosi, H. & Pusca, M. (2009). Digital Simulation for Computer Control of
Radiological Devices: A Preliminary Model Using Partial Differential Equations,
Proceedings of MediTech 2009 International Conference on Advancements of Medicine and
Health Care through Technology, pp. 37-42, ISBN 978-3-642-04291-1, Cluj-Napoca,
Romania, September 23-26, 2009
Roman, N.M.; Colosi, H. & Pusca, M. (2010). Numerical Simulation Using Partial
Differential Equations for Modeling and Control of Medical Radiation Fields.

Medical Imaging 86
Annals of the Academy of Romanian Scientists, Series on Science and Technology of
Information, Vol.3, No.1, (2010), pp. 59-74, ISSN 2066-2742
Ron, E. (2003). Cancer risks from medical radiation. Health Physics, 85(1), pp.47-59, ISSN
0017-9078
Rosenthal, D.I. et al. (2006). Radiology order entry with decision support: initial clinical
experience. Journal of the American College of Radiology. Vol. 3, No. 10, pp.799-806,
ISSN 1558-349X
Sajin, G.; Sajin, M. & Gavriloaia, G. (2003). Aplicaii biologice ale radiaiilor electro-magnetice,
Ed. Academiei Tehnice Militare, ISBN 973-640-010-7, Bucureti, Romania
United Nations Scientific Committee on the Effects of Atomic Radiation (2010). Sources and
Effects of Ionizing Radiation. [UNSCEAR] 2008 Report to the General Assembly with
Scientific Annexes (2010): Vol. 1 Annex A. United Nations Publications, ISBN 978-92-
1-142274-0 New York, NY
United Nations Scientific Committee on the Effects of Atomic Radiation (2010). Sources and
Effects of Ionizing Radiation. [UNSCEAR] 2008 Report to the General Assembly with
Scientific Annexes (2010): Vol. 1 Annex B. United Nations Publications, ISBN 978-92-
1-142274-0 New York, NY
White, S.C. & Pharoah, M.J. (2009). Oral Radiology - Principles and Interpretation (6th ed.),
Mosby Elsevier ISBN 978-0323049832 St. Louis, MO
Yilmaz, M.H. et al. (2007). Female breast radiation exposure during thorax multidetector
computed tomography and the effectiveness of bismuth breast shield to reduce
breast radiation dose. Journal of Computer Assisted Tomography, Vol. 31, No. 1,
pp.138-142, ISSN 0363-8715
Yukihara, E.G. & McKeever, S.W.S., 2008. Optically stimulated luminescence (OSL)
dosimetry in medicine. Physics in Medicine and Biology, 53(20), pp.R351-379.
Part 2
Image Processing Techniques

0
Current Trends in Archiving and
Transmission of Medical Images
Lus S. Ribeiro, Carlos Costa and Jos Lus Oliveira
Universidade de Aveiro, IEETA
Portugal
1. Introduction
The traditional Picture Archiving and Communication System (PACS) model consists of
one PACS serving one healthcare institution (i.e. hospital-centered). Typically, a healthcare
institution does not possess all the modalities or experts to evaluate a patient. Therefore,
patients tend to move across several healthcare institutions to undergo all the required exams
or diagnosis. These high levels of patient mobility produce huge amounts of medical data
spread among different healthcare institutions without being shared conveniently, making the
traditional PACS model obsolete. The current solutions for sharing non-anonymous medical
images (i.e. HIPAAs Protected Health Information - PHI) rely mainly on point-to-point trust
relationships (e.g. the radiologist trusts the physician sending image reports by e-mail) or
ad-hoc solutions between few institutions (Jacq, 2007). The main problem of the ad-hoc
solutions is interoperability, caused by the heterogeneity of methods to exchange information.
To overcome this and other issues, the health industry, research and professionals around
the world joined forces and started an initiative entitled Integrating the Health Enterprise
1
(IHE) with the main purpose of dening which standards (e.g. DICOM, HL7, ISO, OASIS, etc)
may be used in a given situation of the healthcare workow. IHE does not design standards,
but denes integration proles, i.e. blueprints that describe real-world scenarios or specic
characteristics for building integration-ready systems (Strkson & Aslaksen, 2009). Among
several integration proles, one stands out: the Cross-Enterprise Document Sharing (XDS).
XDS is gaining momentum and nowadays there are several implementations working in the
eld. XDS for imaging (XDS-I) is a content prole based on XDS that takes into consideration
the particularities of the medical imaging eld. However, XDS or XDS-I assume that their
architectural components are located inside trustworthy domains, i.e. owned and maintained
by Healthcare institutions. Planning and maintaining the IT infrastructure required to support
the XDS-I architecture is not simple and demands a signicant human and nancial effort.
Therefore, it would be desirable to delegate this task to a third-party entity and pay for it as a
service. For instance, delegating the IT infrastructure to a Cloud Computing provider where
the healthcare institutions would just pay to use the data-sharing service and not for the entire
IT infrastructure required to support XDS.
Cloud is a computing paradigm that intends to deliver computation and data storage as a
utility service (Faruqui, 2005; Michael Armbrust, 2009; Rajkumar Buyya, 2009). A utility
1
http://www.ihe.org
5
2 Medical Imaging
service in the Cloud follows the same approach as a utility service in currently established
utility facilities (e.g. electricity, gas, or telecommunications). Utility services are accessed
so frequently that they need to be available anywhere and whenever the consumer requires
them. This approach brings obvious advantages since consumers no longer need to plan,
invest heavily at the outset or maintain a complex IT infrastructure.
However, combining the concepts of PACS/XDS-I and Cloud computing raises other
problems, mainly regarding protection of the patients private information fromunauthorized
entities. In this chapter we intend to: (i) - describe the XDS-I and Cloud computing; (ii) -
highlight the challenges and opportunities of combining these two concepts; (iii) - present
several complementary solutions allowing discussion of whether the benets of outsourcing
the XDS infrastructure on Cloud are worth the associated risk.
2. Background
Medical imaging is a non-invasive technique used to create internal images of the human
body for clinical or medical science purposes (i.e. virtual dissecting of the human body).
The genesis of medical imaging occurred in the nal decade of the 19th century (1895), when
Professor Wilhelm Roentgen noticed electromagnetic radiation while performing vacuum
tube experiments. Not understanding the plenitude of that radiation he decided to call
it x-rays (Roentgen, 1898). After these rst steps, radiology evolved at a good rate until
World War II. The intense use of x-rays during the Second World War, and the arrival of
the digital computer and new modalities such as ultrasound and magnetic resonance have
triggered a boom in diagnostic imaging techniques in the past years (Hendee & Ritenour,
2002). Digital imaging techniques have been in use since the 1970s after the clinical acceptance
of Computer Tomography (CT scanner). Currently, digital medical imaging technology is
globally acknowledged and an important part of the healthcare workow.
2.1 Current medical imaging scenario
Nowadays, the importance of medical imaging in the healthcare system is irrefutable. To
physicians it represents a key factor in supporting their clinical thesis and, as a result,
delivering high quality healthcare decisions. Images and studies are stored in local
repositories following a PACS concept (Huang, 2010). PACS embraces a set of technologies for
the archiving, distribution, visualization and acquisition of medical images over a computer
network. Compared with the traditional analogue lm, the PACS concept brings signicant
benets for the productivity, economy and management of a healthcare institution (De Backer
et al., 2004; Huang, 2010; Langer, 2009). At the present time, PACS is a widespread concept
in the majority of medical centers. This acceptance was encouraged by introduction of the
DICOM, a standard for handling, storing, printing and transmitting medical images. It
includes data format denition, storage organization and a network communication protocol
(Costa et al., 2007; Mustra et al., 2008; Pianykh, 2008). In this way, PACS devices from different
vendors are able to interact with each other in a transparent manner. The PACS architecture
began mainly on an ad-hoc basis, serving small subsets, called modules, of the radiology
department. Each module functioned as an independent island, unable to communicate
with other modules (Huang, 2010). Later it evolved into a PACS infrastructure solution,
integrating the hospital information system and the radiology information system, serving
the entire hospital (gure 1). The core element of a PACS is one (or more) archive that
holds all the DICOM images and studies. Although the various PACS archives that serve
90 Medical Imaging
Transmission of Medical Images 3
the institution are typically accessible within the institution, they tend to be independent
from each other. This may be useful to create federations of clinical specialities within the
institution, but if a workstation needs to access the full picture of one patient it has to query
archive by archive. Furthermore, when the central PACS archive of the healthcare institution
or department deteriorates (e.g. not capable of delivering an acceptable Quality of Service),
the institution typically replaces it with a more powerful machine (scales up) or appends a
new and independent PACS archive, bringing problems regarding data migration and lack of
a unied view within the archive of the same department. If inside the healthcare institution
such problems still persist, the magnitude and complexity of the problem amplies when
we reach the inter-institutional document-sharing level. Medical imaging exchange among
different institutions and e-health in general poses a range of legal and ethical challenges, such
as ownership, condentiality, privacy and integrity of medical data or licensure, accreditation
and liability of the health professional or institutions. Legal issues have been a major barrier
to inter-institutional medical imaging exchange (Pattynama, 2010). The transmission of PHI
across different institutions is not just transferring data from one spot to another. A major
problem regarding inter-institutional cooperation is the trust establishment (Lovis et al., 2007;
Ruotsalainen, 2009). Due to the rigid laws regarding data privacy and security protection,
institutions are reluctant to exchange sensitive data such as non-anonymous medical images.
Compromising sensitive data is a present risk when data leaves the institution walls because
the institution that controls the data (data controller) may be accountable for any malpractice
performed on it, even if the data is sent to other institutions and the malpractices were
performed there (Mora et al., 2008). This scenario leads to institutional closure and prevents
inter-institutional cooperation due to the lack of trust in third party institutions. Therefore,
in order for a group of healthcare institutions to cooperate and share clinical data they must
trust each other in the rst place. Moreover, besides trusting each other, they must trust in the
IT infrastructure that supports the medical imaging cross-transmission.
RIS
DICOM Print
DICOM Print
Media Storage
Retrieve
Query
Image Archive ModaIity ModaIity
DICOM Storage
Storage Commitment
M
o
d
a
Iity
W
o
r
k
Iis
t
M
o
d
a
Iity
W
o
rk
Iis
t
M
P
P
S
DICOM Storage
MPPS
WorkStation
Printer
Fig. 1. PACS traditional workow.
91 Current Trends in Archiving and Transmission of Medical Images
4 Medical Imaging
2.2 Integrating the healthcare enterprise
As mentioned before, DICOM supports interoperability inside the healthcare institution, but
inter-site interoperability is a different situation. Ideally, an authorized entity would have
access to the entire clinical history of the patient, relevant for one specic episode, whether
or not the exams were performed by the institution. IHE aims to make this possible by
improving system integration and eliminating barriers to achieving optimal patient care.
IHE provides integration proles, every one composed of several actors and transactions.
The IHE actors and transactions are abstractions of the real-world healthcare information
system environment. While some of the transactions are traditionally performed by specic
product categories (e.g. HIS, Electronic Patient Record, RIS, PACS, Clinical Information
Systems or imaging modalities), IHE intentionally avoids associating functions or actors with
such product categories. For each actor, the IHE denes only those functions associated
with integrating information systems. The IHE denition of an actor should therefore not
be taken as the complete denition of any product that might implement it, nor should
the framework itself be taken to comprehensively describe the architecture of a healthcare
information system. The reason for dening actors and transactions is to provide a basis for
dening the interactions among functional components of the healthcare information system
environment (IHE, 2006a).
2.2.1 Cross-Enterprise Document Sharing for imaging
Cross-Enterprise Document Sharing (XDS) is IHEs integration prole that provides core
guidelines for sharing documents among any healthcare institution, more precisely it
supports: querying, retrieving, publishing and registering Electronic Health Records (EHR)
documents. However, XDS is content neutral this means that its architecture was designed to
support any type of EHR document. This document independence allows XDS to be a generic
framework and more resilient to the appearance of future document standards or formats.
In the other hand, being so generic could raise challenges dealing with more specialized
domains (e.g. radiology, cardiology). Therefore, it is possible to extend the XDS integration
prole and create more specic proles, entitled content proles. XDS for Imaging (XDS-I)
is a content prole scoped to the medical imaging domain, where systems like PACS, RIS or
DICOM objects are taken into account on the XDS architecture.
XDS-I prole facilitates the registration, distribution and access of medical images and
imaging related documents across multiple healthcare institutions. Its focus is to provide a
standard-based specication for managing the sharing of documents between any healthcare
provider, ranging from a small physician ofce to a metropolitan Hospital (IHE, 2006a; 2008).
XDS-I may be seen as a content prole of the integration prole XDS specialized in the
radiology domain. Like XDS, XDS-I is document type independent however XDS-I takes in
consideration some radiology particularities such as the integration of PACS and RIS in the
conceptual architecture.
XDS, and as a consequence XDS-I, assumes that the healthcare institutions belong to one,
or more, XDS Afnity Domains (XAD). XAD is a community of healthcare providers that
agreed to cooperate using a common set of policies, share a common infrastructure of
repositories and a common document registry. Inside an Afnity Domain policies must be
dened, such as patient identication (e.g. using IHE Patient Identier Cross-Reference - PIX),
control of access, security model, as well as the format, content, structure, organization and
92 Medical Imaging
Document
Consumer
Document
Registry
Registry Stored Query [ITI18] Patient Identity
Source
Patient Identity Feed [ITI8]
Consumer
Imaging
Document
Consumer
Registry
Register
Document Set [ITI42]
R t i D t id & i i
Source
Document
Repository
Imaging
Document
Source
Retrieve Document
Set [ITI43]
Provide & Register Imaging
Document Set [RAD68]
WADO Retrieve [RAD55]
Retrieve Imaging Document Set [RAD69]
Source
Retrieve Images [RAD16]
Retrieve Presentation States [RAD17]
Retrieve Reports [RAD27]
Retrieve Key Image Note [RAD31]
Retrieve Evidence Documents [RAD45]
Fig. 2. Diagram of the XDS-I content prole with respective actors and main transactions.
representation of the clinical information (IHE, 2006b). Although XAD is a trust community
and authorized participants may access the documents, the healthcare provider that produces
the documents may select which documents are to be shared and which are just for internal
use.
As mentioned before, the IHE actors communicate through well dened IHE Transactions.
These Transactions are based on ebXML messages developed by the OASIS standards
2
. So,
the exchanged XDS documents are wrapped in ebXML. The receiving actor by analyzing
the message will know what type of information the message contains and who the sender
entity was. Figure 2 shows the actors taking part in the XDS-I prole and the transactions
between the actors. The Imaging Document Source is the actor that provides or holds the
clinical documents (e.g. PACS) typically within the healthcare institutions. It is responsible
for pushing the documents and its metadata to the respective Document Repository. If the
document is a DICOM object the Repository will only hold the DICOM Key Object Selection
(KOS). The KOS objects are small DICOMobjects containing a list of UIDreferences (instead of
the image data itself) in order to the Document consumer retrieve the images fromthe Imaging
Document Sources. Besides being responsible for the persistent storing of the documents,
the Document Repository actor is also responsible for registering the documents metadata
in the appropriate Document Registry. The Document Registry actor is the central player
of the entire XDS Afnity Domain. It maintains the metadata of each registered document
and its mapping to the respective Document Repository, where the actual document is stored.
Furthermore, the Document Registry responds to queries from the Document Consumer actor
with the locations of the matching documents. Moreover, the Document Consumer actor
queries the Document Registry to nd the location and the identier of the document. With
this information, it contacts the respective Imaging Document Source (or Sources) requesting
the access to the respective image (or images). If the Imaging Document Source authorizes
the access, the set of images are typically retrieved through the WADO retrieve protocol - a
sub-protocol of the DICOM standard. Finally, Patient Identity Source from the PIX integration
prole, and not from XDS, is the actor that provides a unique patient identier within the
Afnity Domain.
2
http://www.oasis-open.org
6 Medical Imaging
Usage example
Figure 3 illustrates a real world application of the XDS-I approach on the inter-institutional
cooperation within an Afnity Domain. At this example, the XAD is composed by a
shared document registry and each institution owns a document repository and an imaging
document source within its walls. This example highlights possible interactions between
professionals, institutions and patients in an inter-institutional clinical episode enabled by
the XDS-I content prole:
Physician Ofce: A referring physician, working in a private ofce, orders one
examination and the patient goes to the Imaging Acquisition Center to perform the exam.
Imaging Acquisition Center: An imagiology institution with modality equipment and a
RIS/PACS to manage report and imaging information. The healthcare professional queries
the Document Registry for previous studies relevant to this clinical episode. Then, the
images studies (acquisition and report) are preformed and, nally, the new documents are
registered on the Document Registry.
Diagnostic Center: Eventually, the private physician determines that a consult with
a specialized physician is required after analyzing the new exams. At the healthcare
institution with specialized physicians (e.g. oncologist) the physician queries the
document registry and fetches, from the distributed document repositories, all the needed
documents. It performs the evaluation reports of the clinical episode and, nally, registries
the documents on the Document registry of the Afnity Domain.
XDS Document Registry: The Document Registry it is the heart of the afnity domain. It
is this entity that enables the lookup of documents by indexing the documents metadata.
Document
Registry
Q
u
e
r
y

D
o
c
u
m
e
n
t
Imaging Study
P
r
i
o
r

I
m
a
g
i
n
g

S
t
u
d
i
e
s
P
r
i
o
r

I
m
a
g
i
n
g

S
t
u
d
i
e
s
Query
Document
Query
Document Provide Imaging
Information for Sharing
Diagnosic Centre
Physicians Office
Imaging Acquisition
centre
Provide
Imaging
Information
for Sharing
Fig. 3. Data ow within an XDS-I Afnity Domain for a clinical consult.
94 Medical Imaging
2.2.2 Cross-community access for imaging
In the last section, the reader learned how XDS enables the creation of trust communities for
document sharing. However, the scenario of a unied cluster of healthcare institutions sharing
documents is not always possible. As mentioned before, the participants of a XAD have to
agree a priori to several matters that might slow down the creation of XAD or even make it
impossible. For instance, XDS assumes that the patient identiers are globally unique inside
the XAD. However, each institution has their own patient identiers and the effort of merging,
or mapping, the identiers into a master patient identier can be great and, as a consequence,
jeopardize the project. Whether for technical or conjectural reasons it is certain that patients
will ow across several XADs or independent care communities. Therefore, having a way
to access documents from outside the community is a desirable feature. Foreseeing this
scenario, IHE proposed the Cross Community Access (XCA) integration prole that enables
independent communities to exchange documents.
The Cross-Community Access (XCA) integration prole was designed to complement XDS. If
XDS creates islands of institutions and XCA creates bridges between those islands. In other
words, XDS provides the blueprints for building concise domains of document sharing. While
on the other hand, XCA denes a way for extra-domain entities to access the documents
shared within the community.
Although XCA does not specify that communities must be XDS Afnity Domains, the reality
is that its design makes it the natural choice to connect XDS Afnity Domains. However,
the actions allowed to an entity using XCA are not the same as if the entity belonged to the
XAD. XCA only allows search and retrieval of documents among communities and does not
support pushing or registering of documents inside those communities.
2.3 Cloud computing
Cloud computing is a new buzz word for an old dream of computing as a utility, more
precisely the 5th utility after water, electricity, gas and telecommunications. Cloud computing
does not stand for a completely new concept, as several computing paradigms have promised
to deliver this utility computing vision such as Cluster computing, Grid computing, and more
recently Cloud computing. A utility service in the Cloud follows the same approach as a
utility service in the current established utility facilities (e.g. electricity). Clouds business
model is based on Economies of Scale where efciency of the provided service increases as
the number of services being delivered increases. Hence, the average unitary cost of the
service decreases, because the xed costs of the service are shared over the increased number
of provided services. For instance, state of the art magnetic resonance (MR) modalities have
high purchase costs; if every healthcare institution (independently of its dimension) had a
MR modality, the unitary price of each exam would be higher because there would not be
enough patients to ll the modalitys schedule. Therefore, the usage efciency of the provided
service would be low and, as a consequence, the service would have to be more costly. As
result, typically there are imaging centers that own several modalities and sell the service to
the patients of nearby healthcare institutions. This scenario brings several advantages: (i) -
healthcare institutions do not need to invest heavily up-front in modalities to conduct their
healthcare core business; (ii) - the institutions do not need to maintain the modalities; (iii)
- there is more efcient use of the modality and therefore the costs of patients exams may
become lower.
8 Medical Imaging
In addition to economic advantages, the quality of service is a major incentive to use
Cloud services: high availability, high reliability and high scalability (Binnig et al., 2009).
Technologically, Cloud may be seen as a Grid with a different business model, managed by a
single entity and with a virtualization strategy (Foster et al., 2008; Rajkumar Buyya, 2009). Its
economic approach could make Cloud computing more sustainable than the Grid in the long
term, because it is driven by economic goals and does not rely on uncertain and ephemeral
project founding. As a consequence, Cloud computing could be seen as the next step of the
Grids technology to deliver computing as the 5th utility (Rajkumar Buyya, 2009; Rimal et al.,
2009).
Cloud computing relies on virtualization. Virtualization ts the extremely dynamic Cloud
environment very well. With it, computing environments may be dynamically created,
expanded, shrunk, replicated or moved according to demand (Rimal et al., 2009). With
virtualization, it is possible to easily build scalable and fault-tolerant systems according
to the quality of service purchased by the cloud consumer. Virtualization is boosted by
the increasing ability of hardware to run applications within Virtual Machines efciently
(Rajkumar Buyya, 2009), more precisely the recent advance in the eld of multi-core
microprocessor. Cloud Computing stands for the applications, the hardware resources and
everything in between is delivered as a service.
Application
a
a
S
S
a
a
S
I
a
a
S
P
a
Platform
Storage Databases Signalling
Storage Resources Networking Servers
Infrastructure
Virtualization
Storage Resources Networking Servers
Hardware
Fig. 4. Cloud Computing layers.
Figure 4 illustrates Clouds abstract layers (Fu & Chen, 2010):
Hardware layer: refers to the actual physical machines that compose the data center.
Virtualization layer: the hardware resources of the below layer are shared among all
the virtual machines supported (i.e. the hardware capabilities are sliced into virtual
machines). Each virtual machine has a self-contained operating system and may be seen
as an individual and independent machine.
Infrastructure layer: Known as Infrastructure as a Service (IaaS) extends the services
provided by the virtualization layer. It provides the mechanisms to manage, monitor, and
congure all the supported virtual machines into a utility computing manner. IaaS is the
96 Medical Imaging
abstraction allowing access to infrastructure resources (e.g. storage, computation, server,
data center) on-demand and paid according to the required quality-of-service (QoS).
Platform layer: Known as Platform as a Service (PaaS) extends the IaaS layer by hiding
the IaaS complexity. PaaS is accessed as one big system and not by accessing individual
virtual machines. Therefore, it provides an abstraction where the virtual machines are
automatically managed by the cloud service provider. Offering reliable services by default
such as storage, databases and signalization.
Application layer: Layer usually entitled as Software as a Service (SaaS) where the
applications hosted on the Cloud are deployed on local computers typically through
web-browser. On this layer, the payment is associated with the application itself and not
with the platform or infrastructure below it. Typically, this layer is where developers build
their applications.
The services themselves have long been referred to as Software as a Service (SaaS) and Cloud
does not change that (Michael Armbrust, 2009). Nevertheless, Cloud Computing allows the
application providers the choice of deploying their product as SaaS without providing a data
center.
3. Challenges
Cloud promises to deliver nancial benets to enterprises by reducing the costs of the IT
department. The IT infrastructure supporting the storage of medical images is a heavy burden
for healthcare institutions. Besides storing the clinical data in a reliable manner, it is also
required to make that data available 24/7 and provide appropriate access performance - a
true management nightmare (Huang, 2010). When the scope of the IT infrastructure ranges
from an intranet to a cross-enterprise solution, the associated costs, complexity and effort
of deployment also increase. Regarding the nancial dimension, Clouds business model
is attractive to build cross-enterprise solutions, such as XDS-I, due to the fact that the costs
associated with components shared by the community (e.g. XDS Document Registry, XDS
Document Repository) would be paid as a service and according to each institutions use.
In this way, the signicant costs of developing and maintaining such infrastructure would be
eliminated. Unfortunately, the outsourcing of clinical data to the Cloud faces many challenges
that must be considered before moving the clinical data or components of the IT infrastructure
to the Cloud.
3.1 Privacy and condentiality
The privacy and condentiality of the data placed on the Cloud are the main barriers that
delay Clouds general acceptance. The World Privacy Forum (WPF) released a report spelling
out the risks to privacy and condentiality posed by Cloud computing. The report unveils
concerns regarding Clouds current state: what happens to the data after the consumer
uploads it to the Cloud and how its condentiality will be protected. The answer to this
question is somehow disappointing. In the current state of Cloud computing it does not
ensure privacy or condentiality of the stored data. The simple fact of uploading data to
the Cloud makes that data more suitable for disclosure or unauthorized usage. The WPF
published several privacy advices in order to aware the cloud consumers:
Read the Terms of Service before placing any information in the cloud. If the Terms of
Service are not understandable, a different cloud provider must be considered.
10 Medical Imaging
Information in the cloud is legally easier to access through the Cloud than by seizing a
private computer. Therefore, sensitive information should not be uploaded to the Cloud.
In the Term of Services one must notice if the Cloud provider reserves rights to use,
disclose, or make the uploaded information public.
Read the Privacy Policy before placing any information in the cloud. If the Privacy Policy
is not understandable, a different cloud provider must be considered.
Beware if the Cloud provider retains rights over removed data by the consumer.
Beware if the cloud provider noties their consumers when the Terms of Service or Privacy
Policy change.
Furthermore, WPF extends its advices to companies or governments that are considering the
upload of data or the migration of the IT infrastructure to the Cloud:
Caution on ad-hoc Cloud computing is advised. Organizations should have standardized
rules for employees to know which data they may (or not) upload to the cloud.
Sensitive information that is of the interest of the organization to keep away from
the government, other competitive organizations or other governments should not be
uploaded.
Information disclosure of clouds data should be considered before uploading the actual
data.
Hire professional support for understanding the Terms of Service or Privacy Policies of the
Cloud provider.
After analyzing these advices, answering the question: Is the current state of Cloud computing
suitable for storing any PHI? The answer is pretty straightforward: No! Any cloud consumer
should consider that any uploaded data could be used by others e.g. statistics, publicity or
even be sold to competitors. For instance, health insurance companies may buy information
regarding the health history of a patient in order to assert fees or deny the health insurance
to the patient. At this case, the patient privacy right was jeopardized and there was a
condentiality breach. Even if the uploaded medical data is anonymous for safeguarding
the patients privacy, it is still possible to infer statistical analysis to the data identifying
geographical regions propitious to certain pathologies and sell that information to health
insurance companies. Terms of Service and rigid Privacy Policy agreements may raise the
protection of the stored data by adding liability to the Cloud provider. However, is extremely
difcult to prove that the stored data was violated, or not, by the Cloud provider. Therefore,
Cloud computing, without any privacy and condentiality protection system built over the
Cloud, is not suitable to store any data undesirable to be disclosed.
3.2 Interoperability and standardization
Although some efforts towards reaching standardization among for Cloud Computing, such
as the Cloud Computing Interoperability Forum (CCIF) or the European Telecommunications
Standards Institute (ETSI), the reality is that cloud standardization is far from being achieved
and at some levels maybe it never will, as a consequence, the lack of interoperability
inter-cloud provider is a resilient issue. If it is possible and relatively easy to guarantee
cloud interoperability at the IaaS level even without standardization (e.g. DeltaCloud
3
),
3
http://incubator.apache.org/deltacloud/
98 Medical Imaging
accomplishing interoperability between the several PaaS is a more demanding task. Not only
due to possible economical interests of the Cloud providers, but as well at the technical level.
PaaS APIs abstract the complexity of the IaaS and provides the developers with embedded
functionalities (e.g. automatic scaling) that at the IaaS level have to be implemented from
scratch . This facilitated API of PaaS is more convenient for the developer but less exible at
the same time (Michael Armbrust, 2009). PaaS automatic features turn the standardization or
interoperability efforts more complex. Each Cloud provider follows its unique IT architecture
- especially above the IaaS level (PaaS or SaaS). If some features are easy to accomplish in
some cloud architectures, the same features could be extremely difcult to achieve at other
architectures. This heterogeneity of features and architectures may push back standardization,
and turn full interoperability above the IaaS level extremely difcult.
3.3 Geographical distribution
The Cloud providers are private companies adjudicated to a country obligated to follow the
countries laws. However, they compete with each other on a global market, ignoring countries
borders. Enabled by the Internet, a Cloud provider from the USA may easily offer its Cloud
services to a consumer from Singapore or Australia. At the same level, a Cloud provider
may have several data centers around the world. For instance, one of the major expenses of
a data center is the energy consumed for refrigerating the machines and, if the data center
is placed in a cold natural geographical location (e.g. Alaska) the cooling costs will be much
lower and the Economics of Scale enhanced. However, this cross-border logistics of the Cloud
business model may aggravate liability and jurisdiction risks, due to the fact that cross-border
litigation is typically more complex and high-staked than litigations within borders, and is
not clear which court have jurisdiction over the Cloud where the illegality occurred: would it
be treated in the jurisdiction of the consumer, the provider or the vendor?
3.4 Consumer lock-in and bankruptcy
Consumer lock-in and bankruptcy of the Cloud provider are two serious risks and both have
a similar consequence, which is losing control of the data trusted by the cloud consumer to
the cloud provider. Consumer Lock-in may be performed at two different levels: data lock-in
and vendor lock-in. Trusting valuable data to a single provider may lead to opportunistic
reprising, where the provider uses the held data at its data centers to blackmail the costumer
and rise prices while renegotiating contracts. Vendor lock-in is more subtle type of consumer
lock-in. Due to the fact that each Cloud provider offers, to its customers, a unique
development API which turns the applications specic to the Cloud provider. Therefore,
the consumer is locked-in with the Cloud provider since migrating the applications to other
Clouds means recoding the applications following the new cloud API. As a result, the cost
of migration does not compensate the eventual competitive prices of other Clouds. As it
was mentioned before, other risk that could lead to losing control of the consumers data is
the bankruptcy of the Cloud Provider. Cloud Providers are companies and companies may
go bankrupt. The obvious question that arises is what happens to the consumers data and
applications if such scenario becomes a reality? Migrating data (e.g. images or documents)
probably would be less demanding than migrating applications developed with specic PaaS
API. As a result, the lack of interoperability at the PaaS may turn the migration process
extremely difcult, requiring the refractor of the majority of the source code.
12 Medical Imaging
4. Available solutions and new opportunities
As we mentioned at the Challenges section, the architecture of Cloud Computing is not
suitable by itself to sustain PHI. There were identied several issues that turn the outsourcing
of PHI on the Cloud inappropriate or even illegal. However, when the challenges are
identied, opportunities also may emerge to mitigate them. At this following section we
will present some design approaches, built over the Cloud, in order to enable XDS-I IT
infrastructure to be outsourced to the Cloud taking into consideration the previous identied
challenges.
4.1 Encryption and decryption client-side
The public Cloud design permutes the previous trust patterns of older computing paradigms
such as Client-Server. For instance, at the Client-Server paradigm the Server is considered the
most trustworthy entity of the architecture. This trust assumption fades away when the data
to be dealt on the Cloud is PHI. At this scenario of storing PHI on the Cloud the trustworthy
entities are the clients and it must be assumed that the Clouds data center is untrustworthy.
Therefore, besides the transmission channels between nodes, the information must be hidden
from the central node where the information is stored or relayed, more precisely the Clouds
data center. To do so, the encryption must be performed on the client machine and stored
ciphered at the Cloud data center. Consequently, a reverse procedure must be performed
while retrieving the information from the Cloud: download the ciphered data and decrypt
it at the client machine. The described workow may seem trivial but is the core essence
of several systems that use the Cloud to connect trustworthy nodes. With this strategy two
approaches may emerge: Cloud as a Relay or Cloud as a Repository.
4.1.1 Cloud as a eelay
In the Cloud as a Relay approach the Cloud is used to create a virtual bridges between the
intranets of different healthcare institutions enabling the exchange, search and store of medical
images within a wide domain (Ribeiro, Costa &Oliveira, 2010; Ribeiro, Silva, Costa &Oliveira,
2010). They add at each trust node (i.e. intranet of the healthcare institutions) one device that
acts as a proxy for the income data and as a gateway for the outcome data. The proxy/gateway
device installed within the healthcare institution speaks DICOM (i.e. the device is DICOM
compliant and implement several DICOM services) and with the outside it communicates
through Internet standards such as IMAP and HTTPS (gure 5). The mentioned solutions are
interoperable within the institution due to the fact that they implement the DICOM services.
However, at the cross-enterprise scale the communication is ad-hoc. Nevertheless, they
were managed to enable the exchange of medical images through a Cloud bridge without
jeopardizing the condentiality and privacy of the clinical data. Furthermore, because the
Cloud is just used as a relay and the PHI is stored temporally it mitigates identied issues,
namely consumers lock-in, bankruptcy of the Cloud provider, and the geo-distributed nature
of the Cloud is irrelevant. However, the major drawbacks of such solutions are: (i) - the
Cloud is just used as a relay and the IT infrastructure remains at within the institution.
Therefore, the Clouds benet, of outsourcing at lower prices, is not ideal. Furthermore, a
VPN connection would have the same effect (although in several countries opening a VPN
connection in a public healthcare institution is a bureaucratically and procedure and such
solutions may diminish the initial inertia of inter-site cooperation). (ii) - The performance of
the relay strategy decreases while comparing with storing the data at the Cloud in a repository
100 Medical Imaging
approach, since the data must be uploaded to the cloud and downloaded from the cloud for
all inter-site interactions. (iii) - Such solutions may not be considered pure cross-enterprise
because, for instance, they do not deal with cross identication of patients so that task is
managed manually.
Gateway
Image Archive
DICOM
Gateway
DICOM
Workstation
Relay
H
T
T
P
o
r
IM
A
P H
T
T
P
o
r
IM
A
P
Institution A Institution B
Fig. 5. Example of using Cloud as a Relay inter-connecting two different healthcare
institutions.
4.1.2 Cloud as a repository
In the Cloud as a Repository approach the Cloud is used as repository of medical images (i.e.
Image Document Source) (Silva et al., 2011). The architecture is similar to the Cloud as a Relay:
there is a device (e.g. gateway) that encrypts/decrypts the clinical data client-side. However,
the ciphered data is stored in a persistent manner at the cloud (gure 6). Furthermore,
the trivial approach does not enable rich search queries (e.g. by patient demographics)
only queries by unique identier of the document. To overcome this problem Silva et al.
(2011) decoupled the searchable text based data of the images (metadata) from the image
data pixel itself. The pixel component data is encrypted and stored at the Cloud while
the searchable metadata is stored is a server that is owned (therefore trustworthy) by the
healthcare institution. This approach diminishes the IT infrastructure more than the Cloud
as a Relay approach, since the metadata represents a small portion of the data required to be
stored. Furthermore, the privacy and condentiality of the clinical data is guaranteed and the
geo-distribution nature of the Cloud is diminished. However, this approach may be affected
from Cloud bankruptcy or data lock-in due to the fact that the actual pixel data is at the
Cloud-side. These risks are minimized since the gateway of Silva et al. (2011) enables Cloud
provider redundancy, by following a similar approach of DeltaCloud. This way, healthcare
institutions are able to store the ciphered pixel data at more than one cloud provider in
a transparent manner for the institutions DICOM devices, increasing the reliability of the
system. The Silva et al. (2011) PACS solution is ideal for a multi-site healthcare institution
enabling the institutions distributed nodes to share the same PACS archive. However, the
trust community created does not follow the communication standards recommended by
the XDS-I content prole and vendor lock-in is still possible at the cross-enterprise scale.
Nevertheless, XDS-I does not predicts the outsourcing of the IT infrastructure and Silva
et al. (2011) was capable of developing a cross-site solution enabling it without jeopardizing
the privacy and condentiality of the clinical data, and with an acceptable performance
degradation while comparing with local PACS archive.
14 Medical Imaging
Gateway Modality
DICOM
Gateway
DICOM
Workstation
Repository
H
T
T
P
H
T
T
P
Institution A Institution B
Fig. 6. Example of using Cloud as a Repository of clinical information.
4.2 XDS-I infrastructure on the cloud
The current state of Cloud Computing does not guarantee privacy or condentiality of the
stored data. Even if the Service Terms and the Privacy Policy of the Cloud provider claim
privacy and condentiality of the trusted data, it is extremely difcult (if not impossible) to
prove that there was not any leak of information or if there was some data mining processes
to extract patterns of the clinical data. Therefore, the systems design running above the
Cloud must take this issue in consideration. As we mentioned previously, XDS-I assumes
that the nodes where its actors are instantiated are trustworthy. Therefore, the clinical data
may be stored in a readable manner - assumption possible to have in private Clouds owned
by trustworthy entities (e.g. hospital, National Healthcare System) but inappropriate to have
in public Clouds. In order to combine these two concepts (public Cloud and XDS-I) one must
ensure privacy and condentiality of the PHI without removing the interoperability of the
XDS-I content prole. Analyzing the XDS-I actors (gure 2) we have three candidates to be
migrated to the Cloud: Document Registry, Document Repository and Imaging Document
Source. The only central and unique component of the Afnity Domain is the Document
Registry, which searches the registry to locate documents that meet the criteria specied in
the query request by the Document Consumer IHE (2006b). The queries supported by the
Document Registry are well dened by the transaction Registry Stored Query (IHI-18). This
transaction reduces directly or indirectly the scope of the search to the global patient ID of
the respective afnity domain. For instance, to nd documents metadata the patient ID
must always be supplied and within that patient scope the search query may lter according
other attributes (e.g. documents type, authors ID). Therefore, queries only based on patient
demographics are not supported. Patient demographics are an optional eld only intended
as an audit/conrmation mechanism for the Document Consumers (IHE, 2010). Ribeiro et al.
(2011) was able to launch XDS Document Registries on the public Cloud in a secure manner
and keeping the privacy and condentiality of the metadata. Any elds of the metadata on the
XDS Registry possible to extract meaning (e.g. patient demographics, institutions name) are
protected. The level of protection is dened by the administrator of the Afnity Domain that
may vary fromstoring the meaningful elds encrypted to not allowing those elds to be stored
at all. This is possible due to the fact that the readable metadata with meaning is optional.
Nevertheless, from the three above mentioned candidate actors the XDS Document Registry
is the one that offers less consequences if disclosure of the meaningful elds occurs. In the
other hand, the other two candidate actors, Document Repository and Imaging Document
Source, hold valuable clinical information and must be protected accordantly from disclosure.
102 Medical Imaging
The approaches to ensure privacy and condentiality, and at the same time interoperability
while migrating these two actors to the Cloud may pass by:
1. Encrypted Storing and on-the-y Decryption: The PHI is cloud-side encrypted with
one symmetric key generated by the XDS-I actor (unknown to any other entities). The
encrypted PHI is stored on the cloud. Whenever transaction triggers the access to the
encrypted data, the data is decrypted on-the-y and the transaction is answered with
readable data. This approach guarantees interoperability and adds some level of data
privacy/condentiality protection. However, it is possible that when the data is being
decrypted on-the-y the Cloud provider sniff it leaking the PHI.
2. Middleware encryption/decryption: One middleware device is added between the
communication of the Document Consumer and the Document Repository or Image
Document Source placed on the cloud. The middleware device is owned by the healthcare
institution and is responsible for translating (encrypting/decrypting) the ow of data
between the two end actors. The middleware device implements the XDS-I transactions
and therefore it is interoperable. Furthermore, privacy and condentiality of PHI is
ensured due to the fact that PHI is stored ciphered and encrypted in a trustworthy node.
However, this approach may bring performance degradation compared with the previous
approach since the message ow required is always bigger. Finally, the middleware
machine owned by the healthcare institution would be a bottleneck and scalability and
availability issues could rise.
3. XDS for private imaging: The XDS integration prole was designed to be document
format independent, i.e. it supports any document type. When the Document Consumer
retrieves data from the Document Repository like, for instance, a report in the Portable
Document Format (PDF) the document requires a PDF reader in order to be accessed by the
healthcare professional. The same occurs at XDS-I when a DICOM object is retrieved from
the Imaging Document Source. XDS-I is a content prole based on the XDS integration
prole for dealing with medical imaging. Following the same thread of thought and
taking in account that XDS is format independent it is possible to design a new content
prole XDS for private images (XDS-). At XDS- the three candidate actors could be
migrated to the public cloud, storing the documents and the medical images encrypted,
and the encryption/decryption would be performed client-side. This approach ensures
the PHI privacy and condentiality and interoperable at the architecture level since the
transactions and the actors of the XDS-I would be the same. However, the drawback of
this approach would be the lack of interoperability at the document level since it is not yet
predicted by IHE.
Furthermore, the three above approaches may follow a Searchable Symmetrical Encryption
(SSE) to discover wanted documents among the encrypted resources stored on the cloud.
Single private-key encryption inhibits the search document among encrypted blobs of data.
SSE enables the search of keywords over the encrypted data without the need to decrypt the
data or disclosure the keyword (Curtmola et al., 2006). Therefore, SSE ensures privacy and
condentiality of the data and at the same time the ability of retrieve selectively documents
from the Cloud is maintained. Finally, comparing the risk associated with each approach the
1
st
approach is the one that offers less privacy and condentiality protection. In the other
hand, the 2
nd
and 3
rd
approaches offer higher levels of protection from disclosure which, in
our opinion, are adequate for storing PHI on the Cloud.
16 Medical Imaging
5. Conclusion
It is expected that the production of medical imaging will continue to increase in the following
decades. For instance, the PET-CT modality requires space for storing the PET images,
the CT images and the outcome fusion images and, the same situation happens with the
new modality PET-MRI. Furthermore, there is a new research trend of content-based image
retrieval, where it is possible to discover and retrieve images based on the pixel data of the
image. This content-based retrieval is enabled by models describe the image and these models
also require store space. As result, the storing requirements of the medical imaging elds
are demanding and will be even more demanding in the future. Therefore, new storage
solutions with exible business models are needed more than ever. The Cloud computing
paradigm offers an elastic framework to allocate or release computational resources on-the-y
and enabling a more efcient usage and, as a consequence, reducing costs. Current PACS
architectures, hospital oriented, with their own short-term and long-term archives with
no or little interaction with other institutions or PACS are difcult to extrapolate to the
cross-institutional environment. XDS-I allied with XCA integration prole set the roadmap
to enable the cross-enterprise medical imaging sharing. The conjugation of both integration
proles offers to the healthcare institutions exible levels of inter-institutional coupling:
through XDS-I the institutions create a common trust community (XAD) aggregating their
federations into one single federation of medical imaging inter-site sharing or through XCA
allowing access to the documents of other XADs without the need of federation fusion.
Ribeiro, Costa & Oliveira (2010); Ribeiro, Silva, Costa & Oliveira (2010); Silva et al. (2011)
proved that storing and/or distribute medical images and related exams using public Cloud
providers is possible. Although, these solutions are interoperable within institution (since are
DICOM compliant) at the cross-enterprise level they do not follow the transactions dened by
the IHE. Nevertheless, based on their experience and on the hypothesis analysis performed
at the new opportunities section we conclude that the public Cloud Computing utility has
potential to host several actors of the XDS-I content prole safeguarding the privacy and
condentiality of the PHI.
6. Acknowledgment
The research leading to these results has received funding from Fundao para a Cincia e
Tecnologia (FCT) under grant agreement PTDC/EIA-EIA/104428/2008.
7. References
Binnig, C., Kossmann, D., Kraska, T. & Loesing, S. (2009). How is the weather tomorrow?:
towards a benchmark for the cloud, DBTest 09: Proceedings of the Second International
Workshop on Testing Database Systems, ACM, pp. 16.
Costa, C., Silva, A. & Oliveira, J. (2007). Current Perspectives on PACS and a Cardiology Case
Study, Computational Intelligence (SCI) 65: 79108.
Curtmola, R., Garay, J., Kamara, S. & Ostrovsky, R. (2006). Searchable symmetric encryption:
improved denitions and efcient constructions, Proceedings of the 13th ACM
conference on Computer and communications security, CCS 06, ACM, pp. 7988.
De Backer, A., Mortele, K. & De Keulenaer, B. (2004). Picture archiving and communication
systempart 2 cost-benet considerations for picture archiving and communication
system., JBR-BTR: organe de la Socit royale belge de radiologie (SRBR)= orgaan van de
Koninklijke Belgische Vereniging voor Radiologie (KBVR) 87(6): 296.
104 Medical Imaging
Faruqui, S. A. (2005). Utility computing: certication model, costing model, and related architecture
development, California State University.
URL: http://books.google.pt/books?id=wMdENwAACAAJ
Foster, I., Zhao, Y., Raicu, I. & Lu, S. (2008). Cloud computing and grid computing 360-degree
compared, IEEE Grid Computing Environments .
Fu, L. & Chen, T. (2010). Building enterprise application based on cloud computing, E-Health
Networking, Digital Ecosystems and Technologies (EDT), 2010 International Conference on,
Vol. 2, pp. 534 537.
Hendee, W. R. & Ritenour, R. (2002). Medical Imaging Physics; 4th ed, Wiley, New York, NY.
Huang, H. (2010). PACS and Imaging Informatics: Basic Principles and Applications, John Wiley
& Sons.
IHE (2006a). It infrastructure (iti) technical framework integration proles, revision 6.0 - nal
text, Technical report, Integrating the Healthcare Enterprise.
IHE (2006b). Iti technical framework supplement: Cross-enterprise document sharing for
imaging, Technical report, Integrating the Healthcare Enterprise.
IHE (2008). It radiology technical framework integration proles, revision 9.0 - nal text,
Technical report, Integrating the Healthcare Enterprise.
IHE (2010). It infrastructure technical framework: Cross-transaction specications and content
specications, Technical report, Integrating the Healthcare Enterprise.
Jacq, N. (2007). From genes to personalized healthcare: grid solutions for the life sciences : proceedings
of HealthGrid 2007, Studies in health technology and informatics, IOS Press.
Langer, S. (2009). Issues Surrounding PACS Archiving to External, Third-Party DICOM
Archives, Journal of Digital Imaging 22(1): 4852.
Lovis, C., Spahni, S., Cassoni, N. & Geissbuhler, A. (2007). Comprehensive management
of the access to the electronic patient record: Towards trans-institutional networks,
International Journal of Medical Informatics 76(5-6): 466470.
Michael Armbrust, e. a. (2009). Above the clouds: A berkeley view of cloud computing,
Technical Report UCB/EECS-2009-28, EECS Department, University of California,
Berkeley.
Mora, L., Nevid, J. & Chaplin, W. (2008). Psychologist treatment recommendations for
internet-based therapeutic interventions, Comput. Hum. Behav. 24(6): 30523062.
Mustra, M., Delac, K. & Grgic, M. (2008). Overview of the DICOM standard, ELMAR, 2008.
50th International Symposium, Vol. 1.
Pattynama, P. M. T. (2010). Legal aspects of cross-border teleradiology, Eur J Radiol 73(1): 2630.
Pianykh, O. S. (2008). Digital Imaging and Communications in Medicine: A Practical Introduction
and Survival Guide, Springer Publishing Company, Incorporated.
Rajkumar Buyya, e. a. (2009). Cloud computing and emerging it platforms: Vision, hype, and
reality for delivering computing as the 5th utility, Future Generation Computer Systems
25(6): 599616.
Ribeiro, L. S., Blanquer, I., Costa, C. & Oliveira, J. L. (2011). On demand ihe xds document
registries on the cloud, international Journal of Computer Assisted Radiology and Surgery,
Springer, pp. 297298.
Ribeiro, L. S., Costa, C. &Oliveira, J. L. (2010). Aproxy of dicomservices, Advanced PACS-based
Imaging Informatics and Therapeutic Applications, SPIE Medical Imaging.
Ribeiro, L. S., Silva, L., Costa, C. & Oliveira, J. L. (2010). Email-p2p gateway to distributed
medical imaging repositories, 3th International Conference on Health Informatics,
pp. 310316.
18 Medical Imaging
Rimal, B. P., Choi, E. & Lumb, I. (2009). A taxonomy and survey of cloud computing systems,
Networked Computing and Advanced Information Management, International Conference
on 0: 4451.
Roentgen, W. (1898). Ueber eine neue art von strahlen, Annalen der Physik 300: 1217.
Ruotsalainen, P. (2009). Privacy and security in teleradiology, European Journal of Radiology .
Silva, L., Costa, C. & Oliveira, J. (2011). A pacs archive architecture supported on cloud
services, International Journal of Computer Assisted Radiology and Surgery pp. 110.
10.1007/s11548-011-0625-x.
URL: http://dx.doi.org/10.1007/s11548-011-0625-x
Strkson, S. A. & Aslaksen, A. (2009). Pacs: Beyond radiology, International Journal of Computer
Assisted Radiology and Surgery 4: 168170. 10.1007/s11548-009-0321-2.
106 Medical Imaging
6
Large-Scale User Facility Imaging and
Scattering Techniques to Facilitate
Basic Medical Research
Stephen D. Miller
1
, Jean-Christophe Bilheux
1
, Shaun S. Gleason
2
,
Trent L. Nichols
2
, Philip R. Bingham
2
and Mark L. Green
3

1
Neutron Scattering Science Division, Oak Ridge National Laboratory
2
Measurement Science & Systems Engineering Division
Oak Ridge National Laboratory
3
Tech-X Corporation
USA
1. Introduction
Conceptually, modern medical imaging can be traced back to the late 1960s and into the
early 1970s with the advent of computed tomography
1
. This pioneering work was done by
1979 Nobel Prize winners Godfrey Hounsfield and Allan McLeod Cormack which evolved
into the first prototype Computed Tomography (CT) scanner in 1971 and became
commercially available in 1972. Unique to the CT scanner was the ability to utilize X-ray
projections taken at regular angular increments from which reconstructed three-dimensional
(3D) images could be produced. It is interesting to note that the mathematics to realize
tomographic images were developed in 1917 by the Austrian mathematician Johann Radon
who produced the mathematical relationships to derive 3D images from projections
known today as the Radon Transform
2
. The confluence of newly advancing technologies,
particularly in the areas of detectors, X-ray tubes, and computers combined with the earlier
derived mathematical concepts ushered in a new era in diagnostic medicine via medical
imaging (Beckmann, 2006).
Occurring separately but at a similar time as the development of the CT scanner were efforts
at the national level within the United States to produce user facilities to support scientific
discovery based upon experimentation. Basic Energy Sciences
3
within the United States
Department of Energy
4
currently supports 9 major user facilities along with 5 nanoscale
science research centers dedicated to measurement sciences and experimental techniques
supporting a very broad range of scientific disciplines. Tracing back the active user facilities,

1
Computed Tomography and History: http://en.wikipedia.org/wiki/Computed_tomography
2
Radon Transform: http://en.wikipedia.org/wiki/Radon_Transform
3
US DOE Basic Energy Sciences User Facilities:
http://science.energy.gov/user-facilities/basic-energy-sciences/
4
US DOE: http://science.energy.gov/

Medical Imaging

108
the Stanford Synchrotron Radiation Lightsource
5
(SSRL) a SLAC National Accelerator
Laboratory was built in 1974 and it was realized that its intense x-ray beam could be used to
study protein molecular structure. The National Synchrotron Light Source
6
(NSLS) at
Brookhaven National Laboratory was commissioned in 1982 and currently has 60 x-ray
beamlines optimized for a number of different measurement techniques including imaging
and tomography. The next generation NSLS-II facility is now under construction. The
Advanced Light Source
7
(ALS) commissioned in 1993 has one of the worlds brightest sources
of coherent long wavelength x-rays suitable for probing biological samples in 3D. The
Advanced Photon Source
8
at Argonne National Laboratory also has a number of x-ray
beamlines dedicated to imaging and tomography suitable for biological and medical imaging
research. The Spallation Neutron Source
9
(SNS) at Oak Ridge National Laboratory (ORNL)
also has a number of beamlines suitable for studying the structure and dynamics of proteins
and other biological systems. A neutron imaging and tomography beamline is currently being
planned for SNS. Similarly, the High Flux Isotope Reactor
10
(HFIR) also at ORNL has
beamlines suitable for examining biological matter and has an operational imaging beamline.
In addition, the production of medical isotopes is another important HFIR function.
These user facilities have been intended to facilitate basic and applied research and were not
explicitly designed with the intention to scan patients the same way commercial medical
imaging scanners do. Oftentimes the instrument beam power is significantly more powerful
than that produced by medical scanners. Thus the ionizing radiation effects of these beams
must be considered when contemplating how these facilities can contribute to medical
research. Suitable research areas involving user facilities include the study of proteins,
human and animal tissue sample scanning, and in some cases, the study of non-human
vertebrate animals such as various rodent species. The process for scanning biological and
animal specimens must be approved by the facility biosafety review board.
However there is still a significant amount of bio and medical related research being
performed at these national laboratory user facilities and below is a sampling of some of the
published research which utilized imaging and scattering resources available via the above
mentioned national laboratories:
- At NSLS: The study of thrombosis in rats utilizing Micro- and Nano-CT (Stolz et al.,
2011), density quantification of Vasa Vasorum in tomographic coronary angiograms
(Moritz et al., 2010), Micro-CT imaging of the human lung acinus (Litzlbauer et al.,
2010), rat bone imaging (Rao et al., 2009), bone density using iliac crest biopsies
(Jorgensen et al., 2008).
- At APS: sparse data reconstruction of biomedical samples using micro-tomography
(Xiao et al., 2010), trabecular bone sample imaging (Xiao et al., 2008), high-energy x-ray
scattering of a cortical bone specimen (Stock et al., 2008), visualization of trace metals in
biological tissue (D de Jonge & Vogt, 2010).

5
SSRL: http://www-ssrl.slac.stanford.edu/
6
NSLS: http://www.nsls.bnl.gov/
7
ALS: http://www-als.lbl.gov/
8
APS: http://www.aps.anl.gov/
9
SNS: http://neutrons.ornl.gov/facilities/SNS/
10
HFIR: http://neutrons.ornl.gov/facilities/HFIR/
Scattering Techniques to Facilitate Basic Medical Research

109
- At ALS: irradiation effects on human cortical bone fracture behavior (Barth et al., 2010),
x-ray diffraction microscopy of whole biological cells (Nelson et al., 2010), and soft x-
ray tomography to image antifungal drug molecules in action (Uchida et al., 2009).
- At SSRL: cell-cell and cell-matrix adhesion molecules (Choi & Weis, 2011),
crystallographic analysis (Gupta & Kielkopf, 2011), X-ray absorption spectroscopy for
DNA repair (Giri et al., 2011).
- At SNS & HFIR: protein-protein interactions in DNA replication and repair (Hinerman,
2008), lysosome diffusion dynamics of hydration water (Zhang et al., 2009), and
enzymatic studies in aqueous ionic liquids (Baker & Heller, 2009).
Each of these facilities utilizes a similar process for how to get beam time at the instruments.
There are periodic calls for proposals to which competing proposal teams will respond with
descriptions of the experiments they seek to perform while also citing their experiment goals
and objectives. Once the proposal call is closed, the proposals are reviewed by a committee
comprised of technical experts from a variety of different research and academic
institutions. The first pass is a feasibility review to determine if the proposed experiment can
be done at the beamline for which time has been requested. This feasibility review will
consider things such as the sample environment required, the nature of the material should
it be hazardous in some way, and the experiment needs versus the resolution and
capabilities of the instrument. The proposals that pass the feasibility review are then peer
reviewed by an external scientific review committee which rates and rank the proposals.
Subsequently, a beam time allocation committee will make determinations on the amount of
time each experiment will then get according to the scientific review committee findings.
The outcome will be a list of accepted and alternate proposals and the facility User Office
will then initiate making contact with the corresponding experiment Principal Investigators
informing them that their proposals have been accepted. As part of the scheduling process,
the instrument team will interact with the approved experiment teams to determine
experiment dates. As part of scheduling, a local contact will be assigned to each experiment
team. This local contact is a valuable asset to the experiment team as this person will assist
as necessary to help with making the experiment a success.
Each of these user facilities is a part of the national laboratory in which it resides. As such,
there are a number of potential resources available to the research teams which may vary
according to the capabilities, personnel, and support facilities available within a particular
national laboratory. Across the national laboratory system, five Nanoscale Science Research
Centers have been built, and one function of these centers is to help facilitate scientific
research at the experimental user facilities. These centers include:
- Center for Nanophase Materials Sciences
11
(CNMS) at ORNL which provides clean
rooms, wet and dry labs, sample fabrication, and analysis. CNMS is co-located with
SNS to help provide unique capabilities for neutron scattering.
- Molecular Foundary at Lawerence
12
Berkeley National Laboratory which provides
laboratories for material science, chemistry, biology, and molecular biology utilizing
clean rooms, controlled environment rooms, DNA synthesizer and sequencer, NMR
spectrometer, mass spectrometers, along with scanning tunnelling, transmission
electron, fluorescence microscopes.

11
CNMS: http://www.cnms.ornl.gov/
12
Molecular Foundary: http://foundry.lbl.gov/

Medical Imaging

110
- Center for Integrated Nanotechnologies
13
(CINT) jointly administered by Los Alamos
National Laboratory
14
(LANL) and Sandia National Laboratory
15
(SNL). CINT focus
areas are nanophotonics and nanoelectronics, complex functional nonomaterials,
nonomechanics, and the nonoscale/bio/microscale interfaces.
- Center for Functional Nanomaterials
16
(CFN) is located at Brookhaven National
Laboratory
17
(BNL). CFN develops understandings of the chemical and physical nature
of nanomaterials in order to make functional materials such as sensors, activators, and
energy conversion devices.
- Center for Nanoscale Materials
18
(CNM) is located at Argonne National Laboratory
19

(ANL) and focuses on research pertaining to advanced magnetic materials, complex
oxides, nanophotonics, and bio-inorganic hybrid materials.
In addition to these centers, the national laboratories have unique capabilities to interpret the
data that is generated by these facilities. For example, the Imaging, Signals, and Machine
Learning (ISML) group
20
within ORNL has expertise in imaging, image/data understanding,
and machine learning methods to turn large experimental data sets into useful information for
the scientist. A few of the relevant active research capabilities include high-resolution neutron
image capture (new scintillators, coded apertures, etc.), tomographic reconstruction, and post
reconstruction image analysis and data understanding. Application areas include biomedical
imaging
21
, telemedicine for retinal diagnostics, CAD for mammography, small animal high-
resolution micro-tomographic medical imaging, image analysis for organ segmentation,
neuron image analysis and phenotype screening. Such capabilities to analyze and interpret
data are an important component of such user facilities to ensure that the ultimate goal of
extraction of information from the collected data can be realized.
The remainder of this chapter will now examine the imaging and scattering capabilities of
these instruments with special considerations for neutron imaging.
2. Imaging and scattering techniques
The measurement techniques at these user facilities vary more widely than those currently
employed routinely within the field of medical imaging. In addition to traditional planar
and tomographic imaging, additional scattering techniques such as spectroscopy, diffraction
and small angle scattering are employed. In these cases, the scattered neutrons or x-rays are
fit to physical models in order to determine structure or dynamics (Pynn, 1990). Structures
can be at the atomic, molecular, or macro-molecular size scale depending upon the
wavelength of the probing beam. An important distinction between x-ray and neutron
scattering is that x-rays interact with the electron cloud while neutrons scatter off of the
atomic nucleus. Thus in conceptual thinking, x-ray attenuation is approximately linear

13
CINT: http://cint.lanl.gov/
14
LANL: http://www.lanl.gov
15
SANDIA: http://www.sandia.gov
16
CFN: http://www.bnl.gov/cfn
17
BNL: http://www.bnl.gov/
18
CNM: http://nano.anl.gov/index.html
19
ANL: http://www.anl.gov
20
ISML: http://www.ornl.gov/sci/ees/mssed/isml/index.shtml
21
ISML Biomedical Imaging: http://www.ornl.gov/sci/ees/mssed/isml/research-bio.shtml

111
according to the material density. However this is not the case with neutrons as the
scattering cross section
22
is not linear with element density and neutrons scatter quite well
off of hydrogen atoms where, in contrast, these atoms are difficult to image with x-rays, but
aluminum and other metals are relatively transparent to neutrons. Thus these potentially
complementary attenuation properties are useful to exploit for what they can provide.
2.1 Imaging
Contrasting imaging instruments and commercial medical imaging scanners reveals a
drastically different scale required for hosting the devices. Typically the largest portion of
the user facility is required to produce the interrogating beam. These beams have high
power and can typically be shaped to produce a beam bandwidth of interest. Thus the
instruments must be located at the facilities and people must come to them unlike the
situation where many of the medical imaging modalities can be small enough to be mobile.
Another important factor to consider is that imaging biological specimens, laboratory
animals, and in rare cases, human tissue is the exceptions and not the rule for these facilities.
To start with, one must make a convincing case via their proposal that it is worthwhile to
scan the subject of interest. Then one must be prepared for additional scrutiny and review
by the appropriate facility biological and bio-hazard review committees. Upon acceptance,
one must then be prepared to follow proper specimen handling procedures. Note that
science areas such as protein crystallography and soft matter are much more routine than
described above. The instruments at the user facilities have powerful capabilities and one
must carefully prepare in order to utilize them.
Despite these obstacles, there are active communities of researchers which utilize imaging
instruments to help with their research. The context here for the definition of imaging refers
to the various X-ray and neutron imaging instruments. Conceptually these instruments
produce images in similar format as those produced by medical imaging scanners, however
these instruments typically incorporate more flexibility for their capabilities and are less
specialized and optimized as compared to the medical scanners. To illustrate imaging
capabilities, some examples showing the differences in image contrast between neutron and
x-ray imaging are shown in figures 1 through 3.

22
Neutron scattering cross sections: http://www.ncnr.nist.gov/resources/n-lengths/list.html

Medical Imaging

112

Fig. 1. Neutron tomography of a dried horsefly. Top right: outer surface, top left: upper
portion sliced away, bottom right: mid-level slice, bottom left: contour of wing muscles.
Note that the insect is approximately 1cm in length head to tail. Images courtesy of Dr.
Eberhard Lehmann using the ICON beamline, SINQ, Paul Scherrer Institut, Switzerland
23
.

23
ICON Instrument: http://neutra.web.psi.ch/facility/icon_index.html

113

Fig. 2. X-ray (top) and neutron (bottom) images of a miniature transistor radio. Images
courtesy of Dr. Eberhard Lehmann taken using the ICON beamline, SINQ, Paul Scherrer
Institut, Switzerland.

Fig. 3. X-ray (Top) and neutron (Bottom) image of a bullet. Note the difference in contrast
where x-rays image illustrate metal density while neutrons are capable of imaging through
metals to image materials containing hydrogen atoms. Images courtesy of Dr. Eberhard
Lehmann taken using the ICON beamline, SINQ, Paul Scherrer Institut, Switzerland.

Medical Imaging

114
2.2 Scattering
From a medical imaging perspective, the usual case is to observe the attenuation pattern
based upon the beam which transmits through the object being imaged. For tomography, a
set of views are taken at angular increments around the object.
However a portion of the beam can scatter off of the object and can be detected outside the
boundaries of the transmission image, and scattering techniques exist for both neutron and
x-rays. For neutron scattering, a strong scattering object will scatter ~10% or less of the total
transmission flux. The scattering angle can be vary across 4t Steradians (sr) depending upon
a number of factors including beam properties such as wavelength and material properties
such as structure. A conceptual illustration of transmission and small angle scattering is
shown in figure 4. In contrast with transmission imaging, scattering detection systems may

Fig. 4. Cartoon illustrating transmission and scatter detection locations for an object in an X-
ray or neutron beam. Typically only one technique will be employed at a time but shown
here together to illustrate differences between the detection techniques. The transmission
image results from rays which strike the detector having passed directly through the object.
The scatter pattern results from the rays being deflected at some angle here shown as small
angle scattering though the scattering angles can be arbitrary in solid angle depending upon
the sample and imaging beam characteristics.

115
utilize very small samples in comparison to the detector sizes. For example, a 1 cm
3
size
sample or smaller is typical while the detection system sizes may range to 10+ meters tall in
order to obtain the desired solid angle detection range. Other scattering systems fix
detectors about the sample covering as much of the 4t space as possible. The necessary
distance between the sample and the detectors must be considered when spacing the
detectors so as to take into consideration adequate resolution, and this distance may be
several meters for some instruments.
There are a number of elastic scattering techniques including small angle scattering shown
in figure 5, reflectometry, powder and single crystal diffraction. There also exist inelastic
scattering techniques such as direct and indirect geometry spectrometry. Typically the
scatter patterns are not as visually intuitive as the transmission images as one must deduce
the structure via model fitting in the case for scattering science. As previously mentioned, a
good primer for neutron scattering has been written by Pynn (Pynn, 1990). For examining
the statistical nature of structure, scattering techniques can be used to measure a wide range
of length scales ranging from atomic to molecular, to macromolecular and thus has its
advantages over other measurement techniques.

Fig. 5. Neutron scattering image of porous silica from a Small Angle Neutron Scattering
instrument. Left: The raw detector data image note that the center of scattering is offset to the
left center of the image in this case to extend visibility of the data tails at the edge of the
scattering pattern. Right: the corresponding reduced data illustrating intensity versus inverse
Angstroms. Models can be fit to the reduced data to determine structure. Images courtesy of
Dr. William T. Heller of the Spallation Neutron Source at Oak Ridge National Laboratory.
3. Neutron imaging beamline overview
The following section illustrates from acquisition to visualization working with neutron
imaging data. The detection systems are explored as well as the processes for normalizing
data and reconstructing images. A section discussing data management pertinent to the
national laboratories is also included which illustrates some significant differences in data
handling between the research and commercial medical imaging communities.

Medical Imaging

116
3.1 Data acquisition
Neutron radiography systems follow the same basic geometric design as x-ray system, so
those familiar with resolution and magnification in x-ray systems will quickly feel
comfortable working with neutrons. There are very few neutron imaging instruments, and
they are built to accommodate a variety of types of materials research, so these instruments
are designed to be flexible in their geometry in order to satisfy imaging requirements over a
broad spectrum of applications from industrial to medical to archeological applications. This
flexibility requires the user to have a basic understanding of neutron imaging instrument
components and geometries to plan successful experiments. While the x-ray and neutron
system geometries are similar, the commonly used designs are different due to differences
in the available sources and detectors. Since x-ray and neutron imaging are so similar and x-
ray is a well accepted and understood imaging modality, the following discussion of
neutron radiography system will draw parallels and point out differences between the
systems through the description of the sources, detectors, and geometric designs.
3.1.1 Neutron sources
Neutron sources for radiography are large user facilities that are either reactor sources
producing a continuous beam of neutrons or spallation sources producing neutron pulses
(Arai & Crawford, 2009). Since the neutrons from both types of source are initially higher in
energy than would be useful for most scattering and imaging experiments, the neutrons are
passed through a moderator that scatters the neutrons to lower their energy. As a result of
creating a large initial source to emit as many neutrons as possible and use of a moderator,
neutron sources are quite large with ports to instruments having sizes in the tens of
centimeters. In contrast, x-ray sources with similar flux are available having spot sizes down
to 5m in diameter. The large source size for neutrons is a key issue when designing the
imaging geometry as discussed later.
Pulsed sources for both x-rays and neutrons can be used like a flash bulb to stop motion of the
object being imaged; however, the reason for building pulsed neutron sources such as the SNS
is to provide energy resolution. The speed that a neutron travels is directly proportional to the
energy of the neutron; therefore, a neutron pulse spreads out over distance traveled. By
placing a detector with timing resolution a sufficient distance from the source, images can be
obtained over a range of neutron energies (wavelengths). In medical imaging, we see dual
energy x-ray systems used to enhance the contrast between materials in the body and also
achieve some level of elemental analysis. This is possible because a materials x-ray attenuation
coefficient is a function of the x-ray energy, and this function varies between materials. For
example, a dual energy X-ray image can determine if a kidney stone is composed of uric acid
or if it contains calcium. Another common use of dual energy x-ray is to calculate bone mineral
density for diagnosis of osteoporosis or osteopenia. Finally, dual (or more) energy x-rays are
commonly used to enhance images of lung parenchyma because the bones can be removed
from the final image. In this same sense, imaging over a range of neutron energies can provide
contrast for different materials and even identify materials based on the transmission
spectrum. One area of interest for imaging at multiple wavelengths is with respect to crystal
structures and may have applications in imaging of interfaces between tissue and prosthetic
joint devices. The polycrystalline materials of these metal joints has a unique transmission
spectrum with large swings in attenuation at particular energies known as Bragg edges.

117
Imaging on both sides of a Bragg edge will provide two very different contrast images to help
identify the interface between tissue, bone, and prosthetic device.
3.1.2 Neutron imaging devices
Two types of direct digital imaging devices are employed at neutron imaging instruments:
scintillator based and micro channel plate (MCP) based. MCP based detectors provide the
higher resolution than scintillators for neutron radiography, but they are much more
expensive and degrade with use.
Scintillator based imaging devices use a
6
Li based scintillator plate to convert neutrons to
visible light which is imaged by a CCD or CMOS based camera as shown in figure 6. In
these systems, a mirror is used to move the camera out of the direct neutron beam for
protection of the electronics and to remove components that may become activated by the
neutron beam and as a result produce noise in the images due to gamma ray emissions
striking the CCD or CMOS electronics. The right side of this figure is a photograph of a
scintillator based camera in use at HFIR.
The highest resolution of scintillator based neutron imagers is approximately 50m and is a
function of the scintillator, the lens, and the CCD or CMOS detector. Currently the
resolution is limited by the scintillator resolution.
6
Li Scintillator screens are available in
varying thicknesses. Thinner screens provide the highest resolution but interact with fewer
neutrons which leads to longer exposure times.

Fig. 6. Scintillator based neutron imaging device.

Medical Imaging

118
MCPs are thin slabs of material with an array of holes through the slab. Each hole or
microchannel is an electron multiplier, so once a single electron is produced the signal is
amplified by a 1000 times or more to enable detection of the single event. MCPs are glass
based and have been developed for neutron imaging through the incorporation of
10
B in the
glass. Interactions of the neutron with
10
B in the glass produces ions that can escape the glass
into neighboring microchannels. These ions free electrons in the microchannel which then
get multiplied and detected at the output. The latest microchannels are being produced with
diameters <10m leading to resolutions approaching 10m. MCPs doped with
10
B are
efficient for neutron detection due to the high cross section of
10
B. Efforts to improve these
devices further are attempts to further reduce the channel diameters for higher resolution
and doping with Gd to further improve efficiency (Tremsin, 2011).
3.1.3 System design
Due to the difference in available source spot sizes, radiography system geometries in
particular used for CT differ between x-ray and neutron systems in the position of the object
between the source and detector. Neutron systems place the object as close as possible to the
detector while x-ray CT systems place the object half way between source and detector. One
reason x-ray systems will place object in the center is to allow rotation of source and detector
around the object for CT. This isnt possible with the permanent neutron sources, so the
object must be rotated and doesnt need to be centered. Another advantage of placing the
object in the center is a magnification factor of two for the object projection on to the
detector. Magnification eases resolution requirements for the detector, but restricts size of
the source spot to maintain resolution.
Resolution of radiography systems (both x-ray and neutron) is determined by source size,
system geometry, and detector resolution. Detectors for both x-ray and neutron imaging
systems consist of a conversion material, optics, and a pixelated detector. Resolution of the
detector is a function of the point spread of the conversion material, the modulation transfer
function of the optics, and the pixel size of the pixelated detector. Resolution of an imaging
system up to the detector plane is a result of the source and system geometry as shown in
figure 7. In this figure, an edge is imaged by a source of finite size, D, at a distance, L, from
the source. From parts a and b of the figure, the blur of the edge which is called penumbra,
p, is obviously smaller with the reduced source size. The comparison between a and c shows
an inverse relationship between the source to detector distance and the penumbra. As a
result, instrument developers at neutron sources often specify the quality of their neutron
beams with L/D. This value has a direct relationship to resolution of the beam for imaging.
L/D values from 300-500 are typical at neutron imaging instruments. Most instruments will
offer setups at several different L/D values. The L/D values are selected by placing an
aperture in the beam line close to the source to restrict the size of the source. As the diameter
of the aperture gets smaller, L/D (resolution) increases linearly and exposure time increases
as a square of the diameter. Since one could easily produce a very high L/D value with a
small aperture an instrument specifications must also include a neutron flux value for fair
comparison between imaging instruments.
The example in figure 7 kept the object at the same distance from the imager. To understand
why the object is placed close to the imager for neutron imaging with a relatively large spot
size, consider the effect of pulling the object away from the imager. If an object is placed in

119
contact with the imager, penumbra is reduced to zero. As the object moves away, the object
is magnified onto the detector, but so is the penumbra or blur of the edges. For flexibility,
neutron imaging instruments have stages to move the object position setup to accommodate
various size objects and still keep them as close to the imager as possible.

Fig. 7. Resolution of system depends on source size and geometry. Decrease in source size
between a and b; Increase in source to detector distance between a and c.
Before a user submits a proposal to use a neutron imaging instrument, he/she should
understand the capabilities of the beam line to determine whether the instrument provides
the required field of view and resolution for their application. User facilities have
specifications for their beam lines and will provide best resolution and field of view
specifications, but it is difficult to include all potential imaging setups in descriptions of a
neutron imaging instruments due to the flexibility. User facilities have beam line scientists
assigned to each instrument. Consulting with the beam line scientist on a particular
application is a recommended method to determine whether the instrument is capable of
meeting the needs for the proposed research.
3.2 Data management
Data management at large-scale research facilities can be drastically different than
commercially available smaller scale systems. In general commercially available system are
static as well as propriety and perform in a highly regulated environment whereas research
facilities are quite dynamic and utilize open source packages on unregulated open data files.
Many research facilities do indeed have data policies and/or best practices that they follow
but for the most part they follow an open research policy. These large-scale research facility

Medical Imaging

120
data policies support an innovative and collaborative environment while still enabling a fair
and competitive atmosphere.
In recent years we have seen a dramatic increase in data file sizes that are generated from
research instruments. This size increase has afforded researchers the ability to produce high
fidelity images with extremely fine-grained resolutions. As a result, scientific discovery has
been enabled and it is now commonplace to process and transfer data files on the order of
hundreds of gigabytes. These data files are part of collections that in some cases approach a
terabytes of raw and reduced data. The ability to transfer these data sets over high-
bandwidth modern networks is now not trivial especially when remote users require these
data collections to be transferred to their home institutions. In most cases it is necessary to
use multi-threaded streaming data file transfer applications in order to efficiently migrate
these data collections. For instance, Orbiter Commander is currently used at the Spallation
Neutron Source for transferring data collections utilizing streaming compressed mutli-
threaded algorithms as seen in figure 8.

Fig. 8. The Commander file download manager application https://orbiter.txcorp.com
3.3 Data reduction and normalization
Data from scintillation based detectors results in projections stored in common image
storage formats. MCP imagers detect individual neutron interactions resulting in a list that
includes the time and position of each acquired interaction instead of an image. Images are
produced out of this list mode data through binning of the individual interactions based on
the imaging setup (i.e. binning on time for pulsed sources provides energy resolution). X-
ray CT systems encountered in medical imaging applications are well developed such that

121
projection data is captured along with normalization shots and then normalized and passed
to a reconstruction engine without requiring input from the user along the way. The
flexibility of neutron imaging instruments makes full automation of this process difficult,
since the user can select the number of projections, range of angles, number of normalization
images, and rotational stage position. Acquisition will provide the user with a projection
data set and normalization images (light and dark) based on the setup. Beam lines will have
normalization and reconstruction capabilities for the users to analyze their data.
After acquisition, the sample data files produced need to be normalized. The normalization
is a simple mathematical process that cleans up the data by removing imperfection of the
incident beam, signal detected that are not related to the sample measured and noise due to
the surrounding. To perform the normalization, two more data files called open beam and
dark field are required.
The open beam measures the shape and intensity of the incident beam. The open beam is
acquired using exactly the same set up and time duration as the sample data file but without
the sample in the beam. Due to the way they are acquired, background signals of the open
beam and sample data have to match. Because the intensity of the incident beam can vary over
time, it is important to bring the intensity of the data file background to the same level as the
open beam background. This is done by selecting several regions of interest, away from the
sample, in the data file, and calculating their average intensities. The same process is repeated
on the open beam. Average of the ratio of those intensities gives a corrector factor to apply to
the sample data. The dark field measured the electronic noise (thermal noise and dark current)
and background. It is acquired with incident beam turn off and shutter open.
Gamma rays detected also need to be removed from the various files. Gamma rays events
have the particularity of being very concentrated (1 or 2 pixels for each event) and very
intense. Because those events happen randomly, the files have to be treated individually.
Different mathematical processes can be used to do so. In development of normalization
routines at ORNL, we discovered that the Lee filtering works very well. After removing
gamma strike spikes from all images include data, open beam and dark field, the data are
normalized using the following mathematical formula:
Data DF
ND
OB DF

With ND: normalized data, Data: data sample, DF: Dark Field and OB: Open beam. Since
the DF and OB images are used to normalize every projection, any noise in these images
propagates into all normalized projections. To reduce the noise in the normalization
images, multiple DF and OB images are collected and averaged together after removal of
gamma spikes.
To facilitate this whole process, programs such as iMars, (iMAging Reduction Software), as
shown in figure 9, can be used. This program, using the IDL
24
programming language, has
been developed at the Spallation Neutron Source and works as follows. First, the user has to
select the input files (data file, open beam and dark field) and select the regions of interest

24
IDL: http://www.ittvis.com

Medical Imaging

122
using manual or graphical input. The program then automatically performs the
normalization and produced various output file formats (fits, tiff or jpeg).

Fig. 9. iMars (iMAging Reduction Software) user interface in use with the HFIR imaging
beamline.
3.4 Data reconstruction
Neutron radiography and computed tomography reconstruction can fully leverage
techniques of medical imaging with x-rays and has benefited from the development of
analysis and reconstruction algorithms developed by the medical imaging community. Two
discrete classes of reconstruction algorithms are widely employed to generate tomographic
images. Analytic algorithms calculate the image data directly using defined single-pass
mathematical methods. These algorithms have the advantage of relative simplicity and
speed, but tend to produce noisier reconstructed images. In contrast, iterative reconstruction
algorithms employ complex models of the acquisition system and repeatedly estimate
solutions and compare the estimate with a modeled ideal solution until pre-defined
convergence criteria are met. Iterative algorithms typically produce superior images but can
take much longer to execute. We will briefly introduce these two types in this chapter, and a
more detailed treatment can be found in (Gleason et al., 2010).
As an example of an analytic type, Filtered Back Projection (FBP) algorithms are well
developed for cone beam geometries seen in x-ray CT systems (Kak, 1988). Neutron
radiography systems are cone beam systems as well and therefore use the exact same
reconstruction algorithms with a change in system parameters for placement of the center of
rotation closer to the detector. Again, flexibility of the neutron systems will challenge the user
to set up the appropriate system parameters in any reconstruction code used so attention must
be paid to the setup during data acquisition to ensure a proper reconstruction.

123
FBP basically spreads the attenuation measured by a single pixel within a projection equally
across the 3D volume between the source and the pixel. With enough projections around the
object, the attenuation on a true object piles up and the attenuation that is spread to
incorrect positions in the volume becomes insignificant. This direct calculation method
enables high speed reconstruction of 3D volumes, but places requirements on number and
spacing of projections around the object. These requirements are easily met by x-ray systems
due to the high flux from x-ray sources. Neutron sources have lower flux at the camera, so
obtaining a data set with sufficient signal to noise ratio and number of projections may not
be feasible due to time required to collect the projections.
Iterative reconstruction algorithms have seen considerable development for improvement of x-
ray CT data, but do not see common use in medical imaging outside of research environments,
because iterative reconstruction is computationally intense requiring much more time than
FBP, and FBP reconstructions are suited for most medical applications. A CT projection data
set represents a large set of linear equations to be solved for attenuation in each voxel of the
volume. Using iterative methods for solving linear equations, reconstruction codes have been
developed by many research groups with variation in the iterative methods focused on
speeding up reconstruction through quicker convergence or parallel computation. Iterative
reconstruction algorithms allow reconstruction with fewer and non-uniformly distributed
projections and allow inclusion of more complex physics models. For these reasons, iterative
algorithms should be considered as a reconstruction option for neutron radiography where the
number and quality of projections are less than in medical x-ray CT. However, iterative
reconstruction comes at a considerable cost in computation.
3.5 Image analysis and data visualization
In broad use across the national laboratories is a wide range of open-source and
commercially licensed image analysis and data visualization applications. To illustrate,
several are listed below:
IDL
25
: The Interactive Data Language (IDL) is a commercial array-processing language,
which has been designed from its earliest inception for image processing applications
development. IIDL implements an efficient array-processing engine combined with a large
library of analysis routines, and is well suited for rapidly developing a wide variety of
complex mathematical programs. This combination of array processing, image visualization,
and data analysis functionality is why IDL is well suited for the tomography reconstruction.
While a full version of IDL requires a license from ITT Visual Information Systems (ITTVIS),
the IDL Virtual Machine is freely available and able to run compiled IDL programs. IDL is
also capable of exporting compiled procedures as self-contained executables, making it an
ideal platform for developing and sharing tomography algorithms.
MATLAB
26
: MATLAB is commercial a high-level array language with control flow
statements, functions, data structures, file and data input/output, and object-oriented
programming features with an integrated interactive development environment, and has
pre-built extensive facilities for displaying vectors and matrices as graphs, as well as

25
IDL: http://www.ittvis.com/
26
MATLAB: http://www.mathworks.com/

Medical Imaging

124
capabilities for annotating and printing these graphs. It includes high-level functions for
two-dimensional and three- dimensional data visualization and image processing. MATLAB
is also a natural platform for rapidly prototyping and developing reconstruction algorithms.
VisIt
27
: Developed by Lawrence Livermore National Laboratory, VisIt is a free, open
source, interactive parallel visualization and graphical analysis tool for viewing scientific
data on Unix and PC platforms. Users can quickly generate visualizations from their data,
animate them through time, manipulate them, and save the resulting images for
presentations. VisIt contains a rich set of visualization features so that you can view your
data in a variety of ways. It can be used to visualize scalar and vector fields defined on
two- and three-dimensional (2D and 3D) structured and unstructured meshes. VisIt was
designed to handle very large data set sizes in the terascale range and yet can also handle
small data sets in the kilobyte range.
VisIt employs a distributed and parallel architecture in order to handle extremely large data
sets interactively. VisIts rendering and data processing capabilities are split into viewer and
engine components that may be distributed across multiple machines:
VisIt-Viewer: Responsible for rendering and is typically run on a local desktop or
visualization server so that it can leverage the extremely powerful graphics cards that have
been developed in the last few years.
VisIt-Engine: Responsible for the bulk of the data processing and input/output (I/O) and is
typically run on a remote compute server where the data is located. This eliminates the need
move the data and makes high-end compute and I/O resources available to it. The engine
can be run serially on a single processor or in parallel on thousands of processors.
VisIt also supports C++, Python and Java interfaces. The C++ and Java interfaces make it
possible to provide alternate user interfaces for VisIt or allow existing C++ or Java
applications to add visualization support. VisIt can be controlled by its Graphical User
Interface (GUI), through the Python and Java programming languages, and can be
integrated into custom user interfaces as well.
VTK
28
/ITK
29
: The VTK is an open source library, licensed under the GPL (section 1.1). It is
entirely written in C++ and is available for many different platforms (Linux, Mac and
Windows), and has support for writing platform independent applications. The library has a
size of over 700 C++ classes and contains many algorithms for 2D and 3D image processing
and visualization. It implements a data processing pipeline and numerous filters for
reading, modifying and writing data. It also implements algorithms for volume and surface
rendering. Although the library is written in C++, it was designed to be easily extensible
and can be used within other programming languages. There are currently wrappers
available for TCL/TK, Java and Python. For the surface rendering, VTK typically uses
OpenGL to make use of the graphic cards hardware acceleration. In contrast to that, volume
rendering will in most cases be performed in software by the computers CPU.

27
VisIt: https://wci.llnl.gov/codes/visit/
28
VTK: http://www.vtk.org/
29
ITK: http://www.itk.org/

125
ITK is an extension of the VTK, adding techniques for analyzing medical images, for
example and registration/segmentation methods. Like the VTK, the ITK comes with an
extensive class documentation that has the same features as the VTK documentation. ITK
also includes several example applications that demonstrate the functionality of the library.
ImageJ
30
: ImageJ is a public domain Java image-processing program inspired by NIH Image
for the Macintosh. It runs, either as an online applet or as a downloadable application, on
any computer with a Java 1.4 or later virtual machine. Downloadable distributions are
available for Windows, Mac OS, Mac OS X and Linux.
It can display, edit, analyze, process, save and print 8-bit, 16-bit and 32-bit images and read
many image formats including TIFF, GIF, JPEG, BMP, DICOM, FITS and raw. It supports
stacks, a series of images that share a single window. It is multithreaded, so time-
consuming operations such as image file reading can be performed in parallel with other
operations.
ImageJ does calculate area and pixel value statistics of user-defined selections and measure
distances and angles. It can create density histograms and line profile plots and it supports
standard image processing functions such as contrast manipulation, sharpening, smoothing,
edge detection and median filtering. It does geometric transformations such as scaling,
rotation and flips. Image can be zoomed up to 32:1 and down to 1:32. All analysis and
processing functions are available at any magnification factor. The program supports any
number of windows (images) simultaneously, limited only by available memory. Spatial
calibration is available to provide real world dimensional measurements in units such as
millimeters. Density or gray scale calibration is also available.
ImageJ was designed with an open architecture that provides extensibility via Java plugins.
Custom acquisition, analysis and processing plugins can be developed using ImageJs built
in editor and Java compiler. User-written plugins make it possible to solve almost any image
processing or analysis problem.
GPULib
31
: Graphics hardware has been increasing in performance at a much faster rate
than CPUs. Additionally, since they are designed for graphics applications, GPUs already
include a great deal of parallelism and are optimized for vector- and matrix-based
calculations. Until recently this power was used mostly for video games and computer-
generated animations, and now the scientific community is taking advantage of GPU
performance in visualization applications routinely. GPULib is a commercial product
which provides a high-level, generalized interface that allows scientists and engineers to
take advantage of the speed and performance of GPU hardware in their own applications.
Using GPULib, vector calculations, matrix transforms, and array manipulations can be
off-loaded from the CPU to NVIDIA graphics hardware, leveraging the optimizations and
speed of the GPU to increase performance of their applications. Currently, language
bindings for IDL, Matlab, Python, Java, and C exist. Generally GPULib can be integrated
quickly and provide 30-100x speedup for reconstruction algorithms and 3D tomography
visualizations.

30
ImageJ: http://rsbweb.nih.gov/ij/
31
GPULib: http://www.txcorp.com/products/GPULib/

Medical Imaging

126
GPULib can be used within all of these steps for increasing the processing speed
substantially. To use GPULib from within an IDL script, you first need to call the gpuinit
procedure. This will load the needed libraries and give you access to GPU-related functions.
Once gpuinit is loaded, you will have access to over 135 GPULib functions.
4. Case study Neutron imaging
Following is a case study examining the utility and issues for neutron based medical
imaging. Currently there are no known neutron detection based commercial medical
imaging scanners, though medical applications for neutron imaging are explored.
4.1 Medical applications of neutron imaging
Interest in neutron radiography dates back to shortly after Chadwicks discovery of the
neutron (Chadwick 1932). In the 1950s and 1960s, neutron radiography came into use in
non-destructive testing(Berger 1965) but it was the later when it interest in imaging
biological samples emerged(Thewlis 1956; Berger and McGonnagle 1962; Barton 1964;
Atkins 1965). Seminal articles on the subject outlined to problems of neutron detection,
radiation dose, and multiple scattering events for in realistic thicknesses of tissue that
degrade the images for everything but high energy neutrons (> 1 MeV)(Brown and Parks
1969; Parks and Brown 1969; Budinger, Bichsel et al. 1971; Budinger, Howerton et al. 1971).
At that time, images were obtained by using a scintillator that exposed photographic film.
Since that time, detectors have improved and cover much more of the energy spectrum than
would be required to perform any imaging in humans. Despite advances, current detectors
are specific for defined neutron energies, are not linear, and do not cover the entire energy
spectrum (Vartsky, Mor et al. 2005; Marinelli, Milani et al. 2006; Marinelli, Milani et al. 2007;
Mayer, Forkel-Wirth et al. 2007; Dangendorf, Bar et al. 2009; Mor, Vartsky et al. 2009; Ruddy,
Flammang et al. 2009; Ruddy, Seidel et al. 2009; Vartsky, Mor et al. 2010).
The most significant problem for neutron imaging of tissues is multiple neutron scattering
by the protons in hydrogen. The best transmission images, x-ray or neutron, occur when
there is only a single scattering event in the sample before it reaches the detector. This goal
is never practically achievable. A larger the percentage of multiple scattering events will
result in greater the degradation of the image. Since tissue is composed of hydrocarbons,
neutron radiography is currently confined to thin specimens or the neutron energies must
be large. This is best understood by considering the scattering of neutrons from the protons
in hydrogen atoms in the organic molecules that compose tissues.
From elementary kinematics, the energy transfer from the incident neutron to the proton
(where the masses of the neutron and proton are taken to be equal) is given by:
( )
p n
E E
1
1 cos
2
| =
where E
p
is the recoil proton energy, E
n
is the incident neutron energy, and | is the center of
mass scattering angle. Averaging over all scattering angles gives a result that on the average;
the energy is divided evenly between the neutron and proton, i.e.

127
p n
E E
1
2
= .
So even very energetic neutrons are converted to thermal neutrons (energy ~0.025 eV)
within relatively few collisions. It can easily be shown that a 10 keV neutron requires ~19
collisions to reach a thermal energy while a 10 MeV neutron will only 28 collisions before
becoming completely thermalized. After a very few collisions, the neutron distribution
becomes isotropic and thus has lost all structural information. To understand how this
applies to tissues, consider Table 1 which shows an approximate composition of human
tissues (this will vary somewhat between organ systems but it is representative of the
body as an average). The cross sections are for thermal neutrons only and No represents
the total elastic cross section as measured in the laboratory. This shows that the neutrons
are scattered more by hydrogen than the next most significant element, carbon, by a factor
of 20. Deuterium, which is hydrogen with a neutron or H, has a much smaller cross
section for neutrons and therefore will not attenuate the beam as much. Thus, by using
deuterated or heavy water, the number of multiple scattering events will decrease which
will improve the image. The total cross section is important because it is related to the
mean free path by:
i i
i
N
1
o

| |
=
|
\ .

where the is refer to the different elements present. As noted above, most of the scattering is
from the proton in hydrogen so the mean free path in tissues can be approximated by:
( )
H H
N
1
o
~ .
Element
Tissue
Abundance
Atomic
Weight
Number
Density (cm
-3
)
o (b) No(cm
-1
)
H 0.11 1.008 6.51E+22 30.39 1.977084
C 0.51 12.010 2.53E+22 4.746 0.120158
N 0.02 14.010 8.51E+20 12.19 0.010372
O 0.36 16.000 1.34E+22 3.97 0.053320
Na 0.001 22.990 2.59E+19 3.92 0.000102
P 0.001 30.970 1.92E+19 4.37 0.000084
S 0.001 32.070 1.86E+19 1.52 0.000028
Cl 0.001 35.450 1.68E+19 65.32 0.001099
Table 1. A representative elemental composition of human tissues with number densities,
thermal neutron (E
thermal
= 0.025 eV) cross sections, and total elastic scattering cross sections.

Medical Imaging

128
Table 2 shows the mean free path for neutron energies from 1 eV to 10 MeV for hydrogen
scattering only. The table shows that to penetrate a human limb that the incident neutron
spectrum will need to be between 1 and 10 MeV. As might be expected, increasing the
neutron energy also increases the radiation dose (absorbed dose). The exact details of such
radiation doses due to a neutron require a significant effort to calculate. Such calculations
are typically performed by large Monte Carlo codes such as available from the Radiation
Safety Information Computational Center at Oak Ridge National Laboratory
32
. Prior
determinations of neutron doses would indicate that doses would be in an acceptable range
but further work needs to be performed before human trials could be contemplated (Brown
and Parks 1969; Budinger, Howerton et al. 1971; Kry, Howell et al. 2009).
E(keV) o(b) No(cm-1) (cm)
0.001 20.470 1.33256 0.750
0.01 20.181 1.31380 0.761
0.1 20.153 1.31196 0.762
1 20.048 1.30510 0.766
10 19.196 1.24867 0.800
100 12.743 0.82957 1.205
1,000 4.246 0.27643 3.617
10,000 0.935 0.06088 16.427
Table 2. Neutron cross sections, total neutron cross sections, and mean free path lengths for
hydrogen, as a function of energy.
One method to ameliorate the attenuation problem is to change the water composition from
normal water to deuterated water (also referred to as heavy water)(Slatkin, Stoner et al.
1983; Nakagawa, Hatanaka et al. 1994; Kushner, Baker et al. 1999; Medina, Li et al. 2005).
Studies have shown that approximately 20% of total body water (TBW) can be replaced with
deuterated water without deleterious effects(Kushner, Baker et al. 1999). However, it is time
consuming to replace 20% of the TBW, which would require several days of intravenous or
oral deuterated water intake. Since deuterated water is also expensive, this is not currently a
likely method to improve neutron transmission through tissue in vivo.
Even though there are problems with neutron attenuation and tissue dose for neutron
imaging, there are real possibilities for the future. Images have been made of small amounts
of tissue as can be seen in Figure 10. The images are of a rat lung in two views. The trachea,
main stem bronchi and bronchial tree are well defined by the contrast afforded by air
bronchograms. Several orders of bifurcation are clearly seen. Additionally, the cartilaginous
rings in the trachea and main bronchi are clearly visible as well showing a difference in

32
RSICC at ORNL: http://rsicc.ornl.gov

129
contrast between the two types of connective tissue. The cartilaginous bands are separated
and covered by a fibrous connective tissue. The molecular composition and thickness is seen
in the images. As detectors improve and imaging system geometries improve the resolution
of such images will undoubtedly improve. Resolutions of 50-10 m are readily available at
most neutron imaging centers today. Resolutions of 1 m appear to be routinely possible in
the near future with sub-micron imaging a real possibility in the future.

Fig. 10. Two neutron radiographs of the lungs of a rat from lateral (left) and anterolateral
(right) views are shown. The air bronchograms are clearly visible as is the cartilaginous
rings in the trachea and main bronchi. Courtesy: Dr. Burkhard Schillinger,
Neutronentomographie ANTARES, Forschungsreaktor FRM-II, Technische Universitt
Mnchen, Germany.
4.2 Possible uses
Osteoarthritis (OA) is the most common of all joint diseases affecting 68% of persons over 68
years of age as based upon radiological criteria(Goldman and Ausiello 2008). Clinically,
patients suffer from pain and functional limitations such as decreased range of motion and
instability. Plain x-ray radiographs demonstrate osteophytes and joint space narrowing.
Currently, there exist no treatment strategies that prevent or ameliorate the disease process
so therapies are aimed at analgesia and improving the function of the joint. Not
infrequently, the patient has a chronic course that eventually fails medical therapy which

Medical Imaging

130
makes the patient a candidate for joint replacement surgery. When the joint has significant
functional limitations and/or the pain becomes intractable, interfering with sleep or activity
the patient may elect to have a total joint replacement. The bone prosthesis interface is very
important to the success of the procedure. Even though not common, total knee replacement
infection rates are 0.5 to 1.0 whereas total hip replacements have infection rates of 0.5 to 1.0
(1995; Berbari, Hanssen et al. 1998; Sperling, Kozak et al. 2001; Widmer 2001). It is important
to make an accurate diagnosis because the treatment is the removal of the prosthesis
followed by many weeks therapy with antibiotics. New prostheses are not always possible
after infection. Therefore, it is very important to be certain that there an area of infection
next to the metal. X-ray techniques such as CT suffer from significant degradation due to
scattering from the metal prosthetic device. MRI scanning can still produce suitable images
except in the case of cardiac pacemakers and defibrillators. The prevalence of cardiac
pacemakers in all Medicare beneficiaries was increasing in time with 504.4 per 100,000
enrollees in 2000 as compared to 324.5 in 1990 and the prevalence increases significantly
with age (Brown, Croft et al. 2005). So the patients with pacemakers or defibrillators and
joint prostheses are increasing which leads to diagnostic dilemmas when the prosthesis is
believed to be infected. Using a technique called Bragg edge imaging, the bone prosthesis
interface should be amenable to relatively high resolution imaging which should
demonstrate the fluid found in an infection in those patients with pacemakers.
Another case which is becoming far too common is that of a stroke patient who needs an
MRI of the brain but cannot due to a pacemaker. Since both MRI and neutron radiography
map the hydrogen atom distribution to create an image, neutron radiography should be able
to provide images that would allow an accurate diagnosis to be made. Whether this is
actually achievable will require more research.
Like positron emission tomography, magnetic resonance imaging, and x-ray computed
tomography before it, neutron radiography will much research and engineering for it to
become a clinically useful tool. Paul Lauterbur and Raymond Damadian doubtless had people
shake their heads when told that their machines would image water in the body when that is
the major constituent of the human body. They eventually developed the MRI which has
revolutionized medical imaging of certain structures such as the brain. Originally, the MRI
could not image the heart, lungs, or great vessels due to motion and slow acquisition rates.
Those problems with time are being solved. So there is good reason to believe that neutron
imaging could have a similar story. Since the mean free path of a positron is ~1 mm, it has
been felt that PET cannot exceed a resolution of 1 mm. However, improved reconstruction
algorithms are improving the resolution to the sub-millimeter range. Thus, it is reasonable to
believe that the barriers for using neutron radiography for in vivo human imaging will be
overcome. When that occurs, physicians will have new images that will improve their
understanding of some disease processes and aid in the care of their patients.
5. Conclusion
The national laboratories provide a number of imaging and scattering instruments which
can be used to facilitate basic medical research. These resources are available competitively
via the scientific peer review process for proposals submitted through the user programs
operated by each facility. Imaging human and animal tissue occurs but is not routine in
most places, and strict procedures must be followed to do so. However research

131
communities are burgeoning in a number of biomedical areas, and protein crystallography
research is well established in the X-ray and neutron scattering communities. Novel here is
the forward looking work on neutron imaging with potential medical and biomedical
applications. Thus the national laboratories provide a research environment with
capabilities and a culture conducive to researching new methods and techniques suitable for
exploring new frontiers in medical and biomedical imaging.
6. Acknowledgement
The authors wish to thank the following for their contributions:
- The United Stated Department of Energy Basic Energy Sciences for the large scale user
facilities which it designs, constructs, operates, and maintains. In FY 2010, these facilities
cumulatively provided over 32,500 hours of service to over 12,700 unique users
33
.
- Research at Oak Ridge National Laboratory's Spallation Neutron Source was sponsored
by the Scientific User Facilities Division, Office of Basic Energy Sciences, U. S.
Department of Energy.
- This manuscript has been authored by UT-Battelle, LLC, under Contract No. DE-AC05-
00OR22725 with the U.S. Department of Energy. The United States Government retains
and the publisher, by accepting the article for publication, acknowledges that the
United States Government retains a non-exclusive, paid-up, irrevocable, world-wide
license to publish or reproduce the published form of this manuscript, or allow others
to do so, for United States Government purposes.
- DOE BES SBIR Grants DE-FG02-08ER85000
- DOE BES SBIR Grant DE-SC0004586
- Dr. Eberhard Lehmann for providing images from the ICON beamline, SINQ, Paul
Scherrer Institut, Switzerland.
- Dr. Burkhard Schillinger for providing images from the FRM II reactor of Technische
Universitat Munchen
- Dr. William T. Heller for providing SANS images from HFIR at Oak Ridge National
Laboratory.
7. References
Arai, M. & Crawford, K. (2009). Neutron Sources and Facilities, In: Neutron Imaging and
Applications, H. Z. Bilheux, R. L. McGreevy, & , I. S. Anderson, pp. 13-30. Springer,
ISBN 978-1-4419-4619-5, New York.
Atkins, H. L. (1965). "Biological evaluation of neutron radiography." Materials Evaluation
23: 453-458.
Baker, G. A.; Heller, W. T. (April 2009), Small-angle neutron scattering studies of model
protein denaturation in aqueous solutions of the ionic liquid 1-butyl-3-
methylimidazolium chloride, Chemical Engineering Journal, Vol. 147, No. 1, pp. 6-12,
Available from:
http://www.sciencedirect.com/science/article/pii/S1385894708007250

33
BES Operations: http://science.energy.gov/~/media/budget/pdf/sc-budget-request-to-
congress/fy-2012/Cong_Budget_2012_BES.pdf

Medical Imaging

132
Barth H. D., Launey, M. E. MacDowell, A. A. Ager III, J. W. Ritche, R. O. (2010), Irradiation
Effects on Human Cortical Bone Fracture Behavior, ALS Science Highlight #212.
Barton, J. P. (1964). "Some Possibilities of Neutron Radiography." Physics in Medicine and
Biology 278: 33-42.
Beckmann, E.C. (2006). CT scanning the early days, The British Journal of Radiology, Vol 79.
(January 2006) pp 5-8, DOI: 10.1259/bjr/29444122
Berbari, E. F., A. D. Hanssen, et al. (1998). "Risk factors for prosthetic joint infection: case-
control study." Clin Infect Dis 27(5): 1247-54.
Berger, H. (1965). Neutron Radiography. New York, Elsevier Press, Inc.
Berger, H. and W. J. McGonnagle (1962). Progress on neutron radiography, Argonne
National Laboratory. USAEC Report ANL-6279.
Brown, D. W., J. B. Croft, et al. (2005). "Epidemiology of pacemaker procedures among
Medicare enrollees in 1990, 1995, and 2000." Am J Cardiol 95(3): 409-11.
Brown, M. and P. B. Parks (1969). "Neutron radiography in biologic media. Techniques,
observations, and implications." The American journal of roentgenology, radium
therapy, and nuclear medicine 106(3): 472-85.
Budinger, T. F., H. Bichsel, et al. (1971). "Visual phenomena noted by human subjects on
exposure to neutrons of energies less than 25 million electron volts." Science
172(985): 868-70.
Budinger, T. F., R. J. Howerton, et al. (1971). "Neutron radiography and dosimetry in human
bengs: theoretical studies." Physics in Medicine and Biology 16(3): 439-50.
Chadwick, J. (1932). "Possible Existence of a Neutron." Nature 129(3252): 312.
Choi, H. J. & Weis, W. I. (2011), Crystal Structure of a Ridig Four-Spectrin-Repeat Fragment
of the Human Desmoplakin Plakin Domain, Mol. Biol., Vol 409, No. 800, DOI:
10.1016/j.jmb.2011.04.046.
Dangendorf, V., D. Bar, et al. (2009). "Multi-Frame Energy-Selective Imaging System for
Fast-Neutron Radiography." Ieee Transactions on Nuclear Science 56(3): 1135-1140.
D de Jonge, M.; Vogt, S. (2010). Hard X-ra fluorescence tomography an emerging tool for
structural visualization. Structural Biology, Vol. 20, No. 5, pp. 606-614. DOI :
10.1016/j.sbi.2010.09.002.
Giri, N. C.; Sun, H., Chen, M., Costa, M., and Maroney, M. J., (2011), XAS Structural
Investigation of Early Intermediates in the Mechanism of DNA Repair by Human
ABH2, Biochemistry, DOI: 10.1021/bi101668x
Gleason, S.; Paulis, M., & Osborne, D. (2010). Molecular Imaging: Principles and Practice:
Principles of Micro X-ray Computed Tomography, Weissleder, R., et al., eds.,
McGraw Hill, 2010, ISBN 9781607950059
Goldman, L. and D. Ausiello, Eds. (2008). Cecil Medicine. Philadelphia, Saunders, Elsevier.
Gupta, A. & Kielkopf, C. L. (2011). Purification, Crystallization and Preliminary X-ray
Crystallographic Analysis of a Central Domain of Human Splicing Factor 1, Acta
Crystallogr., F 67, No. 486. DOI: 10.1107/S1744309111004623.
Hinerman, J. M. (2008). The Study of Protein-Protein Interactions Involved in Lagging
Strand DNA Replication and Repair, Dissertation University of Toledo, Available
from:http://etd.ohiolink.edu/send-pdf.cgi/Hinerman%20Jennifer%20M.pdf?
toledo1216824884

133
Jorgensen, S.; Eaker, D. Vercnocke, & A. Ritmin, E. (2008). Reproducibility of Global and
Local Reconstruction of Three-Dimensional Micro-Computed Tomography of Iliac
Crest Biopsies, IEEE Transactions in Medical Imaging, Vol. 27, No. 4, pp. 569-575.
Kak, A.C. and Slaney, M. (1988), Principles of Computerized Tomographic Imaging, IEEE Press.
Kushner, D. J., A. Baker, et al. (1999). "Pharmacological uses and perspectives of heavy water
and deuterated compounds." Can J Physiol Pharmacol 77(2): 79-88.
Litzlbauer, H.; Korbel, K. Kline, T. Jorgensen, S. Eaker, D. Bohle, R. Riterman, E. &
Langheinrich, A. (2010). Synchrotron-Based Micro-CT Imaging of the Human Lung
Acinus, Anatomical Record, Vol. 293, No. 9. pp 1607-1614.
Marinelli, M., E. Milani, et al. (2006). "High performance (LiF)-Li-6-diamond thermal
neutron detectors." Applied Physics Letters 89(14).
Marinelli, M., E. Milani, et al. (2007). "Synthetic single crystal diamond as a fission reactor
neutron flux monitor." Applied Physics Letters 90(18).
Mayer, S., D. Forkel-Wirth, et al. (2007). "Response of neutron detectors to high-energy
mixed radiation fields." Radiation protection dosimetry 125(1-4): 289-92.
Medina, D. C., X. Li, et al. (2005). "Pharmaco-thermodynamics of deuterium-induced
oedema in living rat brain via 1H2O MRI: implications for boron neutron capture
therapy of malignant brain tumours." Phys Med Biol 50(9): 2127-39.
Mor, I., D. Vartsky, et al. (2009). "High spatial resolution fast-neutron imaging detectors for
Pulsed Fast-Neutron Transmission Spectroscopy." Journal of Instrumentation 4.
Moritz, R.; Eaker, D. Langheinric, A. Jorgensen, S. Bohle, R. & Ritman, (2010), E.
Quantification of Vasa Vasorum Density in Multi-Slice Computed Tomographic
Coronary Angiograms : Role of Computed Tomographic Image Voxel Size, Journal
of Computer Assisted Tomography, Vol. 34, No. 2, pp 273-278.
Nakagawa, Y., H. Hatanaka, et al. (1994). "Partial deuteration and blood-brain barrier (BBB)
permeability." Acta Neurochir Suppl (Wien) 60: 410-2.
Nelson J.; Huang, X. Steinbrener, J. Shapiro, D. Kirz, J. Marchesini, S. Neiman, A. M. Turner,
J.J. Jacobson, C. (2010), High Resolution x-ray diffraction microscopy of specially
labeled yeast cells, PNAS, Vol. 107, No. 16, DOI: 10.1073/pnas.0910874107.
Parks, P. B. and M. Brown (1969). "Antiscatter grids for low-energy neutron radiography."
Radiology 92(1): 178-9.
Pynn, R. (1990). Neutron scattering A PRIMER, Los Alamos Science, Vol. 19,
http://library.lanl.gov/cgi-bin/getfile?19-01.pdf
Rao, D.; Cesareo, R. Brunetti, A. Akatsuka, T. Yuasa, T. Takeda, T. Tromba, G. & Gigante, G.
(2009). Micro-CT Imaging of Rat Bone and Lumber Vretebra using Synchrotron
Radiation, Application of Accelerators in Research and Industry : Twentieth International
Conference, Vol. 109, pp 896-899.
Ruddy, F. H., R. W. Flammang, et al. (2009). "Low-background detection of fission neutrons
produced by pulsed neutron interrogation." Nuclear Instruments & Methods in
Physics Research Section a-Accelerators Spectrometers Detectors and Associated
Equipment 598(2): 518-525.
Ruddy, F. H., J. G. Seidel, et al. (2009). "Development of Radiation Detectors Based on Semi-
Insulating Silicon Carbide." 2008 Ieee Nuclear Science Symposium and Medical
Imaging Conference (2008 Nss/Mic), Vols 1-9: 5256-5262.
Slatkin, D. N., R. D. Stoner, et al. (1983). "Whole-body irradiation of deuterated mice by the
10B(n, alpha)7Li reaction." Proc Natl Acad Sci U S A 80(11): 3480-4.

Medical Imaging

134
Sperling, J. W., T. K. Kozak, et al. (2001). "Infection after shoulder arthroplasty." Clin Orthop
Relat Res(382): 206-16.
Stock, S.; De Carlo, F. & Almer, J. (2008). High energy X-Ray scattering tomography applied
to bone. Journal of Structural Biology, Vol. 161, No. 2, pp. 144-150. DOI:
10.1016/j.jsb.2007.10.001.
Stoltz E.; Yeniguen, M. Kreisel, M. Kampschulte, M. Doenges, S. Sedding, D. Ritman, E.
Gerriets, T. & Langhinrich, (2011). Angioarchitectural Changes in Subacute
Cerebral Venous Thrombosis, NeuroImage, Vol. 55, pp. 1881-1886.
Thewlis, J. (1956). "Neutron radiography." British Journal of Applied Physics 7: 345-350.
Tremsin, A.S., Phate, J.B., Vallerga, J.V., Siegmund, O.H.W., Feller, W.B., Lehmann, E.,
and Dawson, M.(2011), Improved efficiency of high resolution thermal and cold
neutron imaging, Nuclear Instruments and Methods in Physics Research Section
A: Accelerators, Spectrometers, Detectors, and Associated Equipment, Vol. 628,
No. 1, pp. 415-418.]
Uchida, M.; McDermott, G. Wetzler, M. Le Gros, M. A. Mylls, M. Knoechel, C. Baron, A. E.
& Larabell, C. A., (2009), Soft x-ray tomography of phenotypic switching and the
cellular response to antifungal peptoids in Candida albicans, Proc. Nat. Acad. Sci.,
Vol. 106, No. 46.
Vartsky, D., I. Mor, et al. (2010). "Novel detectors for fast-neutron resonance radiography."
Nuclear Instruments & Methods in Physics Research Section a-Accelerators
Spectrometers Detectors and Associated Equipment 623(1): 603-605.
Vartsky, D., I. Mor, et al. (2005). "Time-resolved fast neutron imaging: simulation of detector
performance." Nuclear Instruments & Methods in Physics Research Section a-
Accelerators Spectrometers Detectors and Associated Equipment 542(1-3): 206-212.
Widmer, A. F. (2001). "New developments in diagnosis and treatment of infection in
orthopedic implants." Clin Infect Dis 33 Suppl 2: S94-106.
Xiao, X.; Xia, D. Bian, J. Han, X. Sidky, E.Y. De Carlo, F. & Pan, X. (2010). Image
reconstruction from sparse data in synchrotron-based micro-tomography of
biomedical samples, 1st International Conference on Image Formation in X-Ray
Computed Tomography, UCAIR, pp 156-164.
Xiao, X.; De Carlo, F. & Stock, S. (2008). X-ray zoom-in tomography of calcified tissue,
Developments in X-Ray Tomography SPIE proceedings, Vol. 7078. DOI:
10.1117/12.796638.
Zhang, Y.; Lagi, M., Liu, D. Z., Mallamace, F., Frantini, E., Baglioni, P., Mamontov, E.,
Hagen, M., Chen, S. H. (2009), Observation of high-temperature dynamic crossover
in protein hydration water and its relation to reversible denaturation of lysozyme,
Journal of Chemical Physics, Vol. 130, No. 13, Available from:
http://jcp.aip.org/resource/1/jcpsa6/v130/i13/p135101_s1
7
Tools to Improve the Patients Processes
at Imaging Centers
Liliana Neriz
1
and Francisco J. Ramis
2

1
School of Economics and Business, Universidad de Chile, Santiago
2
Department of Industrial Engineering, University of Bio-Bio, Concepcion
Chile
1. Introduction
There is a worldwide concern for increased efficiency and cost effectiveness in healthcare
delivery which in the USA represents about 16% of the Gross Domestic Product. The
variability and complexity of the processes within healthcare systems demand the use of
more sophisticated management tools, such as the recommended by a joint study by the
National Academy of Engineering and the Institute of Medicine in the USA (NAE 2005),
to improve the healthcare delivery system. In this chapter it is seen the integration of
three management tools (Activity Based Costing, Lean Healthcare and Process
Simulation) to facilitate the development of Medical Imaging Center (MIC) studies in
order to improve the patients processes, which departs from traditional modeling
because it uses a pull paradigm for the patients. Activity Based Costing (ABC Costing) is
used to identify improvement opportunities from the managers point of view, Lean
Healthcare (Lean) to identify opportunities from the Customers point of view and
Process Simulation (Simulation) to test the potential impact of any proposed alternative
identified with ABC Costing or Lean, prior to its implementation. The center to be
modeled might be as simple as having one x-ray machine, or, more complex having all
kind of imaging machines (magnetic resonance, CT, x-ray, digestive radiology, ultrasound
scan, angiography, mammography and others).
Medical Imaging Centers are important units in every hospital or medical center as they are
a relevant link in generating a patients diagnostic. Because of this, it is necessary that these
centers be managed with high quality standards but also with efficiency. As an application,
an integrated example is provided where ABC costing, Lean and Simulation are applied to
an imaging center to improve the patients waiting times and process costs.
Medical Imaging Centers have made an important investment in PACS systems, which
provide data comprising from the reception of patients to the taking of the image. This kind
of system gives relevant and detailed information, such as: number and kind of patient,
frequency, type of exam, supplies and drugs utilized and personnel involved, all of these
entry data are significantly useful to the proposed tools, and, furthermore complement and
support the information for the management of MIC.

Medical Imaging

136
2. Patient flow
Traditionally, when studying a process, the first step is to build a flowchart of the process
that the patient will undergo. A typical patient's process is presented in Figure 1. The
process begins when a patient and his or her companions arrive to the reception desk at the
MIC. The arrivals could be spontaneous, scheduled or sent from an emergency room. The
receptionist asks the patient about the type of exams that he/she is requesting, proof of
insurance or payment method and all the paper work required. Once this information is
entered, the patient waits to be called by a technician/doctor to carry out a preliminary
evaluation, preparation prior to the exam in case the patient may need anesthesia, contrast
medium or other preparation, and the requested exams. The exams can be classified as:
magnetic resonance, CT, X-Ray, digestive radiology, ultrasound scan, angiography,
mammography and others. Once the exam is performed, the patient leaves the examination
area and meets her/his companion, afterwards, back office activities are performed to
digitalize/print the image or make a CD, finally, the results of the examination are
communicated to the patient or reported to the physician.

Fig. 1. Patient flow diagram at an Imaging Center

Tools to Improve the Patients Processes at Imaging Centers

137
Building and understanding MIC workflow is usually conducted in a brute-force fashion;
that is, by taking a stopwatch and physically timing each step of the clinical examination
processes. This can be observed as taking a snapshot from the doctors/technologists daily
activities focusing on a patient examination, including the room preparation prior to an
examination, the examination itself, the report generation and the discharge of the patient
from the MIC (Wideman and Gallet 2006).
The upper flow diagram is useful for representing the process and the activities involved,
the sequence of such activities, and also, as a communication device for the persons working
within the imaging center. However, it does not provide any guidelines to identify potential
improvements. Because of this, a three steps approach is proposed in this chapter to identify
opportunities for improvement: 1) ABC costing to identify important activities from a cost
point of view, 2) Lean to identify one of the patients concerns, the waiting times and
process times and 3) Simulation to test the impact of proposed solutions before
implementing them. These three tools are explained in the following section.
3. Description of management tools
There are several management tools that could be used to manage an imaging center. In
particular, this chapter illustrates the use and integration of three of them, which are focused
on the processes and generate synergies to support the decision making process more
objectively. This way, the manager can get different points of view, take more effective
decisions and promote the continuous improvement of the processes. The three tools to be
used are described as Activity Based Costing, Lean Healthcare and Process Simulation.
3.1 Activity based costing
Activity Based Costing (ABC) was developed by Kaplan and Cooper (1997). The objective of
ABC is to compute the cost of products or services (cost objects) in industries where indirect
costs are significant and have more than one product. Traditional costing methods assign
indirect costs directly to the products using an allocation criteria (units produced, patients,
square feet, etc.), which do not reflect the current use of resources. This is the case of
healthcare, where most of the costs are indirect, i.e., the resources utilized to produce the
products are used by more than one product.
ABC relates the use of the resources to the products, through the allocation of costs of the
resources to the processes, which are represented by activities. These activities consume
resources, and then the costs of activities are allocated to the products generated (goods or
services). For ABC, a process is a group of activities, i.e., a process is composed of at least
two activities. A definition of the concept of activity is a "task that consumes resources."
The ABC basic elements are: direct and indirect resources, activities, drivers and products.
Products are tangible goods or services. Drivers represent the way used to allocate resources
to activities and then from the activities to the products. The methodology can be
summarized as follows: (1) identify the processes in a company, (2) establish the activities
associated to those processes, (3) choose drivers for each indirect resource and allocate the
costs of the resources to the activities, and (4) calculate the costs of products by assigning
activity costs through the predefined cost drivers.

Medical Imaging

138
The ABC costing process requires identifying the activities and their characteristics as well
as recognizing the direct and indirect resources needed to operate the company. Once the
indirect resources have been identified and valued, their costs should be allocated to the
activities that use the resources by identifying the appropriate drivers (resource drivers).
These drivers relate the activity level to the resources used. Once the activities have been
valued, the next step is to allocate the costs of the activities to the products through other
drivers (activity drivers), which link the activity with the products. Finally, direct costs are
directly allocated to each product. See Figure 2.

Fig. 2. Activity based costing method.
Figure 2 provides valuable information to the manager, as opposed to traditional costing
methods that give the total cost and do not answer why or where the most expensive
activities are going to be found or how the process might be improved. Managing the
activities is called Activity Based Management (ABM), which is the basis for any initiative of
process improvement. As managers know their real costs, therefore they could find original
ways to improve their processes and reduce costs or add more value. ABM supports
continuous improvement, cost reduction program, re-engineering and other process and
quality initiatives (Kaplan and Cooper, 1997).
Imaging procedures are services with a high amount of indirect costs which require many
activities; several authors have provided some evidence such as Canby (1995) who, using ABC
principles and techniques, calculates costs associated with a x-ray process in a midsized
outpatient clinic. Laurila et al. (2000) show an informative and detailed picture of resource
utilization in a radiology department in order to support its pricing and management. Another
research from Laurila et al. (2001), studied the efficacy of continuous quality improvement
(CQI) compared to ordinary management from a radiology department. Cohen et al. (2000)
applied ABC to test the hypothesis that academic radiology can be separated into three distinct


139
businesses clinical activity, teaching and research- in order to determine the effect of the
current teaching paradigm on the clinical productivity. This analysis identifies opportunities to
improve quality of service, productivity and cost within each business.
There are several ABC studies in different imaging areas such as Enzmann et al. (2001) in
mammography services; Gray et al. (2003) in magnetic resonance imaging; Suthummanon et
al. (2005) researched how applying ABC to the nuclear medicine unit at a teaching hospital
for training interns; and Clevert et al. (2007) who analyzed methods to reduce cost in
interventional radiology departments by reorganizing supplies.
3.2 Lean applied to healthcare
Lean healthcare tools are focused on those activities that an institution must develop to
added value to their customers, represented by the patients of the clinical services. Lean is
based in the Toyota Production System (TPS), which, by definition is how to produce more
with less (Womack and Jones, 2005).
Lean has four principles: (i) Only perform activities that add value to the customer, (ii)
Eliminate waste. (iii) Promote flows, and (iv) Continuously improve processes. Adding
value to the customer means only perform those activities that the customer is willing to
pay for, the activities are correctly executed at first, and the activity must change the
product or service in some way. For example, waiting for an x-ray to be taken does not
add value to the customer.
Waste can be eliminated by identifying the seven sources of waste: 1) Production of defects
(wrong dose, wrong medication, rework, correction of defects), 2)overproduction(unnecessary
tests, reading mails, production to inventory), 3) Unnecessary transportation (inappropriate
layout, changing documents), 4) Waiting times(patients waiting for reviews, doctors/nurses
waiting for results of exams), 5) Excess inventories (expired drugs), 6) Motion (travel to the
laboratory, excessive file search) and 7) Extra-processes (data collection that are not used,
defective samples).
The promotion of flows is reached by minimizing the time between activities that add value.
For example, minimizing the waiting times of the patients between successive treatments,
which is triggered by a pull discipline, where the patient requests successive services as
he/she makes progress in the attention process.
The fourth principle states that by doing things a little better every day, this will lead to
continuously improve the processes. Ideally, people should arrive to their workplace in the
morning and ask themselves: what can I do better today?
This study uses a particular tool of lean called Value Stream Map (VSM), to illustrate the
relationship customer supplier, and also to show the waiting and process times for the
entire imaging procedure; as a result, overcrowding and unbalanced flows can be identified.
In order to build a value stream map, the following steps should be followed: 1) Identify the
boundaries of interest, 2) Identify the activities performed within the unit, 3) Use a pencil
and paper to collect field data, 4) Draw the flow based on demand, the customer
requirements, and his employer, 5) It is important to consider the flow of material,
information and production flow, 6) Check the value chain backwards in every step of the
flow, 7) Add boxes to the data on each process, 8) The individual process steps must go from

Medical Imaging

140
left to right in the flow of the value chain, sometimes many paths converge on several
points, 9) Add the number of operators 10) Add the cycle time and processing time, 11)
Calculate the total cycle time and total processing time, 12) Once traced the flow of the chain
and when data has been entered in each step, it is necessary to add the information flow and
13) Finally, add the timeline at the bottom of the page.
For a comprehensive view of lean see Womack and Jones (2005) pioneers in the introduction
of the concept of Lean. For the specific case of Healthcare, an excellent handbook is
provided by Graban (2008), with specific examples and guidelines for implementing lean
healthcare. A general overview about how lean has been applied in North American
hospitals is provided by Poole et al. (2010).
3.3 Process simulation
Simulation is a tool that allows studying complex systems under dynamic conditions (Law,
2007). Healthcare simulation is different from the well known manufacturing simulation,
where the entities go through a well defined and standardized process, requesting resources
at different locations, waiting in a row if the resources are not available and both the
processing and waiting discipline are according to a given priority. In healthcare, patients
(the entities) request the resources at a fixed location, or in other words, are the resources
(doctors, nurses, paramedics and others) that come to the patient and provide him with the
required services. Also, patients cannot wait forever to get service or be stored as
inventories in case they are not given service in a given day. Waits have dynamic priorities
and physicians give preference to the patients with the highest severity.
In order to build a simulation model with any modern simulation software, for example
with FlexsimHC that is the only object oriented simulation software for healthcare processes
(www.flexsim.com), the typical steps are: 1) Import layout of the imaging center, 2) Drag &
Drop objects into layout, 3) Adjust visual properties, names and areas of objects, 4) Create
the Tracks, 5) Make connections according to tracks/patient flow, 6) Configure the
experiment, scenarios and performance measures and 7) Run the model and view results.
Simulation has the flexibility to become a suitable tool for healthcare due to the complexity
of these systems, which are integrated by different actors and a large number of parts
interacting (Monsef, 1997). For a comprehensive healthcare simulation review see (Eldabi et
al., 2007), (Young et al., 2009) and (Neriz et al., 2011). Specific examples of simulation of
imaging centers can be found in (Ramis et al., 2008, 2009), where simulation focuses on
providing an alternative to reduce the waiting time of the patients, improve the use of
physical and human resources, evaluate different alternatives of plant layout, scheduling
patients or simply, to study and better understand the inner working of the system.
4. Using ABC costing and lean to identify opportunities for improvement
To illustrate the integration of tools, it is presented the case of an Imaging Center at a
research hospital, which is a high complexity center with different equipments such as
Magnetic Resonance, CAT's, x-ray, Digestive Radiology, Mammography, Angiography,
Echography among others. In this center, it was applied an activity-based costing model and
as an example in Table 1 it is reported the X- ray section, which shows the annual
breakdown of the costs of activities in this unit.


141
Activity
Total annual costs
(US$)
Activity
Total annual costs
(US$)
Carry envelopes 687.204
Give the report to the
patient
7,805.274
Record times 834.288 Give directions 8,275.712
Give indications to
patients
1,517.640
Deliver reports to
departments
8,648.474
Coordinate peoples
schedules
1,999.228
Call and receive the
patient
8,724.822
Give contrast to patient 1,531.640 Check exams (images) 9,303.936
Assign envelopes 1,999.228 Finish the process 10,217.488
Carry reports (into the
envelopes)
2,227.104
Proceed with the
process
11,315.038
Label envelopes 2,441.644 Do anamnesis 12,025.192
Arrangements 2,459.968 Digitalize exams 12,089.694
Bring reports 2,576.066 Release exams 12,089.694
Carry cartridge 3,511.320 Admit patient 15,208.386
Building Maintenance 3,732.934 Receive payments 23,444.442
Obey safety measures 4,789.020 Print pictures 23,654.790
Remove safety measures 4,789.020 Check inventory stock 27,519.104
Print report 5,749.998 Check report 36,406.510
Dismiss the patient 6,289.120 Do report 38,048.202
Back to the imaging
center
7,397.258
Manage academic
duty
84,917.472
Settle the patient 7,469.516
Equipment
Maintenance
14,6236.436
Check report 7,501.822 Manage the Center 341,832.954
Go to X-ray portable
examination
7,647.588 Take exams 795,188.264
Total Cost 1,721,125.714
Table 1. Activities in X-ray area
The information from table 1 is summarized in Figure 3. It can be observed that the costliest
activity is of taking the exam, a situation that is explained because most of the resources
employed in the X-ray section are fixed costs, and correspond to personnel, equipment and
facility costs, as can be seen in Figure 4.

Medical Imaging

142

Fig. 3. Activities costs in X-Ray area

Fig. 4. Costs of resources - Activity take exams.
Considering that the resources employed in the activity of taking the exam are fixed, this
says that if 1 x-ray or 1000 x-ray are made, the same fixed cost is incurred. Then, the
manager of the MIC realizes that to generate value he must make decisions about the
activity of taking the exam, for example, by increasing the production of X-rays the cost of
that activity is amortized with more x-rays. Now, the manager wants to increase the
production, but the analysis would consider this decision from a customer standpoint, that
is, people need to get an X-Ray without long waiting times. In order to do this analysis, a
value stream map (VSM) will be used.
From the standpoint of the VSM, there are three main activities: 1) the reception of patients
in the reception area. 2) Proof of insurance or direct payment from the patient and 3) Taking
the exam in the examination room. Figure 5 shows the VSM with the three main activities
mentioned, and it can be seen that the bottleneck occurs in the examination room because
the flow of patients, which in the previous stage was an average of 16.76 patients per hour,
drops to 5.95 patients per hour.


143

Fig. 5. Map of stream Value of the main processes in the X-ray area.
A more detailed analysis of what happens within the examination room, using the same
tool again, established that the flow is leveled in the first three activities, having a capacity
of 60 patients per hour. But the flow goes down in the last activity when the patient
returns to the dressing room to put on his/her clothes, now, the capacity is to attend only
17.94 patients per hour, which is shown in Figure 6. This drives to the conclusion that as
the dressing room is the bottleneck, by increasing the number of dressing rooms an
increase in productivity will be achieved.

Fig. 6. VSM in the examination room.

Medical Imaging

144
5. Using simulation to test identified solutions
From the previous section it has been seen that the actual bottleneck is the dressing room,
and then, by adding more rooms it should be improved the productivity of the x-ray section.
To test this hypothesis, a simulation model was constructed, following the steps described
to build a model and using the simulation package FlexsimHC.
The simplified model of the process is shown in figure 7. It is assumed that patients arrive
and go directly to the service if the dressing room and the technician are idle; otherwise they
go to the waiting room. Once the patient is ready, the technician will get him out of the
dressing room, locate him/her into the table and take the image. Then, after some
adjustments, the patient is taken back to the dressing room, where he/she gets dressed and
then leaves ending the process.
As a result of the simulation it can be observed, by having one dressing room the technician
is inactive about 65% of the time, instead by having two dressing rooms this figure goes
down to 35%, and with three dressing rooms it sharply decreased to a 5% idle time. In terms
of production, for a direct digital machine, with one dressing room a total of 50 patients will
be attended, with two a total of 100 and with three dressing rooms a total of 125 patients per
day. It is important to notice that increased productivity is achieved having the same
resources, except for changing the number of dressing rooms.

Fig. 7. Screen of simulation model.


145
6. Conclusions
This chapter shows the integration of three management tools, Activity Based Costing, Lean
Healthcare and Process Simulation, to improve the waiting times and equipment rate of
utilization at a clinical hospital. As a general conclusion we can say that:
1. Activity Based Costing is a valid tool for computing costs of products at healthcare
institutions. ABC provides a powerful tool for Activity Based Management (ABM),
because it shows the cost of the different activities involved in getting a given
service/product.
2. Lean healthcare provides a focus on the wastes from the point of view of the customers,
in this case, the waiting times.
3. Simulation is an appropriate tool for modeling in a healthcare environment, because of
the complexity of the processes and the flexibility of the tool when modelling different
situations. It was used to test solutions identified with ABC and Lean.
Specifically, we can conclude that, the bottleneck of the system is represented by the
dressing rooms. By adding more of them in the X-ray section, the productivity was
increased from 50 patients per shift to 125. Similarly, the average idle time of the technician
taking the images decreased from 65% to 5%.
7. References
Canby, J. (1995). Applying activity-based costing to healthcare settings. Healthcare
Financial Management, 49, 2, 50-56.
Clevert, D. A., Stickel, M. Jung, E. M., Reiser, M. and Rupp, N. (2007). Cost analysis in
interventional radiology. A tool to optimize management costs. European Journal
of radiology, 61, 144-149.
Cohen, M., Hawes, D., Hutchins, G., McPhee, W., LaMasters, M. and Fallon, R. (2000).
Activity-based cost analysis: A method of analyzing the financial and operating
performance of academic radiology departments. Radiology, 215, 3, 708-716.
Eldabi, T., Paul, R. and Young, T. (2007). Simulation modelling in healthcare: reviewing
legacies and investigating futures. Journal of the Operational Research Society, 58,
262-270.
Enzmann, D., Anglada, P. Haviley, C. and Venta, L. (2001). Providing profesional
mammography services: financial analysis. Radiology, 219, 2, 467-473.
Graban, M (2009), Lean Hospitals: Improving Quality, Patient Safety, and Employee
Satisfaction. Taylor and Frances.
Gray, D., Hollingworth, W., Blackmore, C. B., Alotis, M., Martin, B., Sullivan, S., Deyo, R.
and Jarvok, J. G. (2003). Conventional Radiography, rapid MR imaging and
conventional MR imaging for low back pain: Activity- based costs and
reimbursement. Radiology, 227, 669-680.
Kaplan, R. and Cooper, R. (1997).Cost & Effect: Using Integrated Cost Systems to Drive
Profitability and Performance. Harvard Business Press.
Laurila, J., Suramo, I., Brommels, M., Tolppanen, E-M., Koivukangas, P., Lanning, P.,
Standertskjld-Nordenstam, C-G.(2000). Activity-based costing in Radiology.
ActaRadiologica 41, 189-195.

Medical Imaging

146
Laurila, J., Standertskjld-Nordenstam, C-G., Suramo, I, Tolppanen, E-M., Tervonen, O.,
Korhola, O. and Brommels, M. (2001).The Efficacy of a continuous quality
improvement (CQI) method in a radiological department. Acta Radiologica 42,
96-100.
Law, A. (2007) Simulation Modeling and Analysis, Fourth Edition. New York. USA,
McGraw-Hill
Monsef, Y (1997), Modelling and simulation of complex systems, Society for computer
simulation international. In. Jin, X., Kagioglou M., Aouad G. 2006, Towards a
dynamic healthcare process: from requirement capture to simulation. Transactions
of the SDPS JUNE 2006, Vol. 10, No. 2, pp. 1-19
NAE/IOM (2005), Building a Better Delivery System, National Academy of Engineering
and Institute of Medicine, USA.
Neriz, L., Ramis, F. and Sepulveda, J. (2011). A New Approach for Healthcare Simulation",
Proceedings Society of Health systems, SHS 2011, Orlando, USA
Poole, K., Hinton, J. and Kraebber K. (2010). The gradual leaning of health systems.
Industrial Engineer. April, 50-55.
Ramis, F., Neriz, L. and Seplveda, J. (2008). A Simulator to Improve Waiting Times at an
Emergency Unit. Industrial Engineering Research Conference, Vancouver,
Canada.
Ramis, F., Concha, P. Neriz, L. and Seplveda, J. (2009). Improving Services of a Hospital
Imaging Center using ABC Costing, Lean and Simulation .Simulation Solutions,
IIE Annual Conference.
Suthummanon, S., Omachonu, V. and Akcin.M. (2005).Applying activity-based costing to
nuclear medicine unit. Health Services Magaement Research, 18, 141-150.
Wideman, C. and Gallet, J. (2006) Analog to digital workflow improvement: A quantitative
studt. Journal of Digital Imaging, 19 Suppl 1:29-34.
Womack, P. and D. T. Jones.(2005). Soluciones Lean. Gestin2000, Spain.
Young, T., Eatock, J., Jahangirian, M., Naseer, A. and Lilford, R. (2009).Three critical
challenges for modeling and simulation in healthcare. In Proceedings of the 2009
Winter Simulation Conference.
Flexsim user manual (2011), www.flexsim.com.
8
High to Microwave Frequencies
Imaging Techniques
George A. Kyriacou
1
, Ilias N. Aitidis
1
,
Dimitrios G. Drogoudis
1
and John N. Sahalos
2

1
Department of Electrical and Computer Engineering, Democritus University of Thrace
2
Department of Electrical & Computer Engineering, University of Nicosia
1
Greece
2
Cyprus
1. Introduction
Microwave and high frequency tomography constitutes challenging electromagnetic inverse
problems aiming at the reconstruction of its internal -,
r
- and/or
r
- distributions. The object
to be imaged is embedded in a lossy homogeneous background medium. This is surrounded
by a fictitious (or real) surface preferably of canonical shape (circular, rectangular, cylindrical
or spherical) over which a number of antennas (electrodes at lower frequencies) are evenly
distributed. The hardware implementing the modality should be able to successively activate
each one of them, while setting all the other antennas to a receive state operating as sensors.
Instead of requiring all antennas to operate in both transmit and receive modes, a subset of
antennas can be configured in transmit mode only (activated in turn) while a preferably larger
subset is configured in receive only mode (passive sensors), all of them performing
simultaneous measurements. In this manner the object is illuminated each time from a
different angle creating a scattered field to all possible directions, which is sampled by the
receiving antennas. The locations and number of transmit antennas should be designed to
cover the required spatial illuminations (projection angles), while the number of the evenly
distributed receiving antennas should fulfill the spatial sampling laws. An alternative
configuration mimicking the one already used in X-Ray Computer Tomography (CT) and
Magnetic Resonance Imaging (MRI) seems preferable and highly recommended for the
microwave imaging. Spesifically, for two-dimensional imaging a single active antenna along
with an array of sensing-passive antennas could be placed on a circular-colar possibly plastic
platform surrounding the object to be imaged. In turn by rotating this supporting structure the
object can be illuminated from all possible projection angles. For a three-dimensinal imaging
the hosting platform could be a plastic cylinder holding multiple antenna ring arrays, each
ring having one active and multiple passive antennas. Each antenna is activated successively
in time while the whole cylinder is roteated providing all possible illuminations and data
recording-sampling by all sensing antennas simultaneously. The information gathered by
this measurement procedure constitutes the dataset to be exploited by the imaging algorithm
to reconstruct the objects internal properties distributions. This constitutes a challenging
mathematical-computational and engineering problem since it is proved to be a highly non-
linear and ill-posed inverse problem.

Medical Imaging

148
Analytical methods can be employed only for simple canonical geometries but these are
valuable since they may serve as exact reference methods to validate numerical techniques.
Hence, for the practical arbitrary shaped and inhomogeneous bodies numerical techniques
are inevitable. It is on these approaches that the following chapter is elaborating.
For the reconstruction algorithm to be implemented a realistic as far as possible computer
model of the practical structure is necessary. For this purpose the geometry is discretized
including the appropriate antenna (or electrodes) modeling by following an engineering
compromise between accurately reflecting the structure and the required computational
resources. A usual approach is to consider a virtual body area of canonical shape large
enough to contain the unknown irregularly shaped actual object with its unknown ,
r
,
r

distributions. The properties of the background media outside this virtual surface are
assumed known as it is practically selected as desired. The also virtual surface carrying the
transmit/receive antennas is larger than the virtual body area but also embedded in the
known background medium. Theoretically, this background media extends to infinity,
however in order to implement a numerical technique a finite solution domain should be
established. Hence, another fictitious surface enclosing the whole structure constituting the
solution domain truncation surface is considered. In turn this truncation surface must not
disturb the electromagnetic field solution or to be transparent to the impinging waves.
The numerically discretized model includes everything within the truncation surface, but
the inverse problem unknowns are only the constitutive parameters within the virtual
body area. By the aid of the discretization this virtual body is comprised of a number of
pixels for two-dimensional (2-D) or voxels for three-dimensional (3-D) geometries. Each one
(i
th
) of them is assumed locally homogeneous with constant but unknown (
i
,
ri
,
ri
)
properties which are in principle different for each pixel/voxel, forming the so-called piece-
wise constant distribution. Now, the aim of the reconstruction is exactly the evaluation of
these three vectors []=[
i
], [
r
]=[
ri
] and/or [
r
]=[
ri
], by exploiting the dataset acquired
from field measurements carried out on the real object. Numerous different approaches are
established for the exploitation of this information. The most usual approach which is also
elaborated herein is to formulate a cost function based on the least square method, which
will be in turn minimized employing some type of optimization techniques. For this
purpose an initial (
i
o
,
ri
o
,
ri
o
) distribution, usually simply a homogeneous one is assumed
and the measurement procedure is mimicked through computer simulations. Namely, for
each active antenna the forward problem corresponding to the solution of a vector wave
equation or in general the numerical solution of Maxwells equations, yields the field
distribution all over the receiving antennas. Exactly the field calculated on the receiving
antennas, when gathered from all illuminating active antennas, it setup a calculated dataset.
On the other hand the field calculated all over the structure and particularly over the
virtual body area is exploited within the methodology elaborated herein for the
evaluation of a Sensitivity or Jacobian matrix. This is obtained through a closed form
sensitivity equation established through a combination of an Adjoint Network Theorem
following the Electromagnetics Reciprocity Theorem approach. Its entries exactly reflect the
sensitivity of the calculated field at each sensing antenna with respect to a differential
change of each pixel/voxel unknown parameter. With the availability of this information an
algorithm minimizing the differences between measured and calculated fields (the complete
datasets) based on a least square approach is established, which herein is efficiently
implemented exploiting the sensitivity matrix in its closed form expression.

High to Microwave Frequencies Imaging Techniques

149
At this point two serious problems are introduced related to the inherent properties of the
sensitivity matrix. Firstly, its entries depend on the unknown properties (
i
,
ri
,
ri
), hence
the system to be solved is a non-linear one. Secondly, this sensitivity matrix is usually an ill-
posed one or a singular matrix. This latter means that if a singular value decomposition is
performed (not eigenvalue since this is a rectangular MxN matrix, where the number of
measurements M should be much greater than the number (N) of unknowns as M>>N in
order to confront measurement errors, noise and ill-posedeness) the maximum singular
value appears a few orders of magnitude larger than the minimum one. This inherent
property is closely related to the measurement (generally the data collection) setup and can
be significantly improved by intuitive techniques. Returning to the nonlinearity its direct
consequence is that the resulting parameters (
i
,
ri
,
ri
) provided by the Minimization first
iteration are not the true ones but only a better approximation, if the reconstruction process
was successful. Hence, the whole procedure should be repeated again and again until the
difference between the measured and the calculated dataset becomes comparable to the
expected measurement error and/or the required convergence is achieved. Obviously, at
each iteration a new sensitivity matrix and a new data set is evaluated and used.
In the following sections the reader will be introduced to the employed approaches from
both the mathematical-computational as well as the electromagnetics point of view. But
mostly the open research challenges will be pointed out motivating new research and paths
toward RF-Microwave imaging practical implementation.
The forward and inverse problems general characteristics are discussed in the second
section, after the introduction to high and microwave frequencies imaging modalities. The
procedure to formulate and numerically solve the forward problem is given in the third
section, for both the high (MHz) and microwave regimes. Within the foarth section the cost
function is first setup and the Perturbation Methodology for its direct iterative solution is
then introduced step-by-step from static to microwave imaging. The fifth section elaborates
on the establishment of the sensitivity Equation based on the Adjoint Network Theorem
for the microwave band and its equivalent Electrical Networks Compensation Theorem
for the MHz range. Before presenting the forward and inverse problem numerical results
(sixth section) the importance of the study of the sensitivity or Jacobian matrix is pointed
out. This is a very promising research area especially when modern Principal Component
Analysis (PCA) or its counterpart Proper Orthogonal Decomposition (POD) approaches are
employed. Either PCA or POD are based on a Singular Value Decomposition of the
sensitivity matrix (rectangular matrix) by an algebraic manipulation of its eigenvectors.
These techniques present prospects for novel and computationally efficient methodologies
for both the forward and the inverse problem solution.
The last section is devoted to numerical results. A series of successfully reconstructed
conductivity and permittivity distributions for both the MHz and the microwave regimes
will be presented. The algorithm performance will be discussed and possible improvements
constituting future research areas will be suggested.
2. Forward problem & inverse problem characteristics
The first step in the course of establishing an imaging modality refers to the construction of
the appropriate computer model, which should reflect the practical geometry as closely as
possible. This model serves in twofold, first it should enable an approach which mimics the

Medical Imaging

150
measurement procedure and secondly it should offer the ability to represent the unknown
distribution (,
r
or both) as a set of successively improved-updated variables. Mimicking
the measurement procedure refers to the solution of the governing differential equations,
identified as a generalized Laplace or Poisson equation in MHz range, while the full wave
Maxwell equations must be considered in the microwave regime. The practical bodies of
interest are of arbitrary shape and present a complicated inhomogeneous internal structure
in both and
r
.
In general, these objects should be represented as three dimensional (3-D) models, however
in this case both the forward and the inverse problems become very complicated with high
computational demands. Besides these, the data collection or measurement strategy is also
complex. The possibility of a two-dimensional modeling even an approximate one could be
very convenient, since it could simplify the forward problem solution, restricting the data
collection approaches to just a few and overally resulting to a straightforward reconstruction
algorithm. The question is when the involved approximations are acceptable and whether
these introduce any fictitious mathematical or numerical complications? Let as elaborate
next on the issue of a 3-D versus a 2-D modeling.
2.1 Non-linearity and Ill-conditioning of the inverse problem
The issue of two-dimensional modeling is not rhetorical, since it has been extensively exploited
in most of the established imaging modalities like x-rays computerized tomography (CT).
However, there is a significant difference between x-rays and electromagnetic tomography in
that x-rays propagate along straight lines, but in contrary electric current and electromagnetic
waves in general flow (stream-) lines are curved whenever they intercept an interface where
the conductivity or permittivity changes [e.g. from (
r1
,
1
) to (
rz
,
z
)], as it usually happens in
biological structures. This causes severe problems not only in worsening the option of a 2-D
approximation but also in rendering a non-linear inverse electromagnetic problem. A
fundamental reasoning for this difficulty is explained by the fact that current streamlines or
wave propagation directions curvature strongly depends on the conductivity (
z
/
1
) and
permittivity (
z
/
1
) contrast at these interfaces. But these contrasts are indeed the unknown
quantities to be sought by the electromagnetic imaging algorithm. Namely, when the inverse
problem is finally formulated into a discrete system of equations the stiffness matrix elements
will depend on the unknown (,
r
) distributions hence comprising a non-linear system. In
general, this phenomenon can be identified as an internal scattering and/or diffraction which
is equally present in Acoustical or Ultrasonic imaging.
Besides the non-linearity, the streamlines curvature appearing in bodies with very high or
r

contrast (e.g. between blood and fat or bone tissues) results to regions with very high current
densities (e.g. blood) while at others this is very low (in fat or bone). Similar high variations
occur in the electric field intensities mainly due to the high
r
constrast. This is exactly what
causes the high variation in the sensitivity of the measured quantity (voltage or electric field)
with respect to the unknown or
r
values. To understand this phenomenon assume a tissue
(area) which is isolated from the current flowing through the body, this will in turn present a
very low sensitivity as it cannot affect the injected current, which will not be able to see
it. Hence, its or
r
values cannot be reliably reconstructed. Mathematically, this high
variation in the sensitivity over the body to be imaged is depicted as a high singularity or
high ill-conditioning in the Jacobian matrix [J], which comprises the inverse problem


151
matrix (as [J][J]
T
). As will be explained latter the severity of ill-conditioning is defined by the
ratio of the larger to the smaller singular value of the Jacobian matrix (singular instead of
eigenvalue since the Jacobian matrix is rectangular due to the requirement of a number of
measurements higher than the number of unknown ,
r
degrees of freedom).
Summarizing the above, the imaging modality elaborated herein constitutes a highly non-
linear and ill-conditioned (singular) inverse problem. The next issue to be discussed refers to
the assumptions involved in a two-dimensional approximation.
2.2 Two-versus three-dimensional modeling
Theoretically an object can be represented by a two-dimensional model (2-D) if it is
homogeneous and uniformly extends to infinity along a direction perpendicular to the
modeled cross-section, which can be arbitrary shaped and inhomogeneous. The question is
now, under what assumptions the current density, the electric potential or electric field
intensity of a practical three-dimensional (3-D) object can be replicated over one of its cross-
sections (2-D model)? Once again recall that this approach is very well suited for x-rays CT
mainly because x-rays travel along straight lines and hence their measured intensity
involves information relevant to the inhomogeneities (varying tissue) along the straight line
from the transmiter to the receiving sensor. Thus by attaching an array of sensors in the
same plane with the source as shown in Fig.1a, all measured intensities depend on the
specific enclosed cross-section. Besides that the specific source location yields an
illumination from a corresponding angle of view, called a projection angle from physical
optics. It is, thus, found very convenient to locate the source and the receivers-sensors
equidistantly along a circular bar (a plastic collar). Rotating this circular structure results to
an illumination of the imaged cross-section from any possible angle of view, to be defined at
the desired number enabling the image reconstruction.
A similar configuration could be in principle setup for the electrical impedance (EIT) or
microwave tomography by planning an array of electrodes or antennas on a single plane
around the desired cross-section as shown in Fig.1b and 1c respectively.
Focusing on EIT or MHz tomography the current density injected through the driven-active
electrode (source) is curved around the objects of lower conductivity (Fig.1), but the same
phenomenon occurs along the third dimension perpendicular to the cross-section under study.
Consequently, the injected current density which offers the means to extract the information
regarding the conductivity distribution to be imaged, cannot be restricted to flow only across
the electrodes plane. Instead this current is spread in all directions around the active electrode
and respectively it is collimated from all directions toward the current-sink active electrode.
Thus, the coplanar voltage-sensing electrodes yield measurements infected by conductivity
inhomogeneities above or below the studied cross-section. Additionally, it is impossible to
estimate the total current flowing across the electrodes cross section. To explain the difficulty,
assume that a 5mA current (I=5mA) is injected during the actual EIT measurements, the
question is what is the current value to be applied to the active electrode in the 2-D model?
This should be only the fraction of the 5mA actually flown through the electrode cross-section,
otherwise by assuming a 5mA value yields overestimated calculated voltages at the electrodes.
Besides this problem, by restricting the current to flow in a single plane in the 2-D model, its
density becomes higher within the highly conducting areas. This phenomenon results to
higher sensitivity related to these areas and causes the Jacobian matrix to be fictitiously more

Medical Imaging

152
singular than its actual 3-D form. Exactly similar behavior is observed in MHz and microwave
imaging where the role of high conductivity regions is also played by the high permittivity
areas, especially as the source frequency is increased.

(a) (b)
Fig. 1. (a) Current density distribution of an inhomogeneous 2-D model at low frequencies
(b) Electric field distribution of an inhomogenous dielectric (
r
) at microwave frequency.
The above observation regarding the higher singularity of the 2-D model as compared to a
3-D model is very important and should be utilized as a guide toward the establishment of
more robust imaging techniques. In particular, most of these complexities stems from the
attempt to solve (simulate) the forward problem as a 2-D cross-sectional model, rather than
caused from the 2-D inverse problem formulation. Hence, whenever the simplification of the
2-D setup is sought, it is a very good idea to adopt the corresponding electrode/antennas
coplanar setup (e.g. Fig.2b, 2c) along with a uniform (,
r
) distribution in the axial direction,
which yields a 2-D inverse problem formulation. However, a finite axial length is considered
and a 3-D numerical technique is employed to solve the forward problem. The important
benefit of this approach refers to the removal of the deviations between measured and
calculated voltages or electric field intensities. Besides that, the fictitious higher sensitivity
matrix singularity is reduced to the inherently existing as well.

Fig. 2. a) An x-ray setup for cross-sectional imaging, different illumination by rotating the x-
ray platform.
b) An electrode array for cross-sectional impedance tomography.
c) An antenna array for cross-sectional microwave imaging, that could be setup on a plastic
transparent to electromagnetic waves collar just like the x-rays CT.


153
A more sophisticated trend toward singularity reduction or as sometimes assumed toward
optimal imaging setup aims at the establishment of the optimum current density
distribution for EIT or in the MHz range and respectively optimal electric field intensity
distribution. As explained above all the involved complications stem from the current or
field streamline curvature which depends on the unknown (,
r
) discontinuities. But it is
intuitively expected for EIT that these curvature may become smoother (or streamlines
tending toward straight lines) when the active electrodes (both injecting and current sink)
is increased. This phenomenon can be diaesthetically understood by realizing that all
current (or field) streamlines are emanating from the source or converging toward the
sink electrode, traveling toward all possible directions. Thus, their curvature becomes
maximum for point electrodes and smooths down as its size is increased. This approach
was forced to its limits in the EIT case by the groups of Isaacson [7] and Lionheart [8],
where the active electrodes were increased at the limit of almost covering the entire
objects surface, retaining only small gaps between them for isolation. This, voltage or
electric field sensing electrode or antennas do not presume any large surface and they
could retain a small size even almost infinitesimal (point sensing electrodes). Recall that
the primary aim of this trend is to reduce the sensitivity or Jacobian matrix singularity,
explicitly by reducing the ratio of its larger to the smaller singular value. Conversely
means smoothing out the sensitivity over the body surface or volume by forcing more
current to flow through low conductivity regions and lowering the current density within
high conductivity areas. However, the Authors experience shows that one should not
exaggerate in increasing the electrode or antenna surface, but the current should be
allowed to follow its natural paths through the object, since this is indeed the mechanism
of extracting information from the object interior. As a rule of thumb the electrode or
antenna size could be increased only up to the point that the singularity becomes
manageable by the reconstruction algorithm, which should withstand relatively high
sensitivity variations. In turn the rotation of the active antenna illuminating the structure
from all possible angles would ensure a higher overall-total information extraction.
2.3 Dual mesh discretization
Returning back to the different forward (3-D) and inverse 2-D model, another characteristic
should be taken into account, which is related to the calculation accuracy and the desired
imaging spatial resolution. Starting from the latter it depends on the available number of
linearly independent measurements, which is in turn defined by the number of active and
sensing electrodes/antennas, as it will be explained later on. Also, keep in mind that
measured values are corrupted by noise and the measurement inaccuracies, including
quantization noise introduced during the analog to digital conversion. Thus a reliable
inversion scheme requires a quite higher number of measurements (M equal to the number
of equations) than the number of unknowns (N) in order to cancel out these inaccuracies
through minimization. As a rule of thumb the number of equations should be almost double
than that of the unknowns (M>2N). Conversely the number of electrodes/antennas
corresponds to the achievable spatial voltage or filed intensity sampling, but it is actually
defined by the practical setup as the electrode/antenna size and the hosting object
dimensions on which these will be attached. Concluding the above parameters define an
upper limit in the achievable number of unknowns (N) and hence the offered inverse
problem spatial resolution-discretization. Thus the in principle continuous (,
r
)

Medical Imaging

154
distribution should be discretized into N locally homogeneous elements comprising the so
called piece-wise constant distribution (
i
,
ri
, i=1-N). This is actually formed by the
reconstruction-mesh, which is in general a 3-D one (Fig.3a), but it could also restricted to a
2-D assembly of bars with in general arbitrary cross-section (Fig.3b with brick bars).

(a) (b)

(c) (d)
Fig. 3. Examples of reconstruction coarse meshes and forward fine meshes as: reconstruction
(a) 3-D and (b) 2-D cases, forward: (c) 3-D and (d) 2-D cases corresponding to the structures
(a) and (b) respectively.
The reasoning described above yields usually a coarse reconstruction mesh, which is always
inappropriate for the forward problem solution. This one purpose is to ensure calculations
providing an accuracy of the same order as the available measurements. In the low
frequency EIT case it was proved by Barber and Seagar [9] that an accuracy of the order of
0.1% is necessary for a reliable reconstruction. On the other end, in the microwave regime it
is widely understood that an acceptable simulation asks for a discretization with elements
size smaller than one tenth of the wavelength (/10) for Moment Method (MoM) or Finite
Element techniques (FEM), while the finest mesh of /32 is asked by Finite Difference Time
Domain (FDTD) to ensure its stability. In any case trying to fulfill these requirements a very
fine mesh is necessary for the forward problem solution, which also yields piece-wise
constant (,
r
) distributions. However, the inverse problem will be iteratively linearized first
around the
0 0
( , )
r
o c initial guess and subsequently around the most recently updated
( , )
k k
r
o c distribution at its k-th iteration. Multiple forward problem solutions are then
required on this ( , )
k k
r
o c distribution, hence the forward and reconstruction meshes
should be compatible. A very straight forward approach is to subdivide each of the
reconstruction mesh into smaller forward mesh elements, as shown in Fig.3(c)-(d). Even


155
though the forward mesh elements comprising each reconstruction cell have the same
constitutive (
i
,
ri
) parameters, this subdivision in necessary in order to fullfill the
voltage/field interpolation requirements. Explicitly, in most cases linear interpolation
functions are considered within the established numerical techniques (MoM, FEM, FDTD).
In turn the field variation especially around fine structures and around conductors are very
high, thus asking either for highly non-linear interpolation functions or conversely very fine
mesh of linearly interpolated elements (/10 up to /32).
Overally, up to now the basic configuration and limiting characteristics for the forward and
inverse problems are explained. We are, thus, ready to proceed to the description of forward
problem governing equations and their numerical solutions.
3. Forward problem
Electromagnetic problems are obviously governed by Maxwell equations in their full vector
differential form, which in general include a temporal variation [10]. Even well-known, their
solute, ion for complicated extremely inhomogeneous structures like the Human body
constitutes a formidable task, especially when the conductivity anisotropy of skeletal and
myocardial muscle is to be considered. Since, the inverse problem asks for multiple forward
solutions (at least one for each illumination-projection angle) at each of its iterations, then a
computationally efficient technique is inevitable. For this purpose a series of simplifications
are necessary in order to reduce the involved complexity.
3.1 Excitation rationale
The temporal dependence could be employed in imaging modalities, since radar type
techniques could be employed, but these are efficient for electrically large bodies
(dimensions of multiple wavelengths) located at distances of at least a few wavelengths
away from the illuminating source. Besides that pulsed-waveform excitation could be
exploited for imaging in general, where the received pulse delay () in respect with the
transmitted one provides the useful phase difference (). In turn based on the ratio of
imaginary to real part of the voltage or electric field is readily calculated, while the in
principle required phase sensitive sensors are replaced by scalar measuring only their
amplitude. As will be explained next, this information is directly proportional to (
r
/) or
inversely proportional to the dielectric loss tangent [tan=/(
0
r
)] of the media profile to be
reconstructed. This imaging approach could be classified as a multistatic phase radar and
it is very challenging as well as a promising modality, especially regarding the involved
relative simplification of the sensing-measuring instrumentation. However, it has only
received a very limited attention and to the Authors knowledge, there was only one Russian
group working on that, [11]. Conversely, most of the research groups activated in the field
in MHz and microwave imaging (including our own) assume time harmonic (sinusoidal)
excitation due to its simplicity in the forward problem solution, thus the following analysis
will be restricted to this case. Note that this excitation eliminates the temporal dependence
but also avoids (dues not account for) the materials frequency dispersion (),
r
() which
introduce more difficulties. In contrary this dispersion can be exploited as an additional
degree of freedom enabling dynamic imaging or calibration purposes and even a spectral
imaging by performing measurements and reconstructing the media profile at multiple
frequencies ((
i
),
r
(
i
)) selected at appropriate steps (
i
).

Medical Imaging

156
3.2 Time harmonic fields & currents
Maxwell equations for the time harmonic excitation of the form
j t
e
e
are significantly
simplified since the temporal variation is substituted as / t je c c = , while the actual field
( ) ( ) t r t r , , ,
, ,
, ,
and source quantities ( ) t r ,
,
,
are replaced by the corresponding complex
phasors ( ) ( ) ( ) , , E r H r J r
, , ,
, , ,
as:
( ) ( )
( )
j t
t r E r e , Re
e
=
, ,
, ,
(1a)
( ) ( )
( )
j t
t r H r e , Re
e
=
, ,
, ,
(1b)
( ) ( )
( )
j t
t r J r e , Re
e
=
, ,
, ,
(1c)
In turn Maxwell equations for time harmonic fields read:

0 r
xE -j H e V =
, , ,
(2a)

0 r
xH j E J ec c V = +
, , , ,
(2b)
D E c V = V =
, , , ,
(2c)
0 0 0
scalar
B H H
=
V = V = V =
, , , , , ,
(2d)
Where the source quantities current ( ) J
,
and charge () densities should obey the continuity
equation:
/ J t j e V = c c =
, ,
(3)
Additionally, for the conducting media assumed herein there is a conduction current ( )
c
J
,

flowing at any arbitrary point (movement of charges due to coulomb forces and the
presence of electric field) as
c
J E o =
, ,
. Thus, the current density J
,
appearing in (2b) is
comprised of the conduction current
c
J
,
and the current impressed by the source as
imp
J
,
.
Substituting in (2a) we may define an equivalent complex permittivity
c
or an equivalent
complex conductivity
c
as:

0 r 0 rc c
xH j j
imp imp imp
E E J E J E J ec c o ec c o V = + + = + = +
, , , , , , , ,
(4)
where

c 0 rc 0 r 0 r
0 r
(1 tan ) 1 j j
o
c c c c c o c c
ec c
| |
= = =
|
\ .
(5a)

0 r c eff
j o o o ec c = = + (5b)


157
Note that the term related to the actual permittivity () is known as displacement
current
( )
0 d r
J E t j E c ec c = c c =
, , ,
. A usual question regarding the body inherent bioelectric
sources has actually no place herein since their spectral content is restricted to less than
about 10KHz. Thus the impressed current in equation (4) is solely due to the source injecting
current or illuminating the body. For safety reasons the injected current density should be
less than 10mA/cm
2
according to recommendation included in [12]. Besides these, recall the
possible anisotropy issue which only concerns the conductivity of skeletal muscular tissues
and the myocardium. Both of them exhibit a fibrous structure, where each fiber has a high
conductivity interior surrounded by a thin low conductivity membrane. It is this structure
that causes the conductivity anisotropy where parallel (along) the fibers it is
//
=7.3mS/cm
while transverse to them it is only
=0.49mS/cm. The question is how to include this

anisotropy into the computer model. Mathematically the anisotropy is accounted for by a
second rank tensor o , but this could be only convenient if one of the coordinate axis could
be oriented parallel to the muscle fibers. However, these fibers are curved and even twisted,
hence they could be only represented by a full 33 tensorial o . It is, indeed, very easy to
model a diagonal tensor of either o or
r
c parameters, but the mathematical and
computational complexity of full tensor parameters is to high to be considered in most
cases. Thus, herein the inherent muscles anisotropy will not be considered.
Returning back to Maxwell equations (2), it is well understood that the divergence equations
(2c) and (2d) can be derived from the curl equations (2a) and (2b) by using continuity equation
(3). Thus one may easily conclude that it is only necessary to solve the implicit pair of curl
equations (2a), (2b) in conjunction with the boundary conditions for the tangential to the
various media-tissue interfaces electric and magnetic field components. Recall that the latter
are indeed obtained from the corresponding integral form laws resulting from the same curl
equations, [10]. However, one should keep in mind that this conclusion is exactly valid only
when an exact analytical solution is sought. In contrary when numerical solutions are adopted
along with the internal structure discretization accompanied by approximate interpolation
functions within each piece-wise homogeneous element, then the aforementioned exact
conditions can be violated, resulting to field distributions not obeying the divergence
conditions. Even though this could be kept in mind as a possible source of inaccuracies,
usually their effects can be negligible. Thus there is indeed a very powerful method to
discretized and solve the interdependent pair of vector curl differential equations (2a) and (2b),
such as the Finite Difference Frequency Domain (FDFD) method established in our previous
work, Lavranos [13] in two-dimensional curvilinear coordinates and currently been extended
to 3-D structures as in Lavdas [14]. This approach can easily account for material anisotropy as
well, within the usual FD limitations regarding the interfaces spatial resolution. Even though
this is a promising approach we have not yet employed it within inverse problem solutions,
but instead the Finite Element Method (FEM) is mostly employed for this purpose. Since FEM
is formulated in an integral form, it is more convenient to decouple the two curl equations in
order to formulate two separate vector wave equations for the electric and the magnetic field
respectively as, e.g. Jin [15] or Volakis [16](p.5) :

2
0 r 0
r 0
1
x -k j
imp
xE j E J
o
c e
ec
| | | |
V V =
| |
\ . \ .
, , , ,
(6a)

Medical Imaging

158

2
0 r
rc rc
1
x -k
J
xH H x
c c
| | | |
V V = V
| |
\ . \ .
,
, , ,
(6b)
Either wave equation could be solved employing FEM after constructing a functional (or
weak form) to be minimized. This task is achieved utilizing either a variational approach or
the Galerkin approach, e.g. Cangellaris [17]. The choice among the two wave equations is
mainly undermined by the more convenient enforcement of boundary conditions. For the
biomedical applications the media-tissues are non-magnetic (=
0
) and the inhomogeneity
is approximated by step changes in conductivity and permittivity. Thus the related
boundary conditions are directly enforced to the electric field as Dirichlet type. Conversely,
if ones solves for the magnetic field the same boundary condition would be translated into a
Newmann type and a differentiation after the solution (prone to numerical errors) would be
performed to evaluate the desired electric field. Hence, it is preferable to choose a
straightforward solution of the electric field wave equation (6a).
Observing equation (6a) and following its solution in the next section one would realize that
this is anything else but a trivial task, involving huge computational resources to solve over a
Human body or just a Human torso. For this reason two simplifications will be presented next,
a quasi-static approach valid for the frequency range up to a few MHz and a second approach
with unidirectional dipole sources (aligned along the z-axis) and even a 2-D scalar Helmholtz
wave equations excited by line sources which significantly reduce the involved complexity.
3.3 Quasi-static approach
Observing equations (2a) and (2b) one may recall the basic understanding that the
electromagnetic waves are created and propagated due to the temporal variation of
impressed current source which is inherited to the generated field. Explicitly, a time
depended impressed current generates a temporally changing magnetic field through (2b).
In turn the varying magnetic field / H t j H e c c =
, ,
produce an electric field with an
identical time variation through (2a). This / E t j E c ec c c =
, ,
regenerates a changing magnetic
field through (2b) and this cycle is infinitely repeated producing a propagating wave.
However, this chain becomes very weak or breaks when either j=j
0
r
or j=j
0
r

quantities become very small (where
0
=8.854x10
-12
F/m and
0
=410
-7
/m). Biological
tissues are non-magnetic
r
=1 while due to their high water content they present a very high
dielectric constant which starts from about
r
~1000 (or more) at 100KHz range and reduced
to about
r
~80 in the microwave (GHz) range. Based on this reasoning it was long ago
realized, e.g. Price 1979 [18], that magnetic effects can be ignored for frequencies lower than
about 10MHz. Namely, the electric filed generated by magnetic field temporal variations
becomes negligible. From the wave propagation point of view the wavelength in a media
r
~100 at f=10MHz is
0
3
g r
m c = = (and
0
=c/f), which means that at a distance
d=
g
/4=75cm the field or voltages changes due to wave propagation from its maximum
value to zero and vice-versa. Hence for a thorax with larger dimension of about 40cm the
10MHz frequency constitutes indeed an upper limit.
In view of the above, the ignorant question asks then from where comes the varying
electric field and the desired electric potential? Now we have to reconsider the divergence
equation (2c) and the charge conservation law (3) as:


159
0 E E V V ~ = V
, , ,
(7a)
E V c c V = V V =
, , , ,
(7b)
J j E j V j e o e o e V = V = V V =
, , , , , ,
(7c)
Equations (7b) and (7c) seems to be in contradiction since, they call for different voltage
definition or making V to voltage uniqueness. Here comes the original Maxwell observation,
e.g. see Jackson [19]( p.238), that one could differentiate (2c) and substitute in charge
conservation equation as:

0
(2 )
0 (3) 0
D
D
J
c
t
t t
D
or J J
t t
c c
V + V =
V =

c
c c
`
| | c c

V + = V + =
|
c c )
\ .
,
,
, , ,
,
, , ,
(8a)
Substituting the constitutive relations J E o =
,
and D E c = into (8a) and restricting ourselves
to time harmonic fields we get:
( ) 0 j E o ec V + = (8b)
In turn utilizing (7a) the so called generalized complex Laplace equation is obtained:
( ) 0
c
r V c V V =
,
,
(8c)
where:
( ) ( ) ( )
c
r j r c c o e =
, ,
(9)
However, besides this classical approach, (8c) can be obtained by dividing (7c) with j and
adding the result to (7b). Charge density at the operating frequency indeed exist over the
metallic electrodes obeying (7c), but these can be taken into account through the boundary
conditions to be enforced on the voltage distribution during the FEM formulation of the
Newmann type:

on the electrode
on the body air interface
c
o
J
V V
j j j
V
n n j
n
o
ec e c
e ec
c c | |
= =
c
|
c c
\ .

c
(10)
where cV/cn denotes the normal derivative at the interface. At this point is important to
note that (10) applies to both active-driven as well as passive (sensing) metallic electrodes,
since current and charge density is induced on them. Most important is that the field and
current singularity at metallic edges is a local phenomenon and thus it occurs at any
frequency from DC to microwaves. Referring to a classical electromagnetics text book, e.g.
Collin [10](p.25), field components normal to the edge as well as current density
components parallel to the edge exhibit a singular behavior tending to infinity as 1 / ,
where the distance from the edge. Conversely, field components parallel to the edge and

Medical Imaging

160
current density normal to that vanish as 0. This is an important behavior and it should be
taken into account when an accurate modeling is sought. To clarify this behavior the current
density on a patch electrode driven by a current 1 is shown in Fig.4a and the induced current
density on a passive electrode is sketched in Fig.4b. Note that the corresponding integrals
denote the total current I on the driven and zero on the passive electrode. Obeying equation (3)
the charge density on the active or passive electrodes exhibits identical singular behavior.

(a) (b)
Fig. 4. Current density on metallic strip electrodes exhibiting the inherent singularity as a)
emanating from an active electrode and b) induced on a passive electrode, e.g. [19](p.165)
3.4 Wave equation for microwave sources in biological media
The forward problem for the microwave imaging is governed by the wave equations (6a) or
(6b). It was explained before that the electric field wave equation (6a) offers a
straightforward enforcement of boundary condition at the different lossy dielectric
interfaces as a rigid-Dirichlet type. Besides that, observing the first term of equation (6a) this
could be significantly simplified for the non-magnetic biological media as =
0
=const or
r
=1. In contrary the corresponding first term of (6b) is complicated as it contains the
spatially dependent complex ( )
rc rc
r c c =
,
. In view of this clarification the electric field wave
equation (6a) can be simplified by utilizing the identity:

2
( ) E E E VV = V V V
, , , ,
(11)
For the divergence of E
,
we may again exploit the original Maxwell observation [19](p.238)
as given in equation (8c) which is generally valid and using the symbolism (5a)
c
=+j,
then (8c) reads:

( )
( ) ( ) 0 ( ) ( ) ( ) ( ) 0
c c c
r E r r E r E r r c c c V = V + V =
, , , , , ,
, , , , , ,
(12a)
or

( ) ( )
( )
( )
c
c
E r r
E r
r
c
c
V
V =
, ,
, ,
, ,
,
,
(12b)


161
Substituting equation (12b) into (11) and the resulting expression back into (6a) the desired
electric field wave equation for biological media is obtained:

2 2
0 0
0
( )
r r r imp
E j
E k j E j J
j
o ec o
c e
ec o ec
| | | | V +
V + + V =
| |
|
+
\ . \ .
, ,
, , , , ,
(13)
It is now convenient to define a complex inhomogeneous wavenumber ( ) K r
,
,
in the usual
form as:

2 2
0
0
( )
( ) ( )
r r
r
K r k r j
o
c
ec
| |
=
|
\ .
,
, ,
(14)
With the aid of (14) the electric wave equation reads

2
2 2
0
2
( ) ( )
( ) ( ) ( ) ( )
( )
r imp
E r K r
E r K r E r j J r
K r
e
| |
V
' V + + V =
|
|
\ .
, ,
, ,
, , , ,
, , , ,
,
(15)
where ( ) r'
,
denotes the source spatial vector.
3.5 Two dimensional structure-inhomogeneous cross-section
The convenience offered by a two dimensional imaging are already explained in section II.2
along with the necessity of solving a three-dimensional forward problem. The geometry
considered in this case is a cylindrical structure with an arbitrary shaped cross-section but
uniform along its axial
z -direction. Likewise the material complex permittivity to be sought

is inhomogeneous in the (xy) cross section but uniform along the z-direction, like an
assembly of different dielectric bars, as ( ) ( , )
c c
r x y c c =
,
. Due to the uniformity in the axial
z -
direction the structure can be considered as an inhomogeneously loaded lossy dielectric
open waveguide of arbitrary cross-section which is excited-illuminated by different type of
sources. In view of this uniformity the electromagnetic field within and outside this open
waveguide could be expressed as a superposition (expansion) of an in general infinite
number of modes either propagating or evanescent just as classically done in the mode-
Matching technique. Even though this is a very promising approach to our knowledge it has
not yet being exploited for imaging purposes. Let us give a formal description of such a
methodology. For each one of the possibly excited modes the field dependence in the
z -
direction can be denoted as:

( , , ) ( , )
i
j z
i i
E x y z e x y e
|
=
,
,
(16a)

( , , ) ( , )
i
j z
i i
H x y z h x y e
|
=
, ,
(16b)
where
i
is the i
th
mode complex propagation constant. When is real or exhibits a small
imaginary part due to losses, then it represents a propagating wave. Conversely, when is
purely imaginary then it represents an evanescent or non-propagating wave which can be
excited by the source but is exponentially attenuated away from that. Namely, it exists only

Medical Imaging

162
around the source, it does not transfer energy but it only stores electric and magnetic energy
in its neighborhood, thus contributing only to the reactive part of the source (antenna) input
impedance. In order to utilize the eigenmode expansion approach the characteristics of
every possible mode, its eigenfunctions ( , ), ( , )
i i
e x y h x y
,
,
and the complex propagation (
i
)
constant should be estimated first. The corresponding expansion formally reads:

1 1
( . )
i
j z
ei i ei i
i i
E w E w e x y e
|

= =
= =

, ,
,
(17a)

1 1
( . )
i
j z
mi i ei i
i i
H w H w h x y e
|

= =
= =

, , ,
(17b)
Where w
ei
and w
mi
represent the modal amplitudes or weighting factors which depend on
the source type and its location, see for example our work [20]. These can be estimated
through a power conservation low which is the basis of a mode matching technique.
Explicitly, each specific source can be enclosed inside a fictitious box inside which the field
is described with the aid of a numerical technique on a fine mesh exploiting the knowledge
of source modeling, e.g. Volakis [16](p.238). Outside the fictitious surface the eigenfunctions
(17) is considered and the weighting factors w
ei
and w
mi
are evaluated through field
continuity conditions across this psudo surface. It is important to keep in mind that the
power conservation low should be enforced, which is preferably formulated in conjunction
with the modes orthogonality properties, e.g. [21].
The above mode matching approach pressumes the knowledge of the eigenvalues (
i
) and the
eigenfunctions
i
, h
i
e
,
,
which for the considered inhomogeneous cross-section can only be
acquired numerically. For this purpose the wave equation (6a) shall be formulated as an
eigenvalue problem and the usual approach is based on the separation of both the electric field
e
,
and the nabla
( )
V
,
differential operator into axial (
z ) and transverse (t) components as:

,
t z t x y t z
e e ze and e e x e y h h zh = + = + = +
, ,
, , ,
(18a)

,
t t t
z j z x y
z x y
|
c c c
V = V + = V V = +
c c c
, , , ,
(18b)
The detailed procedure given in a lot of textbooks, e.g. Volakis et al [16](p.98) is summarized
as follows:

t t t z t
e e e j e z
( ) | V = V + V +
, , ,
, , ,
(19)
Which can be substitute into (6a) to yield a pair of differential equations for the axial (e
z
) and
transverse (
t
e
,
) field components as:

( )
2
0
1
0
t t t t z t rc t
r r
e j e j e k e
|
| c

| |
V V V + =
|
\ .
, , ,
, , ,
(20a)


163

2
0
1

( ) 0
t t z t rc z
r
e j e z k e z | c
| |
V V + =
|
\ .
, ,
,
(20b)
The pair of equations (20) constitutes the eigenproblem to be solved numerically. Observing
that the transverse component
t
e
,
mainly occurs a product j
t
e
,
, the eigenproblem in further
simplified by letting
t t
j e e | '
, ,
and multiplying (20a) by j yields:

( )
2 2
0
1 1
t t t t z t rc t
r r
e e e k e | c

| |
' ' ' V V + V + =
|
\ .
, , ,
, , ,
(21a)

2 2 2
0
1

( )
t t z t rc z
r
e e z k e z | | c
| |
' V V + =
|
\ .
, ,
,
(21b)
Equations (21) constitute the desired eigenproblem for the eigenvalue. The interested
reader may contact our previous work [22] on open waveguides which was carried out
within Dr. Allilomes doctoral thesis [23] also available online (in Greek).
Even though the above 2-D analysis offered some simplification the forward problem still
remains complicated and computationally demanding. A truly two-dimensional approach
presumes the sources to be two dimensional, i.e. tending to infinity along the z-axis, just as
in the case of an infinite line source.
3.6 Two dimensional object excited by infinite line sources
The field of an infinite electric line source embedded in a homogeneous media is well
studied and can be found in any classical electromagnetics textbook, e.g. Balanis [24]( p.571).
Since the line source is infinite its current density must be constant and if it is considered
aligned along the z-axis, then:
Line Source:

2

( ) ( )
L
I
J r J z z
A
o ' ' = =
, , ,
(22)
where I is a sinusoidal current (I=I
0
e
jt
) and the polar radial coordinate transverse to
z , as
A let the line source cross section assumed very small.
For a homogeneous medium the field generated by an infinite line source is a Transverse
Magnetic (TM
z
), H
z
=0 with only and axial electric field
z
E E z =
,
and its transverse
eigenfunction is identical to a cylindrical Hankel function of the second type
(2)
0
( ) H k
assuming that the line source is at =0, for details in [24]( p.571). For the case of the
inhomogeneous cross-section the TM nature of the wave is exactly preserved, thus
z
E E z =
,

and E
x
=E
y
=0. Besides that the two-dimensional complex dielectric profile is uniform along the
z-axis or ( ) / 0
c
d r dz c =
,
and consequently from equation (14) it is ( ) / 0 K r z c c =
,
. In view of the
above simplifications the third term of equation (15) vanishes as also explained by Fang [25] :

2 2
2 2
0
( ) ( ) 0
( ) / 0
x y
E E
E r K r
K r z
= =
V =
`
c c =

)
,
, ,
,
(23)

Medical Imaging

164
Consequently, the wave equation (15) is reduced to a scalar Helmholtz equation:
2-D & Line Source:

2 2
2 2 0
( ) ( )
r ZL
E r K E r j J e V + =
,
, ,
(24)
A more practical structure to be considered herein is comprised by the same lossy dielectric
profile
c
(x,y) but with a finite extend in the z-direction and illuminated by a circular array of
infinitesimal
z -directed dipoles located t the mid z-plane and encircle the object to be
imaged. Even though this can again be classified as a 2-D inverse problem and solved
accordingly, the corresponding forward problem is governed by the 3-D wave equation (6a)
or (15) and it will be solved with the aid of a 3-D finite element method.
3.7 Variational formulation
The finite element will be employed for the numerical solution of either the generalized
Laplace equation (8c), the scalar wave equation (24) or the general vector wave equation
(6a). For each case the structure will be discretized with the aid of the appropriate forward
fine mesh. The first step of this procedure refers to the formulation of the differential
equation into an appropriate integral functional or weak formulation which will, in turn, be
minimized to obtain the numerical solution. Let us start from the simpler case.
3.7.1 Functional for the generalized laplace equation
For these relatively simple scalar problems the variational formulation is usually adopted.
According to the approach, e.g. Jin [15]( p.72), the differential equation (8c) along with its
boundary conditions yields the functional to be minimized. These are as follows:
1. Neumann boundary conditions at the active current electrodes and along the body-
surrounding air interface:

0
, ondrivenelectrodes
, onbody-airinterface
c
J
V
j
V
n j
n
0
ec
0
0 ec
0
(25)
where / V n 0 0 is the potential derivative in the direction normal to the unknown object
surface.
2. Mixed Dirichlet and Neumann boundary conditions between different media regions-
object elements (i,j) as:

i j
V V =
and
j
i
ri rj
V
V
n n
0
0
c c
0 0
- -
=
(26)
These are natural boundary conditions which are an inherent property of the differential
equation (8c). Namely, these are automatically imposed through the FEM solution of (8c).
Note that the above does not include any Dirichlet condition, namely only the derivatives
are defined (delta change), hence the resulting solution will be floating and thus non-
unique. To avoid this problem at least one point must be grounded (V=0)


165
Absorbing boundary conditions
At this point recall that the Generalized Laplace equation solution domain is basically
infinite, since for high frequency sources (i.e. in the MHz range) the electric field
( )
E
,
and the
related scalar potential (V) extends in the air to infinity obeying in general the
Sommerfeld radiation condition. Explicitly, for very low frequencies the static potential in
the air surrounding the structure vanishes since the conduction current
c
J E o =
, ,
is zero and
0 o = in the air. However, as frequency () is increased a significant displacement current
d o
J j E ec =
, ,
flows in the air around the structure and this phenomenon must be taken into
account. But, for a numerical solution the analysis domain must be finite and restricted if
possible to the actual body of interest. For this purpose the solution domain is enclosed
within a fictitious closed surface of canonical shape on which transparent Absorbing
Boundary Conditions (ABC) are enforced. Transparent ABCs means that they should not
disturb the field distribution but conversely to behave like perfect absorbers (if possible),
absorbing any wave incident on them without any reflection, just like a termination load.
For the two dimensional forward mesh of Fig.5a enclosed within a fictitious circular
contour-C, the quasi-static potential ABCs read [15](p.97):

1 V
V
ln
0
0
~ on circle
c
= (27)
where
2 2
x y = +
the radial polar coordinate.
The three dimensional structure of Fig.6b is enclosed within a fictitious cylindrical surface
on which the appropriate ABCs should be imposed. In order to understand the ABCs in
both 2-D and 3-D structures it worths to recall that these are extracted from a generalized
type of boundary conditions, [15](p137). For this purpose let us write the generalized
Laplace (8c) or Poisson equation in the general form of:

cx cy cz
V V V
V f
x x y y z z
c c c |
| |
c c c c c c | | | |
+ =
|
| |
c c c c c c
\ . \ .
\ .
(28a)
Considering the solution domain enclosed within a closed surface
1 2
S S S = + on which two
type of boundary conditions can be imposed as:
Dirichlet Boundary Condition:
V = on S
1
(28b)
General (mixed Dirichlet & Newmann) Boundary Condition

cx cy cz
V V V
n V q
x y z
c c c
| |
c c c
+ + + =
|
c c c
\ .
on S
2

(28c)
where
n the outward unit normal vector. Besides these the field and consequently the scalar
electric potential should obey the Sommerfeld radiation condition at infinity which reads,
[15](p.8) or [16]( p.8):

Medical Imaging

166
3-D:
( ) ( ) { }
lim 0
o
r
r jk r
V + + + =
, ,
(29a)
where or E H + =
, , ,
and
2 2 2
r x y z = + + ,which states that the field is comprised of an
outgoing wave with dependence
o
jk r
e
as r . For a two-dimensional case (29a) where
z
E E z =
,
(e.g. TM
z
modes) or
z
H H z =
,
(for TE
z
modes) equation (29a) reduces to:
( ) ( ) lim 0
z o z
r
jk
c
+ + + =
`
c
)
(29b)
where or
z z z
E H + = . The question is now how to express (29b) in the general form of (28c)
to be applied on the fictitious boundary truncating the solution domain, by keeping in mind
that E V = V
, ,
. Additionally it has been proved that for a 2-D case the potential (the static or
the quasi-static approximation) asymptotically behaves as [15](p.97):
2-D: ln V A = as (30a)
where A is a constant.
For a three-dimensional domain the potential in the homogenous background (i.e. air) and
at a large distance from its sources behaves as [15](p150):

A
a
r
~ (30b)
where
2 2 2
r x y z = + + .
For the two dimensional case let 0 V z c c = and the 2-D ABC (27) is obtained from (28c) by
letting q=0 and =1/(ln). Note, that inside the fictitious surface-S there is always an air
layer (
rx
=
ry
=
rz
=1), hence (28c) reads:

1
ln
V V
n V
x y
| |
c c
+ =
|
c c
\ .
(31a)
For a circular fictitious surface it is

n = and for a large enough radius the left hand side of
(29) is reduced to V c c to yield the ABCs of (27), (the related vector identities see e.g.
Balanis [24] (appendix II)
Likewise, for the three dimensional domain eq.(28c) yields the absorbing boundary
conditions.
For a spherical fictitious surface with radius r
c
large enough the absorbing boundary
conditions take the form:

1 V
V
r r
c
~
c
(31b)
which, in turn, can be expressed through (28c) by letting =1/r, q=0 and
n = .


167

(a) (b)
Fig. 5. The object to be imaged is surrounded by the illumintating or sensing antennas array
and is enclosed within a fictitious surface transparent to electromagnetic waves on which
ABCs are imposed: (a) 2-D and (b) 3-D configuration.
For the scalar microwave case where =
z
or H
z
equation (29a) yields the absorbing
boundary condition along a cruved fictitious surface as,

1
2
z
o z
jk

| | c
=
|
c
\ .
(31c)
3.7.2 Functional for generalized Laplace equation
With the aid of the variational technique the generalized Laplace equation (8c) along with
the boundary conditions (25), (26) or their general form (28) and the corresponding ABCs in
(27), are found to be equivalent to the minimization of the following functional:

2 2
1 1
1
( ) ( ) ( ) ( )
2
e
M M
e e
c
e e
V V
F V dxdy JVd F V
x y
0 0
c
0 0
= =
O
(
= + + =
(

} } }
/
(32)
where M is the total number of elements and
e
O is the area of each e-element. Note that the
variational formula (32) is valid for either real or complex quantities according to [15](p.76)
and the references therein. For F(V) to be minimized, its partial derivatives with respect to
the elements nodal voltages must be zero, namely:

( )
0 1, 2,...,
i
F V
for i n
V
0
0
= = (33)
where n is the total number of nodes.
3.7.3 Functional for the vector and scalar wave equations
A variational technique can be employed to formulate the microwave functional for the
vector wave equation (6a) along with the appropriate boundary conditions and the
corresponding Absorbing Boundary Conditions. The resulting functional reads [15]:

Medical Imaging

168

( ) ( ) ( ) ( ) ( )
2
1 1

3-D:
2 2
o
o rc
r
V S
ABCs
s
V
source
jk
F E E E k E E dV r E r E dS
j J EdV
o
c
e
(
( (
= V V +
( (

(

+
}}} }}
}}}
, , , , , , , , ,
_
,
_
(34)
Likewise, the scalar wave equation (24) when a 2-D structure is illuminated by a line source
results to the functional:

( )
2
2
2 2
1 1 1
2-D:
2 2 2
s
z z
o rc z o z z z
S S
source
ABCs
E E
F E k E dS jk E dl j J E dS
x y r
o
c e
I
(
| | c c | | | |
(
= + + + +
|
| |
c c (
\ . \ .
\ .

}} } }}
,
_
_
(35)
The surface integral in eq.(34) is defined over the fictitious cylinder truncating the
solution domain. Likewise the line integral in eq. (35) is defined along a fictitious circle
(
S
) terminating the 2-D mesh. Both these integrals express the absorbing boundary
condition.
3.7.4 Finite element solution of the forward problem
For both the quasi static and the microwave approach the structure must be discretized
utilizing a dual mesh. A fine one for the forward problem and a coarse mesh for the
inverse problem, as shown in Fig.3. The coarse mesh is defined only over the object to be
imaged (Fig.6), while the fine-forward mesh covers the whole solution domain up to the
fictitious line or surface enclosing all constituents including antennas of electrodes.
Besides that every coarse element is subdivided in a number of smaller fine mesh
elements. The nodal FEM approach is employed for the MHz potential estimation as well
as the 2-D scalar Helmholtz equation involving only Ez component referred bellow as
scalar FEM. Conversely, the edge elements technique is adopted for the 3-D vector wave
equation. Since, for the reconstruction mesh the body under consideration is split into
cubic (or rectangular) elements with constant and
r
, so a piecewise homogeneous model
is constructed as:
( )
1
1 withink-thelement
, ,
0 elsewhere
E
c ck k k
k
x y c c
=
= =

(36)
Even though FEM is well described in numerous textbooks, e.g. [15], [16] and [17] for
multigrid approaches, a short description of the basic interpolation functions and the related
element matrices is given next, since they are necessary for the definition and evaluation of
the Jacobian-Sensitivity matrix.
3.7.5 Scalar nodal 2-D approach
Either the potential in the functional (32) or the E
z
component of (35) are discretized
employing first order linear triangular elements with the unknowns (degrees of freedom)
defined on their nodes as:


169
2-D triangular:
3
1
e e e
i i
i
N
=
=
(37)
where
e e
V = or
e
z
E respectively. The scalar interpolation or basis function are given by:

( )
1
, 1, 2, 3
2
e e e e
i i i i
e
N a b x c y i = + + =
A
(38)
e
A = area of the e-th element,
1 2 3 2 3 1 2 3 1 3 2
, ,
e e e e e e e e e e e
a x y y x b y y c x x = = = ,
2 3 2 3 2 3
, , , , ,
e e e e e e
a a b b c c
can be found by cyclic interchange. The above interpolation functions include unknown
coefficients related to the element geometry and its material constitutive parameters. These
are estimated by applying the interpolation functions at the points where the field or
potential is sampled, in this case the element nodes. The resulting system of equations is
analytically solved for these unknowns in terms of the potential or field values at the nodes.
These are, then, substituted back into the interpolation function. In turn, this can be readily
exploited in the conditions minimizing the FEM functional like that of equation (33) through
an analytical evaluation of the derivatives with respect to the unknown potentials (V
i
) or
field components E
zi
. Observe at this point that the involved surface integrals over the
solution domain (e.g. (32) or (35)) are discriminated into a sum of integrals over each
element. The latter yields a system of equations of the form:

( )
e e e e
c
Y V I c
(
( (
=

(39)
or respectively:

( )
( )
e e e
c z z
K E B J c
(
(
( =

(40)
Within the classical FEM the above element equations are assembled to from a unified
system matrix representing the whole solution domain. Hence, this matrix should include
all boundary conditions along with the illuminating sources and passive substructures like
sensing electrodes or antenna models as well. The latter is useful and it should be accounted
at least when the methodology is matured, but it is for now neglected. Regarding the
important absorbing boundary conditions these are usually imposed through the closed
contour integral over
S
in equations (32) and (35). For this purpose the contour
S
is
discretized into line segments coinciding with the corresponding edge of the peripheral
triangles (as depicted in Fig.6)
The corresponding line element matrices are as follows:
Quasi-static ABCs:
2
ln
s
s o
V
K j dl ec

cI
=
}
(41)
2-D microwave ABCs:
| |
1
2
s o i j s
K jk N N d
r
cO
| |
= + I
|
\ .
}
(42)

Medical Imaging

170

Fig. 6. Example of coarse mesh overlapping a fine mesh, for 2-D rectangular object enclosed
in a fictitious circle (
S
)
.
The contour
S
is discretized using line segments which coincide
with the corresponding edge of the peripheral triangles
The last term to be included here is the feeding sources contribution through the
corresponding terms of equations (32) and (35). Herein the simplest possible sources are
considered, i.e. infinitesimal electrodes or dipoles respectively. In both cases the current
density is approximated as uniform ignoring the field/current singularity at the edges (a
phenomenon which should be included in a most complete version), thus given by the ratio
of the feeding current (I) to the antenna or electrode cross section (A) like J=I/A. The related
integrals over the source yield the right hand side [I] or B(J
z
) respectively. The resulting FEM
system of equations takes the form:
Quasi static FEM: ( ) | | | |
c
Y V I c ( =

(43)
2-D microwave FEM: ( ) ( ) ( )
c z c z
K E B J c c ( ( ( =

(44)
Classical numerical techniques can be employed to solve systems (43) or (44), but it is useful
to keep in mind that multiple solutions are necessary (for each inverse problem iteration),
one for each illuminating antenna or active electrodes pair position. Hence a technique
based on the inversion of the system matrix like the LU decomposition is very convenient,
since the solution for each k
th
-right hand side is obtained by a simple multiplication as
1 k k
V Y I
( ( (
=

.
3.7.6 Vector edge elements FEM
For the volume integral of the electric field vector in the functional of (34) first order
tetrahedral vector-edge elements are employed. The electric field E
,
within each tetrahedral
is expanded in terms of the FEM basis functions as, [15]:


171
3-D:
6
1
e e e
i i
i
E N E
=
=
, ,
(45)
e
i
E denotes the tangential field values along the i-th edge and
e
i
N
,
is the vector interpolation
or basis function given by:
( )
1 2 1 2 2 1
e e e e e e e
i i i i i i i i i
N W l L L L L l = = V V
, , ,
(46)
The edge numbers and the associated nodes i
1
and i
2
are defined in Table 1 and in Fig.7. For
a detailed definition of the quantities involved in (46) one may consult [15, 17]

Edge i Node i
1
Node i
2
1 1 2
2 1 3
3 1 4
4 2 3
5 4 2
6 3 4
Table 1. Edge definition for a tetrahedral element

(a) (b)
Fig. 7. Finite elements employed for the discretization, (a) tetrahedral edge element for 3-D
and (b) triangular node element for 2-D.
Following an approach similar (but more complicated) to that for the 2-D case the resulting
volume element matrix reads:
vector-tetrahedral:
{ } { } { } { }
2
1
T
e e e e e e
i j o c i j
e
r V
K N N k N N dV c
(
(
= V V
(

(

}}}
, , , , , ,
(47)

Medical Imaging

172
The absorbing boundary conditions stem from (29a) and take the form of the corresponding
term in (34) through an asymptotic approximation. In order to evaluate this integral, the
closed fictitious surface (S) is discretized into triangles coinciding with the outer surface of
the peripheral tetrahedrals as shown if Fig.8. The resulting element matrix reads:
3-D ABCs:
( ) ( )

2
s o
i j
S
jk
K r N r N dS
(
(
=
(

}}
, ,
(48)
where
r is the outward unit normal vector. Note that the above involved triangular edge
elements shape functions are not trivial, since each triangle has a different orientation in a 3-
D space. The most convenient way to extract them is through the degeneration of the
corresponding peripheral tetrahedral into a surface triangle. Regarding the source modeling
an infinitesimal dipole approximation is again considered as for the 2-D case.

Fig. 8. Triangles coinciding with the outer surface of the peripheral tetrahedrals
The linear system solution in the 3-D case constitutes a major difficulty due to the large
number of unknowns. Thus the performance of direct inversion methods like LU
decomposition becomes questionable and iterative techniques are also tried herein.
Specifically, in the 3-D case the inversion of the matrix K will consume all the system
memory due to the size of the matrix, so an iterative solver is preferred with the appropriate
preconditioner. The Generalized Minimum Residual method (GMRES) is used for the
solution of the linear system with a symmetric successive over-relaxation-vector (SSOR-
vector) preconditioner.
Solving this system the electric fields (or the potential) on the receiving antennas (or
electrodes) and at all the internal edges or nodes is calculated and stored, to be used latter
within the reconstruction algorithm. An example of the electric field distribution when one
of the infinitesimal dipole or a line-source antenna is activated is shown in Fig. 9.


173
(a) (b)
Fig. 9. (a) Electric field distribution when a 2D structure is illuminated by an infinitesimal
dipole operated at 800MHz and (b) electric field distribution at two planes when a 3D
structure is illuminated by an infinitesimal dipole operated at 1 GHz.
4. Inverse problem
The reconstruction algorithm is based on the modified perturbation method that was
initially developed for the conductivity imaging in Electrical Impedance Tomography [26]
and later on extended to higher frequencies (up to 10MHz),[27]. This research was carried
out within the doctoral thesis of D. Drogoudis [28], which was primarily focused to extend
MPM and prove its validity in the microwave regime. Indeed this proof of concept was
achieved and published in [29] and [30]. For the introduction of the reader to the inverse
problem approach but also to provide him with the knowledge to proceed further and
ultimately achieve its practical implementation, the reasoning behind its implementation
and the logical steps for its extension will be given next. Thus, let us start by reviewing the
static MPM and proceed to its extension to the MHz and finally the microwave range.
4.1 Review of the static MPM algorithm
First a review of the static algorithm will be given and based on that the steps towards its
present time harmonic formulation will be described. According to [26] the -imaging
algorithm reads:

1 1 1
1
1
| |
M
mi c i
i
mi j i n n n
j j j
M
k
j k
V V
V
V
k
V
0
0o
o o o
0
0o
=
=
= +

(49)
where M is the total number of linearly independent measurements, V
mi
and V
ci
are the
measured and calculated voltage differences at the i
th
port (electrodes pair) and k
1
is the
relaxation factor to be chosen in the range 0<k
1
<2 in order to provide faster convergence.
The partial derivatives /
i j
V 0 0o or /
k j
V 0 0o constitute the elements of the Jacobian matrix.
These are calculated from the voltage distributions all over the object cross section through a

Medical Imaging

174
closed form expression given by Yorkey et al. [31]. The latter is based on the electrical circuit
compensation theorem as applied on a resistor network equivalent for each -element, as
shown in Fig.10. In turn these expressions are given in [31] as:

( ) ( )
1
1
L
i
ij ij kj
j i
V
J S V V
I
0
0o
=
= =
/ / /
/
(50)
where
( ) ij
V
/
and
( ) kj
V
/
are the voltages developed across the l
th
branch of the j
th
element
Fig.10 when a sinusoidal current with amplitude
i k
I I = is successively injected through the
ports -i and -k respectively. The constants S
/
are the normalized geometrical weights
arising from the finite element formulation. S
/
are actually given by the non-diagonal
entries of the | |
e
Y -element matrix as: | | | |
j e
e
Y K o = and | |
e
S =
/
non-diagonal entries of | |
e
K .
An indicative example of these branch admittance values is through the definition of the
resistance network for the rectangular element (Fig.10) and its related element matrix:
static triangular element

| | | |
11 12 13
22 23
3 3
33
j
G G G
Y G G K
symmetric G
o
(
(
= =
(
(

(50a)

| |
2 2
1 1 1 2 1 2 1 3 1 3
2 2
2 2 2 3 2 3
3
2 3
3 1
1
4
b c b b c c b b c c
K b c b b c c
symmetric b c
(
+ + +
(
= + + (
A
(
+
(

(50b)
static rectangular element

| |
11 12 13 14
22 23 24
4
33 34
44
G G G G
G G G
Y
G G
symetric G
(
(

(
=
(
(

(50c)

| |
2 2 2 2 2 2 2 2
2 2 2 2 2 2
4 2 2 2 2
2 2
2 2 2
2 2
2 2 2
2 2
a b a b a b a b
a b a b a b
K
a b a b
symetric a b
(
+
(
+ (
=
(
+
(
(
+

(50d)
and ( )
1 1
2 2 1 2 2 1
3 3
1
1 1
1
2 2
1
x y
x y b c b c
x y
A = =
1 2 3 2 3 1 2 3 1 3 2
2 3 1 3 1 2 3 1 2 1 3
3 1 2 1 2 3 1 2 3 2 1
, ,
, ,
, ,
a x y y x b y y c x x
= = =
= = =
= = =

A detailed derivation of the Compensation theorem can be found in Yorkey et al. [31] as
well as in Sahalos et al. [32](p.229). Guidelines for its implementation are also given therein.


175

(a) (b)

(c) (d)
Fig. 10. Equivalent resistor network for each j
th
element with locally homogenous
conductivity
.
: (a) Triangular element defined by its coordinates (x
i
,y
i
), (b) equivalent
network of (a) (c) Rectangular element of dimensions ab, (d) equivalent network of (c).
4.2 MPM extension to high frequencies
Working on the extension of the algorithm toward complex permittivity imaging, equations
(49) and (50) should be modified for complex voltages and complex admittances.
Analyzing the constituents of MPM algorithm: Let us clarify equation (49) starting from each
term in the numerator sum which runs over all i=1, to M voltage measurement ports. The i-
th term corresponds to the i-th port where the difference between measured
,
( )
m i
V and
calculated
,
( )
c i
V voltages
, ,
( )
m i c i
V V is normalized by the measured value and this term is
multiplied by the sensitivity of the i-th port with respect to the j-th pixel conductivity
j
o (being currently updated), namely / /
i j i j
V V o oo 0 0o = . As for the denominator sum in
(49) it is just a term normalizing the sensitivities to unity. It is exactly the same as saying that
that the infinite integral of a probability density function is equal to unity. From a different
point of view, each measurement port contribute to the total updating summation by its
normalized voltage deviation from the measured value weighted by the port's normalized
sensitivity. The same principle can be readily applied to the complex or two variables
( , )
e e
j rj
o c update (this could be even generalized to multiple variables) scheme, provided that
the corresponding sensitivities are available.

Medical Imaging

176
Extending MPM to complex quantities: Recall that when Laplace equation is solved only for a
conductivity distribution the calculated voltage will be real or in-phase with the injected
sinusoidal current (source), just as injecting a current in its equivalent network in Fig.10. In
contrary, when permittivity is included in the model a 90
phase shifted (quadrature)

component appears in the developed voltage as well. This is in turn equivalent to injecting a
sinusoidal current on a complex admittance network (Fig.11) and measuring the complex
voltage across any i
th
admittance, which reads:
( , ) ( , )
re im re im
i i i i r i r
V V jV V jV o c o c = + = + (51)
In view of the above a complex sensitivity
ij i cj
J V c = c c is required in order to establish an
algorithm similar to (49). For this purpose a complex derivative can be defined as [33-36]:

( ) ( )
i i i
ij
cj j j
V V V
J j
ec o ec
c c c
= = +
c c c
(52)
with:

cj j j
j j ec ec o = + or
0 0 crj rj
j j ec c ec c o = + or
cj j j
j ec ec o = + (53)
This complex derivative definition has already been employed by many researchers, e.g
Franchois et al. [33] and Rekanos et al. [34-36], for the definition of a complex gradient
(complex Jacobian matrix) working toward the establishment of a complex permittivity
microwave imaging. The definition of (52) is exploited herein in order to identify the
complex Jacobian matrix
ij
J into four real submatrices as presented by Polydorides [37]

( )
( )
re re
RR RI
ij ij j j
ij
IR II im im
ij ij
j j
V V
J J
J
J J V V
0 0
0o 0 ec
0 0
0o 0 ec
(
(
(
(
(
= =
(
(
(

(

(54)
An alternative approach, instead of using a complex derivative is to expand the problem
into two real functions ( , )
re
r
V o c and ( , )
im
r
V o c each one depended on two real unknown
variables ( , )
r
o c . In turn the basic MPM algorithm (49) can be employed four times for each
one of the sensitivities just as those occurring in (54). As explained above this is possible not
only for two but even for any arbitrary number of variables. Attention must be paid to the
normalized contributing terms (summations) similar to that of (49), since all terms referring
to the same variable ( , )
r
o c must be added together after being weighted by an appropriate
relaxation factor k. Hence, measurements at each i
th
port along with the solutions of the
complex generalized Laplace equation yields complex voltages which are separated into real
( )
re
i
V and imaginary parts ( )
im
i
V . These numerical solutions yield complex voltage
distributions at the FEM nodes all over the model of the body to be imaged. Before
proceeding to the complex MPM algorithm, the analytical expressions for the evaluation of
the above four sensitivities must be extracted.


177

(a) (b)

Fig. 11. Equivalent admittance network for complex permittivity elements, (a) Rectangular
and (b) its equivalent networks.
4.3 Complex MPM formulation for the MHz range
The complex derivative of equation (52) presents the important decoupling of its real and
imaginary parts, thus actually providing separate real and imaginary sensitivities.
Specifically its real part constitutes the sensitivity of the measured complex voltage with
respect to the conductivity
j
. Respectively, the imaginary part comprises the sensitivity
with respect to the j
th
element permittivity
rj
. Additionally each one of them can be
discriminated into two terms by substituting the complex
i
V from (51). Consequently, the
complex permittivity imaging is readily provided by applying the basic MPM algorithm (49)
for each one of the four sensitivity terms. Specifically, considering the conductivity
j
o update we may set-up a summation like that of (49) comprised of the contributions of the
real voltage part
re
i
V weighted by the sensitivities /
re
i j
V 0 0o . But, now the imaginary part of
the voltage
im
i
V weighted by its sensitivity /
im
i j
V 0 0o comprises another sum similar to
(49). The real and imaginary voltage contributions are summed together and multiplied by
the proper relaxation factor to form the formula for -imaging. Likewise, the real and
imaginary voltages are weighted with the corresponding sensitivities with respect to
permittivity / ( )
re
i j
V 0 0 ec and / ( )
im
i j
V 0 0 ec to formulate the -imaging algorithm. These
two could be given either separately or combined together in a complex
c
c imaging
algorithm as:

1 1 1 1
0 0 1 1 0 2 2
( ( ) )
n n n n n n n n
cj j r j rcj j j r j r j
j j j k W j k W e c o ec c ec c o o ec c c

( (
= + = = + + +

(55)

1 1
1
1 1
re re im im
re im M M
mi c i mi c i
i i
re im
j j i i mi mi
re im M M
k k
j j k k
V V V V
V V
V V
W
V V
0 0
0o 0o
0 0
0o 0o
= =
= =

= +

(56)

Medical Imaging

178

0 0 1 1
2
0 0 1 1
( ) ( )
( ) ( )
re re im im
re im M M
mi c i mi c i
i i
re im
rj rj i i mi mi
re im M M
k k
rj rj k k
V V V V
V V
V V
W
V V
0 0
0 ec c 0 ec c
0 0
0 ec c 0 ec c
= =
= =

= +

(57)
M is the total number of linearly independent measurements,
mi
V and
ci
V are the measured
and calculated voltage differences at the i
th
port (electrodes pair) and
1 2
, k k are the
relaxations factors that may provide faster convergence.
The optimum values of
1 2
, k k can be obtained through a numerical investigation. If these
are not known then is preferable to set
1 2
1 k k = = (no relaxation). Even though a relaxation
factor could in general speed up convergence, however in all performed numerical
experimentation the best performance was obtained without relaxation.
4.4 MPM for microwave imaging
Working toward the extension of MPM algorithm to microwave imaging we simply need to
substitute the complex voltage by the complex electric field. Indeed, the expressions of
equations (55)-(57) are readily applicable for the 2-D microwave imaging when the electric
field is comprised of only one component
z
E E z =
,
or it is a complex scalar function.
However, the proposed procedure is general and applies for the arbitrary 3-D imaging
where the electric field is a complex vector quantity. Restricting ourselves to the 2-D case we
may repeat the definition of the complex derivative just for clarity as:

i i i
ij
real imag
cj cj cj
E E E
J j
c c c
c c c
= = +
c c c
(58)
In turn, by following exactly the same rationale as for the MHz imaging the microwave
reconstruction algorithm reads:

( 1) ( 1) ( 1) ( 1)
1 1 2 2
imag n imag n real n real n n
ck ck ck ck ck
k W j k W c c c c c

( (
= + + +

(59)

1 1
1
1 1
imag imag imag M real real real M
mi ci i mi ci i
imag real real real
i ck i mi ck mi
real imag M M
k k
real real
i ck i ck
E E E E E E
E E
W
E E
c c
c c
= =
= =
c c
c c
= +
c c
c c

(60a)

1 1
2
1 1
imag imag imag real real real M M
mi ci i mi ci i
real imag imag imag
i i mi mi ck ck
imag real M M
k k
imag imag
i i
ck ck
E E E E E E
E E
W
E E
c c
c c
= =
= =
c c
c c
= +
c c
c c

(60b)


179
where M is the total number of linear independent measurements, E
mi
and E
ci
are the
measured and calculated fields at the i
th
antenna and k
1,2
are the relaxations factors that may
provide faster convergence. The optimum values of k
1
, k
2
can be obtained through a
numerical investigation. Observing the MPM reconstruction equations e.g. (59) to (60) or
even the original static MPM in (49) one may discuss on the expected robustness or the
immunity of the method to the problem of ill-posedness. As it is well understood the ill
posedness is due to very large variations (of the order of 10
6
) of the sensitivity. For example
very large sensitivity values ( )
i ck
E c c c may be observed for the voxels-k around the
transmitting antenna and along the axis of transmit-receive antenna pairs. But the sensitivity
away from these object regions become very small and may be even lower than the
measurement or calculations inaccuracies. When exact inverse problem formulations are
employed these large variations in sensitivity yields an ill posed Hessian matrix
accompanied by a lot of difficulties during its required inversion. Actually, the very large
variation in sensitivity corresponds to very large variation in the Hessian matrix eigenvalues
which reflects to a high singularity degree. In contrary, these very low sensitivities occur as
very small weighting factors in the summations involved in MPM, e.g. (60). Thus, the very
small sensitivity has negligible contribution in the complex permittivity update, which from
a different point of view is shadowed by the higher sensitivities. This is equivalent to
discarding very small singular values of the Jacobian (or Hessian), or using relatively large
regularization parameter. Hence, the resulting reconstruction algorithm is expected to be
robust during the first few iterations and it was indeed proved to converge always in our
previous works, e.g. [26-30, 38]. The observed monotonous convergence was independent of
the electrodes or antenna locations which greatly affects the sensitivity distribution. This is
in turn a clear evidence of the MPM immunity against the problem of ill-posedness. Based
on these observations MPM ensures convergence without the sophistication of
regularization techniques. However, the penalty paid by MPM for its robustness is the
absence of any control of the implicitly involved regularization. Instead, an exact method
may involve a controllable regularization parameter which is gradually reduced from one
iteration to the next. The reduction of the regularization parameter allows for the small
eigenvalues or low sensitivities to be exploited.
According to Meaney et al. [39], most regularization methods require the use of a varying
degree of a priori information in order to ensure convergence. In contrary MPM ensures
convergence without any a priory information, but its accuracy is compromised by its
inability to exploit the hidden low sensitivities. Thus, a logical hypothesis toward accuracy
improvement is to formulate an exact inverse problem (e.g., a Gauss-Newton scheme) and
exploit the final image provided by MPM as an initial guess (a priori information). The
regularization parameter within this scheme can be gradually set to zero or to a very small
value, uncovering the very low sensitivities (exploiting the very small eigenvalues) which
are expected to fine tune the image within a few iterations. The proof of this hypothesis
constitutes one of our next tasks.
Observing the above MPM reconstruction expression it is obvious that the Jacobian-
sensitivity matrix constitutes the fundamental basis of the proposed method. Its accurate
and computationally efficient calculation is of primary importance in ensuring successful
imaging. For this purpose the next section elaborates on an Adjoint network for the
microwave imaging and its counterpart network compensation theorem for the MHz range,
both resulting in closed form expressions for the Jacobian matrix entries.

Medical Imaging

180
5. Jacobian or sensitivity matrix
As explained in the introductory section the current paths following through the object
depend on the unknown and
r
profiles. Also the current density (or field intensity) is very
high around the active electrodes (or antennas). These two phenomena make the - and
r
-
imaging a strongly non-linear inverse problem by creating zones with very high and others
with very low sensitivity. This severe singularity presents ratios of maximum to minimum
singular values
max
/
min
as high as 10
6
. Mathematically this reflects to a strongly non-linear
Jacobian function J
ij
(,
r
). The present method handles the non-linearity, since it calculates
the Jacobian matrix at each n-th iteration, namely for each (,
r
) distribution. The additional
computation cost of this task is negligible, since the Jacobian is calculated from closed form
expressions from already available voltage or electric field distribution data.
Let us proceed to the presentation of the methodologies resulting the closed form
expressions for the Jacobian matrix.
5.1 Compensation theorem on an admittance network
The sensitivities of the MHz range imaging can be obtained by extending the resistive
network compensation theorem (50) originally given by Yorkey et al. [31] and later on
reviewed in detail by Sahalos et al. [32](p 229). This is equivalent to the application of the
circuits compensation theorem on an admittance network. For this purpose an equivalent
admittance network is considered for each triangular or rectangular element
rj
( , )
j
o c as
shown in Fig.11. The admittance values are obtained from the non-diagonal entries of the
corresponding element matrix as:
| | | | ( )
j j
e e
Y j K o ec = + where as defined
| |
e
S =
/
non-
diagonal entries of | |
e
K .
For a triangular element (Fig.11a-b)

| | | |
11 12 13
22 23
3 3
33
( )
j j
Y Y Y
Y Y Y j K
symmetric Y
o ec
(
(
= = +
(
(

(61)
For a rectangular element (Fig.11c-d):
| | | |
11 12 13 14
22 23 24
4 4
33 34
44
( )
j j
Y Y Y Y
Y Y Y
Y j K
Y Y
symmetric Y
o ec
(
(

(
= = +
(
(

(62)
Where | |
3
K and | |
4
K are defined in (50) and
ij ij ij
Y G j C e = +
According to the network compensation theorem, a differential change of the l-th branch
admittance Y
l
by Y
l
causes a differential variation V
i
at the i-th port voltage as:

1

i ij kj
i
V Y V V
I
A = A
/ / / /
(63)


181
Assuming that the l-th branch belongs to the equivalent network of the j-th element, then:
( )
j j j
Y Y S j o e c A A = A + A
/ / /
(64)
Since the voltage is linearly dependent on the network admittances then the superposition
principle can be applied in two-fold. Summing over all element branches (Fig.11) one may
obtain the total variation of the i-th port voltage V
i
. Additionally, in order to evaluate the
four Jacobian submatrices, the above approach can be applied once when only the
conductivity of the jth element is varied ( )
j j
Y S o A = A
/ /
and again when only the
permittivity is varied (
j j
Y S je c A = A
/ /
). Applying (62) for each case and separating into real
and imaginary parts one may end up to the desired Jacobian submatrices as:

1
1
re L
RR i
ij Rl
j i
V
J S V
I
0
0o
=
= =
/
/
(65a)

1
1
im L
IR i
ij Il
j i
V
J S V
I
0
0o
=
= =
/
/
(65b)

1
1
( )
re L
RI i
ij Il
j i
V
J S V
I
0
0 ec
=
= = +
/
/
(65c)

1
1
( )
im L
II i
ij Rl
j i
V
J S V
I
0
0 ec
=
= =
/
/
(65d)
where

re re im im
Rl ijl kjl ijl kjl
V V V V V =
(66a)

re im im re
Il ijl kjl ijl kjl
V V V V V = +
(66b)
As a point of interest showing the potential of extending this method to microwave imaging,
it is proved that the employed Network Compensation Theorem constitutes a particular
case of the more general ``Adjoint Network Theorem'', which is in turn a direct consequence
of the electromagnetism ``Reciprocity Theorem''.
5.2 Sensitivity equation based on Adjoint Network Theorem
Regarding the microwave sensitivity a more general approach is employed based on the
electromagnetic field of a particular illuminating antenna combined with an adjoint
configuration field (herein that for a different transmitting antenna location) through the
Reciprocity Theorem. Explicitly the components of the Jacobian are the partial derivatives
(or the sensitivities) of the electric field
r
E
,
measured at the r-th antenna to the complex
permittivity
e
c
c of the e-th element-pixel, when the s-th antenna is activated. As explained
below this is in turn evaluated through closed form expressions resulting from the

Medical Imaging

182
reciprocity theorem and the employment of an adjoint problem. For this purpose an
approach similar to that given by Oldenburg [40] and the original research cited therein is
adopted. Namely, the two Maxwell Curl equations are written for the source
( )
s
J
,
at s-th
antenna and the involved fields
( )
, E H
, ,
are differentiated with respect to the e-th complex
permittivity. For the adjoint fields
( )
,
a a
E H
, ,
these two curl equations are written considering
a source
( )
r
J
,
at the r-th antenna (Fig.12). The four curl equations are in turn combined
following the reciprocity theorem procedure. Specifically, for a dipole antenna at the s-th
location the curl equations reads:

E j H e V =
, , ,
(67a)

c s
H j E J ec V = +
, , , ,
(67b)
Taking the derivatives of (67a) and (67b) with respect to
ck
c defined in Eq. (36):

ck ck
E H
je
c c
c c
V =
c c
, ,
,
(68a)
( )
c k
ck ck
H E
j j r E ec e
c c
c c
V = +
c c
, ,
, ,
,
(68b)
Consider an auxiliary (adjoint) Maxwell problem with a dipole source
( )
r
J
,
located at the
observation point
r
r r =
, ,
as shown in Fig.12.

a a
E j H e V =
, , ,
(69a)

a a
c r
H j E J ec V = +
, , , ,
(69b)
For the application of the Reciprocity theorem procedure to (68a), (68b) and (69a), (69b)
according to Balanis [24](p. 324), lets take the inner product of
a
H
,
(68a) and
ck
E
c
c

c
,
(69b)
and subtract:

a a a a
c r
ck ck ck ck ck
E E H E E
H H j H j E J e ec
c c c c c
c c c c c
V V =
c c c c c
, , , , ,
, , , , , , ,
(70)
Making use of the identity:

( ) ( ) ( )
A B B A A B V = V V
, , , , , , , , ,
(71)
For
a
B H =
, ,
and
ck
E
A
c
c
=
c
,
,
the left hand side of (70) can be written as
a
ck
E
H
c
| |
c
V
|
c
\ .
,
, ,
.
In turn, one may take a volume integral of (70) over a domain enclosed by a sphere with a
radius tending to infinity and apply the divergence theorem on the left side.


183

( )
V S
F du F dS V =
}}} }}
, , , ,
(72)
to get:

a a a
c r
ck ck ck ck
S V
E H E E
H dS j H j E J dV e ec
c c c c
| | | |
c c c c
=
| |
c c c c
\ . \ .
}} }}}
, , , ,
, , , , ,
(73)
Likewise, dot-multiplying
a
E
,
(68b) and subtracting
( )
ck
H c c c
,
(69a) yields:
( )
a a a a a a
c k
ck ck ck ck ck
H H H E H
E E E j E j r E E j H ec e e
c c c c c
| |
c c c c c
V = V V = + +
|
c c c c c
\ .
, , , , ,
, , , , , , , , , ,
,
(74)
Adding (74) and (70) the similar terms are canceled, then the use of vector identities
A B B A =
, , , ,
and A B B A =
, , , ,
results to:
( )
a a a a a
r k
ck ck ck ck ck
H E E H E
E H H E J j r E E e
c c c c c
| | | | | |
c c c c c
V + V = V = +
| | |
c c c c c
\ . \ . \ .
, , , , ,
, , , , , , , , , ,
, ,
(75)
Taking the volume integral over a domain enclosed within a sphere of an infinite radius and
applying the divergence theorem ends up to:
( )
a a a
r k
ck ck ck
S V
E H E
H E dS J j r E E dV e
c c c
| | | |
c c c
= +
| |
c c c
\ . \ .
}} }}}
, , ,
, , , , , ,
, ,
(76)
The left hand side is identically zero, since both fields intensities tend to zero at infinity by
simply considering artificially small unbounded medium losses. Hence, (76) finally gives the
so called sensitivity equation as:
( )
a
r k
ck
V V
E
J dV j r E EdV e
c
c
=
c
}}} }}}
,
, , ,
, ,
(77)
The sensitivity equation (77) can be further simplified to yield a closed form for the Jacobian
matrix entries
( )
ck
E c c c
,
by first introducing the FEM basis functions and by considering the
specific receiving- sensing antennas. Starting from the integrals of (77), these are in general
over the whole solution domain. But, actually the left hand side is restricted over the current
carrying volume (V
r
) of the r-th antenna. In turn, considering the definition (36) for the
complex dielectric expansion the right hand side integral of (77) is obviously restricted over
the volume (V
k
) of the k-th element. Further a closer look at the dot product of (77) left hand
side reveals that the involved electric field is that produced by a radiating antenna at the
s
r r =
, ,
position (Fig. 12) illuminating the receiving antenna at
r
r r =
, ,
position (sensing port)
with only the field component parallel to its current
r
J
,
producing a net effect. Hence, the
derivative
( )
ck
E c c c
,
can be identified as the sensitivity of the r-th receiving antenna (r-th
sensing port) with respect to the k-th element complex permittivity ( )
ck
c when the s-th

Medical Imaging

184
antenna illuminates the structure, or specifically the (s, r, k) entry of the Jacobian matrix,
( )
, , s r k ck
J E c = c c
,
.

Fig. 12. Geometry for the application of the reciprocity theorem
Since the integration is restricted over the k-th element volume (V
k
), then the fields E
,
and
a
E
,
can be expanded into the FEM basis (shape) functions according to (45). In view of the
above description (77) can be rewritten in the form:

k k k
a k k ak k
r r k i i j j k
ck i j
V V V
E
J dV j E EdV j E N E N dV e e
c
c
= =
c

}}} }}} }}}
,
, , ,
(78)
Working toward a closed form expression for the Jacobian matrix, the next step is to
consider specific antenna types. For the two-dimensional (2-D) case when the complex
permittivity is assumed homogeneous in the z-direction, along which the structure is
assumed infinite, it is convenient to employ infinitely extending thin line sources as
illuminating antennas with current density defined by:
Line source: ( )
r r
J I r r z o =
,
, ,
where I is a constant current (79)
In this case the sensitivity entries are readily simplified as:
2-D:
( ) ( )
( )
, ,
k
k k ak k
i i j j k
s r k
ck i j
S
E r j
J E N E N dS
I
e
c
c
= =
c

}}
,
,
(80)
For a three dimensional structure (3-D) illuminated by thin elementary dipoles of length (l)
the current density reads:
z-oriented elementary dipole:
( )
'
r
J I z o =
,
(81)


185
where
2 2
x y = + . Similarly, the sensitivity equation becomes:
3-D:
( ) ( )
( )
, ,
k
k k ak k
i i j j k
s r k
ck i j
V
E r j
J E N E N dV
I l
e
c
| |
| | c
| = =
|
|
c A
\ .
\ .

}}}
,
,
(82)
The above z-oriented current excitations yields a primarily z-polarized electric field which is
expected to interact mainly with the
zz
component of a possibly anisotropic complex
permittivity. In general, the orientation of the radiating dipoles could be exploited as an
additional degree of freedom to extract information from dielectrically anisotropic
structures, e.g. [41, 42].
The above sensitivity expressions can be written in matrix form as:

( ) ( ) , ,
k k ak
s r k
J E F E
( ( (
=

(83)
where =j/I or j/Il is the constant term for the line source and elementary dipole
excitations and
{ } { }
k
k k k
ij i j k
V
F N N dV
(
=

}}}
(84)
Matrices (vectors) [E
k
], [E
ak
] represent the tangential electric fields along the edges of the k-th
tetrahedral element in three dimensional case or the E
z
electric field on the nodes of the k-th
triangular element for the two dimensional case. These fields are already computed during
the multiple forward problem solutions performed during the setup of the calculated data-
set (once for each illuminating antenna). The matrix
k
ij
F
(

is a 6 6 matrix in 3D or a 3 3
matrix in 2D and its entries can be constructed from the FEM element matrices, [15].Recall at
this point that FEM is applied on a fine mesh of tetrahedral or triangular elements while the
image reconstruction is carried out on a coarse rectangular (pixels) or cubical (voxels) mesh.
Each reconstruction element consists of a number of forward elements and similarly a
number of nodes or edges that belong to these forward elements. Equation (83) results to a
partial Jacobian matrix for each triangular or tetrahedral forward element. However, the
desired Jacobian is that of the rectangular-pixel or cubical-voxel reconstruction element. For
each evaluation the Fk matrices are assembled together according to a classical FEM
procedure to yield a mm matrix M
e
, where e the global number of the reconstruction
element and m the number of edges or nodes that are inside the e-th element. Note that this
matrix depends only on the geometry (independent of
rc
distribution) and its calculated
only once and stored for multiple usage. Consequently, the Jacobian matrix of the e-th
element reads:

( ) ( ) , ,
e e ae
s r e
J E M E
( ( (
=

(85)
where
1 2
, ,...,
e e e e
m
E E E E
( (
=

and
1 2
, ,...,
ae ae ae ae
m
E E E E
( (
=

.

Medical Imaging

186
5.3 Sensitivities and features of the Jacobian matrix
The Jacobian matrix is a rectangular one with dimensions M>N where M the number of
linearly independent measuring points and N the number of unknown (
i
,
ri
) pairs. The
study of tis characteristics is quite important especially in the course toward inventing the
best data collection strategy, i.e. active and sensing antenna or electrode pairs locations and
their subsequent activation. Additionally, the sensitivity matrix study with the aid of the
recently revisited Proper Orthogonal Decomposition [43] techniques constitutes an
interesting research challenge. Even though our group is already working on this subject it
will not be considered herein as it requires a separate chapter of its own. For some
preliminary results regarding the SVD analysis of the Jacobian matrix please contact our
previous work [30] and wait for al follow up publication.
6. Numerical results & discussions
The proposed algorithm was applied for the imaging of numerous conductivity and
permittivity distributions for both circular and rectangular models and satisfactory
reconstructions are obtained. The background and anomaly conductivity and permittivity
values resemble these of typical human tissues.
Before proceeding to the presentation of reconstructed images the convergence criteria
should be first defined.
6.1 Inverse problem convergence criteria
Computer Test Approach: The so called computer test was employed in all cases
throughout this chapter. First a target model is considered and the forward problem is
solved for each illumination position. Namely, the first antenna is activated and the
forward problem is solved to calculate and store the electric field at all the remaining-
receiving antennas. Each one antenna is activated in turn and the received electric fields
are stored to form a complete dataset labeled as measurements. The reconstruction
algorithm starts from a homogeneous model and the desired complex permittivity profile
is sought.
Convergence Criteria: Two convergence criteria are adopted. The more general concerns the
available information, which is determined by the difference between fields measured
on the target model (E
m
or V
m
) and fields calculated at the n-th iterative solution of the
forward problem. As in every minimization approach, the sum of squares (SSQ) is the
appropriate figure taking signs in to account. Hence the summation over all measurements
ports (M) gives:
( )
2
1
i i
M
meas calc
i
SSQ V V
=
=
(86)
or ( )
2
1
i i
M
meas calc
i
SSQ E E
=
=
(87)


187
For comparison purposes it is more convenient to present its normalized value (SSQ
N
) with
respect to its maximum (SSQ
max
) occurring at the first iteration of the inverse problem as:

max
N
SSQ
SSQ
SSQ
= (88)
While SSQ is general and can be calculated in all cases, it is only an indirect indication of
convergence. Namely, its minimization ensure
c
c convergence when a unique solution is
safeguarded. But, this condition may be disturbed by the problem singularity degree, which
in turn depends on the data collection strategy. For this purpose the well- posedeness or the
singularity degree, of the sensitivity-Jacobian matrix should be preliminary examined.
However a further elaboration is required. Besides this for the particular case of the
computer test the target or true (,
r
) distributions are available. Hence, the estimated
normalized deviations of
( )
,
n n
calc calc
o c from their true values at the n-th iteration can be
defined as a norm of relative error:

( ) ( )
2 2
1 1
P P
i i i
true calc true true
i i
error o o o o o
= =
| |
=
|
\ .

(89)

( ) ( )
2 2
1 1
P P
i i i
r r true r calc r true r true
i i
error c c c c c

= =
| |
=
|
\ .

(90)
where
true
o and
r true
c

are the average values of the target profiles and P is the number of
elements of the reconstruction mesh. We should keep in mind that the -error,
r
-error
criteria are not applicable for practical in-vivo or even for the laboratory test cases where
the objects ( ) ,
r
o c distributions are unknown.
6.2 Quasi-static MHz reconstruction
The algorithm was applied for the circular cross-section of Fig.13a and for the rectangular
cross-section of Fig.13b. A total number of 32 electrodes were used, where only two of them
are active in each projection angle. An indicative example with a double anomaly is presented
herein. The target - and
r
-profiles are shown in Fig.14a. The values for conductivity are
1
21 / mS cm o = and
2
14 / mS cm o = and for permittivity
r1
300 c = and
r2
150 c = , in a
homogeneous background of 7 /
o
mS cm o = and
ro
100 c = . The frequency of the injected
current was 8 f MHz = . The relaxation factors
1 2
, k k are set equal to unity. The reconstructed
image after 15 iterations are shown in Fig.14b
The algorithm was also tested for the rectangular model of Fig.6. An anomaly with
conductivity
1
20 / mS cm o = and permittivity
r1
300 c = was introduced in a homogeneous
background of 7 /
o
mS cm o = and
ro
150 c = . The frequency of the injected current is
now 9 f MHz = . The relaxation factors are set equal to unity. The target model and the
reconstructed image after 15 iterations are shown in Fig. 15.
For details on the algorithm convergence rate and its performance, please wait for a follow
up publication which is now in preparation.

Medical Imaging

188

(a) (b)

Fig. 13. The FEM meshes: (a) fine forward mesh and (b) coarse reconstruction mesh.

(a) (b)

Fig. 14. Computer phantom with two anomalies in both
r
and . (a) The target model and
(b) the reconstructed - and
r
-profiles after 15 iterations.


189
2
4
6
8
2
4
6
8
150
200
250
300

cm
Electrical Permmitivity (
r
)
cm
100
120
140
160
180
200
220
240
260
280
300
2
4
6
8
2
4
6
8
6
8
10
12
14
16
18
20

cm
Conductivity ()
cm
m
S
/
c
m
4
6
8
10
12
14
16
18
20
2
4
6
8
2
4
6
8
140
160
180
200
220
240
260
280
300

cm
Electrical Permmitivity (
r
)
cm
2
4
6
8
2
4
6
8
6
8
10
12
14
16
18
20

cm
Conductivity ()
cm
m
S
/
c
m
100
120
140
160
180
200
220
240
260
280
300
4
6
8
10
12
14
16
18
20

(a) (b)
Fig. 15. (a) The target model and (b) the reconstructed and
r
-profiles after 15 iterations for
the rectangular model.
6.3 2-D microwave reconstruction
The target model simulated as a computer phantom is presented in Fig.16. A total number
of 32 antennas (line sources) were used, where only 28 are exploited as receivers for each
projection angle. A single offset anomaly with conductivity
1
0.15 S m o = and permittivity
1
15
r
c = was introduced in a homogeneous background of 0.3
o
S m o = and 30
ro
c = . The
frequency of operation was assumed at 1.1 f GHz = . The image reconstructed after 9
iterations is presented in Fig.17 for the conductivity and the permittivity profile
respectively. The correct location of the anomaly as well as its and
r
peak values are
obtained, but some artifacts are caused.

(a) (b)

Fig. 16. Target model (a) conductivity profile, (b) permittivity profile.

Medical Imaging

190

(a) (b)
Fig. 17. Reconstructed profiles for the example of Fig.16, (a) conductivity and (b)
permittivity distributions after 9 iterations. The object is discretized into 100 pixels and is
illuminated by 32 line sources at a frequency of 1.1 GHz.
6.4 3-D microwave reconstruction
The target model comprised of 4 layers as showin in Fig.18. A total number of 48 antennas
arranged in three rings of 16 antennas (elementary dipoles) were used, where only 39 are
exploited as receivers for each projection angle. A single offset anomaly with conductivity
1
= 0,15 S/m and permittivity
r1
= 15 was introduced at z = -0.5 cm (only at the second layer)
in a homogeneous background of = 0.6 S/m and
r
= 60. The frequency of operation was
assumed at f = 1.4GHz. The image reconstructed after 9 iterations is presented in Fig.19 for
the conductivity and the permittivity profiles respectively.

(a) (b)


191

(c) (d)
Fig. 18. Target model comprised of 4 layers to test the 3-D microwave imaging.

(a) (b)

(c) (d)
Fig. 19. Reconstructed profiles for the 3-D model of Fig.18 (a) 2nd layer, (b) 3rd layer
conductivity, (c) 2nd layer, (d) 3rd layer permittivity distributions, after 9 iterations. The object
is discretized into 400 voxels and is illuminated by 48 dipole sources at a frequency of 1.4 GHz.

Medical Imaging

192
For details on the 3-D algorithm performance and its convergence characteristics please
contact our work [30].
7. Conclusions
A modified perturbation method reconstruction algorithm for MHz and microwave
tomography is successfully established at a computer test level. The key constituent of this
algorithm is a close form evaluation of the sensitivity or Jacobian matrix based on an adjoint
network and reciprocity theorem approach. For the MHz frequency range th Jacobian is based
on an extension of the electrical networks compensation theorem but it can be obtained as a
particular case fo the general Adjoint network approach. The established algorithm is proved
robust and able to withstand a large amount of inverse problem ill-posedeness. Despite its
simplicity this algorithm is able to successfully localize the target anomalies reaching
conductivity and permittivity patterns very close to the expected global minimum at about the
6th iteration. The penalty paid for this simplicity and robustness is a compromise in the finally
achieved solution mostly appearing as artefacts around the target anomalies. A further
improvement, which also constitutes our next task, refers to the formulation of an exact
inverse problem, which may start from the finally obtained image herein and iteratively fine
tune it. This can be readily formulated exploiting the exact Jacobian matrix established herein.
8. References
[1] G. V. Keller, "Electrical properties of rocks and minerals", Handbook of Physical Constants
(S.P. Clark, ed.),N.Y., Geological Society of America, pp. 553-770, 1988
[2] R. Pethig, "Dielectric properties of biological materials: Biophysical and medical
applications", IEEE Transactions on Electrical Insulation, vol. EI, pp. 453-474, Oct.1984
[3] C. Gabriel, S. Gabriel, and E. Corthout, "The dielectric properties of biological tissues:
Part I, II and III", Phys. Med. Biol., vol. 41, 1996
[4] Q. Fang, P. M. Meaney, S. D. Geimer, A. V. Streltsov, and K. D. Paulsen, "Microwave
image reconstruction from 3-D fields coupled to 2-D parameter estimation", IEEE
Trans. Medical Imaging, vol. 23, pp. 475-484, 2004
[5] P. M. Meaney, K. D. Paulsen, A. Hartov, and R. K. Crane, "Microwave imaging for tissue
assessment: initial evaluation in multi-target tissue-equivalent phantoms", IEEE
Trans. Biomed. Eng., vol. 43, pp. 878-890, 1996
[6] S. H. Zainud-Deen, W. M. Hassen, E. E. D. ali, and K. H. Awadalla, "Breast cancer detection
using a hybrid finite difference frequency domain and particle swarm optimization
techniques", Progress in Electromagnetics Research B, vol. 3, pp. 35-46, 2008
[7] J. Goble and D. Isaacson, "Optial current patterns for three-dimensional computed
tomography," in Proc. Int. conf. IEEE Trans. Eng. in medicine & biology soc., 1989, pp.
463-464,
[8] W. R. B. Lionheart, J. Kaipio, and C. N. McLeond, "Generalized optimal current patterns
and electrical safety in EIT", IOP publ, Physiological Meas., vol. 22, pp. 85-90, 2001
[9] D. C. Barber and A. D. Seagar, "The Sheffiled data collection system", Clin. Phys. &
Physiol. Meas. , vol. 8, pp. 91-97, 1987
[10] R. E. Collin, Filed Theory of Guided Waves, 2nd ed.: IEEE Press, 1991.
[11] A. V. Korzhnenevskii, V. N. Kornienko, M. Y. Kultiasov, Y. S.Kultiasov, and V. A.
Cherepenin, "Electrical Impedance Computerized Tomography for Medical
Applications", Instruments for Experimental Tech., vol. 40, pp. 415-421, 1997


193
[12] B. H. Brown, "Tissue impedance methods," in Imaging with non-ionizing radiations, Ann.
of the NYAS, 1983,
[13] C. S. Lavranos and G. A. Kyriacou, "Eigenvalue analysis of curved waveguides
employing an orthogonal curvilinear frequency domain finite difference method",
IEEE Microwave Theory and Techniques,, vol. 57, pp. 594-611, March 2009,
http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=04783093
[14] S. Lavdas, C. Lavranos, and G. A. Kyriacou, "A finite difference frequency domain
method for the eigenanalysis of anisotropically loaded curved periodic structures,"
in 32nd Antenna Workshop on Antennas for Space Applications, ESA/ESTEC,
Noordwijk, 2010,
[15] J. Jin, The finite element method in electromagnetics: John Wiley & Sons, 1993.
[16] J. L. Volakis, A. Chatterjee, and L. C. Kempel, Finite Element Method for Electromagnetics:
IEEE Press, 1998.
[17] Y. Zhu and A. C. Cangellaris, Multigrid Finite Element Methods for Electromagnetic Field
Modeling: IEEE Press, 2006.
[18] L. R. Price, "Electrical Impedance Computed Tomography (ICT): A New CT Imaging
Technique", IEEE Transactions on Nuclear Science, vol. 26, pp. 2736-2739, 1979
[19] J. D. Jackson, Classical Electrodynamics, 3rd ed.: Wiley, 1999.
[20] C. Zekios, P. Allilomes, and G. Kyriacou, "Eigenfunction expansion for the analysis of
closed cavities," presented at the Loughborough Antennas and Propag. Conf., 2010,
http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=5666915
[21] S. G. Diamantis, A. P. Orfanidis, G. A. Kyriacou, and J. N. Sahalos, "Horn Antennas
Analysis Using a Hybrid Mode Matching-Auxiliary Sources Technique " in Proc. of
PIERS PISA March 2004, pp. 457-460,
[22] P. C. Allilomes and G. A. Kyriacou, "A Nonlinear Finite-Element Leaky-Waveguide
Solver", IEEE Transactions on Microwave Theory and Techniques, vol. 55, pp. 1496-
1510, July 2007,
http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=4268422&tag=1
[23] P. Allilomes, "Electromagnetic Simulation of open-radiating structures using the Finite
Element method," Ph.D. Thesis, Dept. of El. & Comp. Eng, Democtirus University
of Thrace, Xanthi, 2007,
[24] C. A. Balanis, Advanced Engineering Electromagnetics: John Wiley & Sons, 1989.
[25] Q. Fang, "Computational methods for microwave medical imaging," Ph.D. Thesis,
Dartmouth College Hanover, New Hampshire, 2004,
[26] G. Kyriacou, C. Koukourlis, and J. Sahalos, "A reconstruction algorithm of electrical
impedance tomography with optimal configuration of the driven electrodes", IEEE
Trans. Medical Imaging, vol. 12, pp. 430-438, Sept. 1993
[27] D. Drogoudis, G. Trichopoulos, and G. A. Kyriacou, "A modified pertubation method
for three-dimensional time harmonic impedance tomography", PIERS Online, vol.
1, pp. 151-155, 2005
[28] D. Drogoudis, "The inverse electromagnetic problem in high frequency tomography,"
Ph.D. Thesis, Democritus University of Thrace, Dec. 2009 (in Greek),
http://thesis.ekt.gr/thesisBookReader/id/18153#page/4/mode/2up
[29] D. Drogoudis, G. A. Kyriacou, and J. N. Sahalos, "A three dimensional microwave
tomography exploiting an adjoint network theorem based sensitivity matrix", IEEE
Trans. Magnetics, vol. 45, pp. 1686-1689, 2009,

Medical Imaging

194
http://piers.org/piersonline/pdf/Vol3No8Page1217to1221.pdf
[30] D. G. Drogoudis and G. A. Kyriacou, "Microwave Tomography Employing an Adjoint
Network Based Sensistivity Matirx," presented at the Progress in Electromagnetic
Research, 2009, http://www.jpier.org/PIER/pier94/13.09060808.pdf
[31] T. J. Yorkey, J. G. Webster, and W. J. Tompkins, "Using the GRID to improve the
computation speed of electrical impedance tomography (EIT) reconstruction
algorithms", Transactions on Biomedical Engineering, vol. 26, pp. S209-S215, 1987a
[32] J. N. Sahalos, G. A. Kyriacou, and C. Koukourlis, "Techniques in Electrical Impedance
Tomography (EIT)," ed, 1997.
[33] A.Franchois and A. G. Tijhuis, "A quasi-Newton reconstruction algorithm for a complex
microwave imaging scanner environment", Radio Science, vol. 38, pp. 8011-8016, 2003
[34] I. T. Rekanos, M. Efraimidon, T. Yioultsis, C. Antonopoulos, T. Tsiboukis, and E.
Kriezis, "Antennas, Propagation and EM Theory, ISAPE 2000," in 5th International
Symposium on 2000, pp. 187-190,
[35] I. T. Rekanos, M. Efraimidon, and T. Tsiboukis, "Microwave imaging: Inversion of
scattered near-field measurements", IEEE Trans. Magnetics, vol. 37, pp. 3294-3297,
2001
[36] I. T. Rekanos and T. Tsiboukis, "A finite element based technique for microwave
imaging of two-dimensional objects", IEEE Transactions on Instrumentation and
Measurement, vol. 49, pp. 234-239, 2000
[37] N. Polydorides, "Image reconstruction algorithms for soft field Tomography," Ph.D.
thesis, University of Manchester, 2002,
[38] D. Drogoudis, G. A. Kyriacou, and J. N. Sahalos, "A sensitivity matrix based microwave
tomography exploiting an adjoint network theorem", PIERS Online, vol. 3, pp. 1217-
1221, 2007
[39] P. M. Meaney, K. D. Paulsen, B. W. Progue, and M. I. Miga, "Microwave image
reconstruction utilizing log-magnitude and unwrapped phase to improve high-
contrast object recovery", IEEE Trans. Medical Imaging, vol. 20, pp. 104-116, 2001
[40] D. Oldenburg, "Practical strategies for the solution of large-scale electromagnetic
inverse problems", Radio Science, vol. 29, pp. 1081-1099, 1994
[41] J. J. Shyu, C. H. Chan, M. W. Hsiung, P. N. Yang, H. W. Chen, and W. C. Kuo,
"Diagnosis of articular cartilage damage by polarizations sensitive optical
coherence tomography and extracted optical properties", Progress in
Electromagnetics Research, vol. PIER 91, pp. 365-376, 2009
[42] P. Mauriello and D. Patella, "Resistivity tensor propability tomography", Progress in
Electromagnetics Research B, vol. 8, 2008
[43] M. Scherg, "Functional imaging and localization of electromagnetic brain activity", Brain
Tomography, vol. 5, pp. 103-111, 1992
9
A Mutual Information-Based Image Quality
Metric for Medical Imaging Systems
Du-Yih Tsai, Eri Matsuyama and Yongbum Lee
Niigata University
Japan
1. Introduction
Information on physical image quality of medical images is important for imaging system
assessment in order to promote and stimulate the development of state-of-the-art imaging
systems. In this chapter, we present a method for quantifying overall image quality of
digital imaging systems using mutual information (MI) metric. The MI which is a concept
from information theory is used as a measure to express the amount of information that an
output image contains about an input object. The MI value is considered that it can be used
to express combined physical properties of image noise, resolution and contrast of an
imaging system. The higher the MI value, the better the image quality. The advantages of
using the MI metric are: (1) simplicity of computation, (2) simplicity of experimentation, and
(3) combined assessment of image contrast, noise and resolution.
The structure of this chapter is as follows. Section **.2 provides a basic overview of factors
that affect medical image quality. Section **.3 describes the mutual information-based
evaluation framework utilized in this work. An example of how to calculate MI is also given
to provide a deep understanding of applying MI to the evaluation of medical imaging
systems. Section **.4 shows a series of computer simulations, followed by investigating the
utility and superiority of MI method by evaluating the performance of two imaging-plate
detectors. Section **.5 presents the results that were obtained. Section **.6 ends with a
discussion and conclusions.
2. Background
In medical imaging, image quality is determined by at least five factors: contrast, resolution,
noise, artifacts, and distortion. Of these factors, resolution and noise are the most commonly
used physical characteristics. As is well known, they are described by the modulation
transfer function (MTF) and noise power spectrum (NPS), respectively. The MTF describes
the ability of an imaging system to reproduce the frequency information contained in the
incident x-ray signal. The NPS describes the frequency content of the noise of an imaging
system. However, one of the dilemmas in medical radiography is the extent to which theses
characteristics affect image quality. In comparison of two imaging systems, for example, an
imaging system may only be superior in one physical characteristic while being inferior to
another in the other characteristic. To deal with this issue, the noise equivalent quanta or
detective quantum efficiency (DQE), which can be calculated if the MTF, NPS, and the input

Medical Imaging

196
signal-to-noise ratio of the x-ray beam used to measure the NPS are known, is used as a
single parameter to describe the general quality of the system. Measurements of the MTF
and NPS are conceptually straightforward but difficult to carry out experimentally and
accurately. Moreover, the results of these measurements vary with the methods employed.
Therefore, a simple and synthetic method for measuring image quality has been desirable.
In this chapter, we present a simple and straightforward method for synthetically evaluating
digital radiographic images using MI. MI originating from information theory has been used
as an effective similarity metric in medical image registration tasks and template matching
schemes, and used as a feature selection criterion in computer-aided detection (Last et al.,
2001; Pluim et al., 2003; Saunders et al., 2003; Tourassi et al., 2007). From the diversity of
modalities (for example, computed tomography, magnetic resonance, and positron-emission
tomography) and objects (the imaged anatomy) found in the literature, it is clear that MI has
become a generally applicable measure. Here, the difference between the MI employed in
image fusion and that employed in the present work needs to be clarified. In image fusion,
MI is a similarity measure and usually serves as a criterion of alignment between two
images. MI reaches its maximum value when two images are well aligned. In contrast, in the
current work, MI is a physical measure of image quality and serves as a metric of overall
physical quality of the imaging system being investigated. MI reaches its maximum value
when the detected image (output of the transmission channel) completely corresponds to
the image object (input of the transmitted channel).
Several studies have been published on the relation between MI and image quality. Using
two Lucite step-wedges as phantoms to study the relations between MI and image noise as
well as image blurring was made (Tsai et al., 2008) . However, the study did not examine the
effect of image contrast on MI. Moreover, it did not make a comparison of MI with the MTF,
NPS, and DQE. Investigating the combined effect of noise and resolution degradation on MI
value by employing two imaging plates used for computed radiography was also conducted
(Matsuyama et al., 2008). These pervious studies concluded that MI has close correlation
with both image noise and image blurring. However, the study has not taken into account
the effect of image contrast on MI. Furthermore, it did not make a direct comparison of MI
with other image quality metric such as DQE.
The current study includes the following contents: (1) investigating individual and
combined effects of contrast, noise and blur on images obtained from medical imaging
systems; (2) conducting various simulation studies with a parametric model to verify the
relationship of MI among the three major physical factors affecting image quality in medical
imaging systems; and (3) comparing the evaluation results obtained using the MI metric to
that using the DQE metric. In addition, two imaging plates for computed radiography were
used for verification of the potential usefulness of the MI metric. The verification was made
by showing clinical images with discussion.
3. Mutual information-based evaluation framework
MI is briefly described as follows.
Given events S
1
,.. S
n
occurring with probabilities p(S
1
), .p(S
n
), then the average
uncertainty associated with each event is defined by the Shannon entropy as

A Mutual Information-Based Image Quality Metric for Medical Imaging Systems

197

Fig. 1. Relationship among H(x), H(y), H(x,y), H
x
(y), H
y
(x), and MI(x;y).

n
i i
i
H S p S p S
2
1
( ) ( ) log ( )

(1)
Considering x and y as two random variables corresponding to an input variable and an
output variable, the entropy for the input and that for the output are denoted as H(x) and
H(y), respectively. For this case the joint entropy, H(x,y), is defined as

x y
H x y H x H y H y H x ( , ) ( ) ( ) ( ) ( )
(2)
where H
x
(y) and H
y
(x) are conditional entropies. They are the entropy of the output when
the input is known and that of the input when the output is known, respectively. In this
situation, we can compute MI, MI(x;y), as:

y x
MI x y H x H x H y H y
H x H y H x y
( ; ) ( ) ( ) ( ) ( )
( ) ( ) ( , )

(3)
A useful way of visualizing the relationship between these entropies is provided by a Venn
diagram as shown in Fig.1. Consider an experiment in which every input has a unique
output belonging to one of various output categories. In this study, for simplicity, the inputs
may be considered to be a set of subjects (e.g., phantoms in simplicity) varying in
composition, while the outputs may be their corresponding images varying in optical
density or gray level. An orderly system is employed in the present study to calculate the
entropies of input, output, and their joint entropies (Attneave, 1959). With this orderly
system, the amount of MI is easily computed. The frequency with which each output is
made to each input is recorded in Table 1.

H

y

(

x

)

H

x

(

y

)

MI

(

x

;

y

)

H

(

x
,

y

)

H

(

x

)

H

(

y

)

Medical Imaging

198

Input x
Output y x
1
x
2
x
i
X Frequency
y
1
n
11
n
21
n
i1
n
X1
n
j=1

y
2
n
12
n
22
n
i2
n
X2
n
j=2

y
3
n
13
n
23
n
i3
n
X3
n
j=3

y
j
n
1j
n
2j
n
ij
n
Xj
n
j=j

Y n
1Y
n
2Y
n
iY
n
XY
n
j=Y

Frequency n
i=1
n
i=2
n
i=i
n
i=X
n

Table 1. A data matrix of occurrence frequency for Y outputs to X inputs.
The columns and lows of this table represent various inputs and outputs. The various
inputs, x
1
, x
2
x
i
..X, are assumed to take discrete values of input variables x. Likewise,
the various outputs, y
1
, y
2
y
j
.Y are discrete values of output variables y. The upper-
case X and Y stand for the number of input and output categories, respectively. Note that
the subscript i refers to any particular but unspecified input, whereas the subscript j refers to
any particular but unspecified output. The number of times input x
i
is presented will be
symbolized by n
i
, the frequency of output, y
j
, by n
j
, and the frequency, with which the input
x
i
corresponds to the output y
j
, is given by n
ij
. The total of all frequencies is given by n. It is
apparent from Table 1 that

ij i
j
n n

(4)

ij j
i
n n

(5)

ij i j
ij i j
n n n n .

(6)
Referring to the definition of information entropy as shown in Equation (1), three
informational quantities, namely, H(x), H(y), and H(x,y), can be calculated from Table 1.

i i
i
H x p p
2
( ) log (1 / )

(7)

j j
j
H y p p
2
( ) log (1 / )

(8)


199

ij ij
ij
H x y p p
2
( , ) log (1 / )

(9)
where p
i
=n
i
/n, p
j
=n
j
/n, and p
ij
=n
ij
/n. For simplicity, we can rewrite the above equations as
follows:

i i
i
H x n n n n
2 2
( ) log (1 / ) log

(10)

j j
j
H y n n n n
2 2
( ) log (1 / ) log

(11)

ij ij
ij
H x y n n n n
2 2
( , ) log (1 / ) log

(12)
Then, the MI MI(x;y) can be obtained from Equation (3) together with Equations (10), (11),
and (12). The MI conveys the amount of information that y has about x.
Table 2 gives an example of how to calculate MI. Assume that a subject (e.g., a step-wedge)
having five steps with different thickness was used for the experiment. The five steps
correspond to five inputs present equiprobably. The gray-scale pixel values of 100 pixels in
each step after imaging were measured randomly. The distributions of the pixel values are
considered as the corresponding outputs and their respective frequencies are given in the
table. The frequencies will be referred to by means of the symbols given in Table 1; for
example: n
12
=60, n
j=3
=118, n
i=2
=100, n=500, and so on. Now, there are three information
quantities, namely, H(x), H(y), and H(x,y), that can be calculated directly from Table 2 by
using equations (10), (11), and (12).

Input x
Output y 1 2 3 4 5 Frequency
1 20 20
2 60 4 64
3 20 88 10 118
4 8 76 14 98
5 12 80 2 94
6 2 6 8 16
7 90 90
Frequency 100 100 100 100 100 500
Input x
Output y 1 2 3 4 5 Frequency
1 20 20
2 60 4 64
3 20 88 10 118
4 8 76 14 98
5 12 80 2 94
6 2 6 8 16
7 90 90
Frequency 100 100 100 100 100 500

Table 2. An example of how to calculate the mutual information. The frequencies shown in
the table is referred to by means of the symbols given in Table 1, for example, n
23
=88,
n
j=2
=64, n
i=1
=100, n=500, and so on.

Medical Imaging

200
For the data given in Table 2,
i
i
H x n n n n
2 2 2
( ) log (1 / ) log log 5 2.323
(since inputs are equiprobable)

H y etc
2 2 2 2
( ) log 500 (1 /500) (20log 20 64log 64 118log 118...... .) 2.575
H x y etc
2 2 2 2
( , ) log 500 (1 /500) (20log 20 60log 60 4log 4..... .) 3.235
Applying Equation (3) to the values calculated above, we have
MI x y H x H y H x y ( ; ) ( ) ( ) ( , ) 2.323 2.575 3.235 1.663 .
This is the estimate of the amount of information transmitted by the subject from input to
output: 1.633 bits, out of a possible of 2.323 bits.
If the output is identical to the input, then knowing the output provides complete
information about the input. In this case, the MI is maximized and equal to the input
entropy, and the uncertainty of the input is reduced to 0. It means that knowing (or viewing)
the image of an object (subject) receives complete information about the object (subject).
Thus the quality of the obtained image reaches to a maximum in terms of the MI. If, on the
other hand, the output and the input are independent, then knowing the output does not
help make any conclusions about the input. In this case, MI value is zero, and therefore the
uncertainty about the input remains unchanged. This means that the obtained image has the
lowest quality from the point of view of the MI.
4. Materials and methods
4.1 Image simulation
Simulation studies were designed, and the framework is as follows. In mathematical terms,
a simulated image g(x,y) is the convolution of a uniformly distributed signal (an object) f(x,y)
and the blurring function B. If the noise u(x,y) is also taken into consideration, the simulated
image may be represented by the following formula:

k
g x y k f x y B u x y W
5
1
( , ) {[ ( , )] ( , ) }

(13)
where the symbol * represents the convolution operation, and k is an integer representing
the number of steps of the simulated image (k=1,2, .5). In the simulation studies, the input
signal is a five-step wedge or a five-gray-level grid pattern with a specific intensity or pixel
value on each step. The term of W is a weighting coefficient used to adjust the extent of
noise, and u(x,y) is a zero-mean Gaussian noise with a standard deviation of 0.5.
An image of a simulated step-wedge generated by using Equation (13) is shown in Fig.2(a).
Five regions of interests (ROIs) indicated with rectangles near the boundaries of two
adjacent steps were chosen for calculation of MI. The five steps of the image are numbered
from the right as step 1, step 2, and so on. The left band without a rectangular box is
considered as the background of the image. The corresponding pixel-value distributions
measured from the ROIs are given in Fig.2(b). The area of each ROI used in this study was


201
50200 pixels. As a result, a total of 10,000 data for each step was obtained. As shown in
Fig.2(a), the number of inputs is five, and the number of outputs is the range of gray levels
shown on the horizontal axis of the pixel-value distributions [see Fig.2(b)].
0
200
400
600
800
1000
1200
1400
1600
1800
300 400 500 600
Pixel Value
F
r
e
q
u
e
n
c
y
Step 5
Step 4
Step 3
Step 2
Step 1

(a) (b)
Fig. 2. (a) Computer-simulated step-wedge. A region of interest (ROI) shown with a
rectangle at each step of the step-wedge was chosen for entropy computation. (b) The
corresponding pixel-value distributions measured from the ROIs shown in (a).
0
200
400
600
800
1000
1200
1400
1600
1800
2000
300 400 500 600
Pixel Varue
F
r
e
q
u
e
n
c
y
Step 5
Step 4
Step 3
Step 2
Step 1

4 3 2 1
5 5
5 5
3
2
1 4
2
3
2 3

(a) (b)
Fig. 3. (a) Simulated grid-pattern image with five different gray levels. (b) The
corresponding pixel-value distributions measured from the ROIs of (a).
Grid-pattern images with various noise levels and different size of blur were also generated
using Equation (13) for another simulation study. A simulated grid-pattern image is
illustrated in Fig.3(a). Five different gray levels were used to construct the grid-pattern
image. The image consists of 16 blocks and is symmetric with respect to the main diagonal.
An ROI with a size of 5050 was selected from the central area of the blocks numbered 2, 3,
and 5, while two nonoverlapped ROIs near the central area of the blocks numbered 1 and 4
(at the four corners) were chosen for MI measurement. As a result, a total of 10,000 data for a
specific gray level could be obtained. Fig.3(b) illustrates the corresponding pixel-value
distributions measured from the ROIs of the simulated image [Fig.3(a)].

Medical Imaging

202

Fig. 4. Photograph of the three step-wedges (phantoms A, B, and C) used in the experiments.
Three different simulations were performed using the simulated step-wedge and grid
pattern. The first simulation was carried out to investigate the relationship between image
contrast and MI for various noise levels and different extent of blurring. In this study, we
defined image contrast as the difference of the mean pixel values between two adjacent
steps of a simulated step-wedge [Fig.2(a)] or the difference between two consecutive
numbers of gray-level steps [Fig.3(a)]. We employed signal-to-noise ratio (SNR) to describe
the extent of noise level. The signal and noise used for SNR calculation were f(x,y) and
u(x,y)W, respectively, as given in Equation (13). As a blurring function, we used a Gaussian
filter with a size of dd (d is an odd integer). The extent of blurring was adjusted by varying
the filter size. The second simulation was performed to investigate the relationship between
the image noise and MI for different extent of blurring and various levels of contrast. The
third simulation was conducted to investigate the relationship between the blurring and MI
for various levels of noise and contrast.
4.2 Real images of step wedges
In addition to the simulation studies, phantom studies were also conducted (Matsuyama et
al., 2009). Three Lucite step-wedges with 0, 0.5, 1.0, 1.5, 2.0, and 2.5 mm (phantom A); 0, 1, 2,
3, 4, and 5 mm (phantom B); and 0, 1.5, 3.0, 4.5, 6.0, and 7.5 mm (phantom C) in thickness
were used as objects for experiments (see Fig.4). Two imaging plates (IPs) for computed
radiography (standard resolution type ST and high resolution type HR, Fuji Film Inc.,
Tokyo, Japan) were used as detectors to record x-ray intensities for performance evaluation.
It is known that the intensity of the transmitted x-ray beam is reduced when the thickness of
the step-wedge increases. The area of each ROI and the number of data used for calculation
of MI were the same as those used in the step-wedge simulation studies.
4.3 Detective quantum efficiency measurement
In order to connect MI to the commonly used image quality metric, the DQE that is usually
obtained from presampling MTF and NPS was measured. As is well known, the
presampling MTF and NPS are used to describe the spatial resolution properties (blur) and


203
noise properties of imaging systems, respectively. The presampling MTF of IPs was
measured with an angled-edge method (Samei et al., 1998). The edge is made of a 100-m-
thick sharp-edged-tungsten plate, and its dimension was 1010 cm
2
. After the image of the
edge was acquired, the digital image data were transferred to a computer for computation.
The details of the processing method are given elsewhere (Flymn & Samei, 1999). NPS
measurements were made by exposing IPs to a uniform beam of radiation. For the
calculation, the central portion of each obtained uniform image was divided into multiple
non-overlapping regions, 256256 in size. A total of 25 regions were used. The details of the
methodology are reported elsewhere (Monnin et al., 2007; Samei & Flynn, 2002).
The DQE is a spatial frequency-based measurement of the ability of the imaging device to
convert the spatial information contained in the incident x-ray fluence to useful image
information (Fetterly & Hangiandreou, 2001; Neitzel et al., 2004; Spahn, 2005). It is
defined as

out in
DQE SNR SNR
2 2
/
(14)
where SNR
out
and SNR
in
are the spatial frequency-dependent signal-to-noise ratios of the
imaging device at the output and input, respectively. It was calculated using the following
formula (Fettery & Schueler, 2006).
DQE MTF q NNPS
2
/( )
(15)
where q is the x-ray photon fluence density (mm
-2
) used for the uniform exposure image,
and NNPS is the normalized NPS. For a perfect imaging detector, DQE can reach a
maximum value of 1.0.
As can be seen from Equation (15), three quantities must be measured to obtain DQE. It is
obvious from the equation that DQE value would be high when any of the following
situations occurs: (1) high MTF value (high spatial resolution), (2) low NNPS value ( low
noise level), and (3) low x-ray photon fluence density. Because the calculation of DQE
includes a complicated set of measurements, there is thus a need to provide an easier and
less complicated methodology for the use of assessing overall image quality. The present
work was just motivated by this need.
5. Results
Simulations were performed to investigate individual effects of contrast, noise and blur on
MI. Fig.5(a) shows the relationship between the contrast and MI for different levels of SNR,
when the filter size (FS) of blurring function was 11 (FS=1). On the left is the result
obtained from the simulated step-wedges, while on the right is that obtained from the
simulated grid patterns. Fig.5(b) illustrates the relationship between the contrast and MI for
various levels of blur, when the SNR was fixed at 35 dB.
As a whole, the results show that MI increases with the increase of image contrast at
constant levels of noise and blur. It is seen from Fig.5(a) that the MI at low noise levels (high
level of SNR) shows remarkable increase as compared to high noise levels (low level of
SNR). For example, the MI value at SNR of 31 is considerably lower than that at SNR of 40.

Medical Imaging

204
This means that the MI value is greatly influenced by the noise level. As shown in Fig.5(b),
the MI curves of different levels of blur (filter size; FS) are similar in shape. The difference in
MI values at low contrast level is not obvious, even if the filter sizes change. However, the
difference becomes notable at high contrast levels. The results demonstrate that the effect of
blur on MI value is more obvious at higher contrast levels as compared to that at lower
contrast levels. It is noted that the results obtained by using the simulated step-wedges and
that by the grid patterns have a similar tendency.

simulated step-wedge
simulated grid pattern
(a)
(b)
0
0.6
1.2
1.8
2.4
0 10 20 30 40 50
M
I

(
M
u
t
u
a
l

I
n
f
o
r
m
a
t
i
o
n
)

[
b
i
t
s
]
Contrast
SNR40
SNR35
SNR31
FS=1
0
0.6
1.2
1.8
2.4
0 10 20 30 40 50
M
I

(
M
u
t
u
a
l

I
n
f
o
r
m
a
t
i
o
n
)

[
b
i
t
s
]
Contrast
SNR40
SNR35
SNR31
FS=1
1.2
1.4
1.6
1.8
2.0
2.2
2.4
0 10 20 30 40 50
M
I

(
M
u
t
u
a
l

I
n
f
o
r
m
a
t
i
o
n
)

[
b
i
t
s
]
Contrast
FS1
FS31
FS61
SNR 35 dB
FS 1
FS 61
1.2
1.4
1.6
1.8
2.0
2.2
2.4
0 10 20 30 40 50
M
I

(
M
u
t
u
a
l

I
n
f
o
r
m
a
t
i
o
n
)

[
b
i
t
s
]
Contrast
FS1
FS31
FS61
SNR 35 dB
FS 1
FS 61

Fig. 5. Relationship between the contrast and MI. Left column: simulated step-wedge. Right
column: simulated grid pattern. (a) For various levels of SNR at a size of blurring function
FS of 1. (b) For various levels of blur at an SNR of 35dB.
Figs.6(a) and 6(b) illustrate MI as a function of SNR for various levels of blurring at image
contrast of 20 and for different contrast levels at FS=1, respectively. The results from the
figures indicate that the MI value increases with the increase of SNR (decrease in noise


205
level). The figures on the left were the results from the simulated step-wedges, while those
on the right were from the simulated grid patterns. It is seen from Fig.6(a) that the difference
in MI values among various filter sizes at low SNR levels (high noise levels) is not
significant. Similar results were obtained at high SNR levels (low noise levels). MI reaches to
its maximum when SNR is higher than 45 dB. This implies that MI could be almost the same
value when noise level is lower than a certain level. As illustrated in Fig.6(b), the MI curves
for different contrast levels are in similar shape. MI reaches to the maximum when SNR is
approximately 38 dB at contrast of 30, and similarly, when SNR is 40 dB at contrast of 20.
This means that the two images would provide the same image quality in terms of MI
metric. It is noted that the results obtained from the simulated step-wedge and those from
grid patterns have the same tendency.
simulated step-wedge simulated grid pattern
(a)
(b)
1.4
1.6
1.8
2.0
2.2
2.4
30 35 40 45 50
M
I

(
M
u
t
u
a
l

I
n
f
o
r
m
a
t
i
o
n
)

[
b
i
t
s
]
Signal-to-Noise Ratio [dB]
FS=1
FS=31
FS=61
Contrast 20
FS 1
FS 61
1.4
1.6
1.8
2.0
2.2
2.4
30 35 40 45 50
M
I

(
M
u
t
u
a
l

I
n
f
o
r
m
a
t
i
o
n
)

[
b
i
t
s
]
FS=1
FS=31
FS=61
Contrast 20
FS 1
FS 61
0
0.6
1.2
1.8
2.4
15 20 25 30 35 40 45 50
M
I

(
M
u
t
u
a
l

I
n
f
o
r
m
a
t
i
o
n
)

[
b
i
t
s
]
Contrast 30
Contrast 20
Contrast 10
FS = 1
0
0.6
1.2
1.8
2.4
15 20 25 30 35 40 45 50
M
I

(
M
u
t
u
a
l

I
n
f
o
r
m
a
t
i
o
n
)

[
b
i
t
s
]
Contrast 30
Contrast 20
Contrast 10
FS = 1

Fig. 6. Relationship between the SNR and the MI. Left column: simulated step-wedge. Right
column: simulated grid pattern. (a) For various levels of blur at an image contrast of 20. (b)
For various levels of contrast at a size of blurring function (FS) of 1.

Medical Imaging

206
Fig.7(a) shows MI as a function of filter size of blurring function for various levels of SNR at
image contrast of 20 for the simulated step-wedges (left) and grid patterns (right). Fig.7(b)
shows MI as a function of filter size of blurring function for various contrast levels at SNR of
35 dB for the two various simulated images. The results from the figures show that MI value
decreases when filter size of the blurring function increases, although the decrease is
relatively small. This means that the effect of the level of blur on the MI is not so obvious in
comparison to noise and contrast. Fig.8 illustrates images of the simulated step-wedges and
grid patterns with different sizes of blurring (FS=7, 21, and 41), while the SNR and image
contrast were kept constant at 30 and 20, respectively. The images demonstrate that image
resolution degrades with the increase of filter size of blurring function. Therefore, MI values
decrease with the increase of filter size.
simulated step-wedge simulated grid pattern
(a)
1.8
1.9
2.0
2.1
2.2
2.3
2.4
0 20 40 60
M
I

(
M
u
t
u
a
l

I
n
f
o
r
m
a
t
i
o
n
)

[
b
i
t
s
]
Filter Size of Blurring Function
SNR40 SNR38 SNR35
Contrast 20
1.9
2.0
2.1
2.2
2.3
2.4
0 20 40 60
M
I

(
M
u
t
u
a
l

I
n
f
o
r
m
a
t
i
o
n
)

[
b
i
t
s
]
Filter Size of Blurring Function
SNR40 SNR38 SNR35
Contrast 20

(b)
1.6
1.8
2.0
2.2
2.4
0 20 40 60
M
I

(
M
u
t
u
a
l

I
n
f
o
r
m
a
t
i
o
n
)

[
b
i
t
s
]
Filter Size of Blurring Fnction
Contrast 30 Contrast 25 Contrast 20 SNR 35
1.6
1.8
2.0
2.2
2.4
0 20 40 60
M
I

(
M
u
t
u
a
l

I
n
f
o
r
m
a
t
i
o
n
)

[
b
i
t
s
]
Filter Size of Blurring Fnction
Contrast 30 Contrast 25 Contrast 20 SNR 35

Fig. 7. Relationship between the filter size of blurring function and the MI. Left column:
simulated step-wedge. Right column: simulated grid pattern. (a) For various levels of SNR
at an image contrast of 20. (b) For various levels of image contrast at a SNR of 35 dB.


207

FS=41
FS=21
FS=7
FS=41
FS=21
FS=7

Fig. 8. Images of the simulated step-wedge and grid pattern with different sizes of blurring
(FS=7, 21, and 41), while the SNR and image contrast were kept constant at 30 and 20,
respectively.
Fig.9 shows MI as a function of exposure dose for the images of a Lucite step-wedge
(phantom B) obtained with ST and HR IPs for computed radiography. The results illustrate
that MI increases with the increase of exposure dose. The rise of MI value might be mainly
due to the decrease of noise, resulting from the increase of radiation dose. The trend of the

0 5 10 15 20 25 30
0
0.6
1.2
1.8
2.4
Exposure Dose [mAs]
M
I

(
M
u
t
u
a
l

I
n
f
o
r
m
a
t
i
o
n
)

[
b
i
t
s
]
* ST
+ HR
0 5 10 15 20 25 30
0
0.6
1.2
1.8
2.4
Exposure Dose [mAs]
M
I

(
M
u
t
u
a
l

I
n
f
o
r
m
a
t
i
o
n
)

[
b
i
t
s
]
* ST
+ HR

Fig. 9. Mutual information as a function of exposure dose for the images of a Lucite step-
wedge (phantom B) obtained with ST and HR imaging plates.

Medical Imaging

208
MI curves is similar to that shown in Fig.6(b), although the two figures are plotted with
different units: one is the exposure dose, and the other is the SNR. However, it is reasonable
to say that the two units are associated with noise levels and are closely correlated. As
shown in the figure, the MI value for the ST plate is higher than that for the HR plate at the
same exposure dose. This can be explained by the fact that combined effects of blur and
noise lead to a higher MI value for the ST plate at a given image contrast. This suggests that
the image obtained with the ST plate transmits more information in comparison to that with
the HR plate under the same exposure conditions.
Fig.10(a) illustrates the measured presampling MTFs of the two IPs at 42 kV. The result
indicates that the HR plate has higher MTF as compared to the ST plate. This is mainly due
to the difference in the spatial resolution of the two IPs: HR is a high-resolution plate, while
ST is a standard one.
0
0.2
0.4
0.6
0.8
1
0 1 2 3 4
Spatial Frequency [cycles/mm]

P
r
e
s
a
m
p
l
i
n
g

M
T
F
HR
ST
1.0E-06
1.0E-05
1.0E-04
1.0E-03
0 1 2 3 4 5 6
N
P
S

[
m
m
2
]
HR
ST

(a) (b)
0
10
20
30
40
0 0.5 1 1.5 2 2.5 3 3.5 4
D
Q
E

[
%
]
ST
HR

(c)
Fig. 10. (a) Experimental results for the presampling MTF measurement with the edge method
for the two imaging plates. (b) NPS versus spatial frequency for the two imaging plates. (c)
DQE versus spatial frequency at 76-Gy exposure level for the two imaging plates.


209
In Figs.11 and 12, we display the real images of the femur and metacarpus acquired with ST
and HR IPs under the same exposure conditions. In these two figures, the left column
illustrates the original images, while on the right are the magnified images of the white
squares indicated in the original images. It is clear from the magnified images of Fig.11 that
the lesser trochanter (with a white arrow) obtained with HR plate shows better resolution as
compared to ST plate. Similarly, the magnified images of Fig.12 (the carpal bone indicated
by white arrows) obtained with HR plate shows better resolution as compared to ST plate.
The experimental validation provides confirming evidence for the MTF results.

ST ST
HR
HR

Fig. 11. Clinical images of the femur acquired with ST and HR image plates.

Medical Imaging

210

ST
ST
HR
HR

Fig. 12. Clinical images of the metacarpus acquired with ST and HR image plates.
Fig.10(b) illustrates the NPS versus spatial frequency at 42 kV acquired with ST and HR IPs.
As shown in the figure, the NPS of the HR IP is higher than that of the ST plate at the same
radiation dose. It can be seen from the magnified images of Figs.11 and 12 that the images
acquired with HR plates show higher noise levels. The perceptual results correctly reflect
the outcome of the NPS shown in Fig.10(b).


211
The DQEs of ST and HR IPs versus spatial frequency at 76-Gy exposure level are presented
in Fig.10(c). In the present study, the DQEs were assessed using Equation (15) in order
words, the results of Fig.10(c) were obtained from the measured results shown in Figs.10(a)
and 10(b). It is known that the MTF values are generally independent of exposure levels.
Thus the MTF values shown in Fig.10(a) were used for calculating the DQEs of IPs at
different exposure doses. In spite of having lower MTFs, the DQEs for the ST plates are
higher than those for the HR plates at the same exposure level. The higher value in DQE
might be attributed to the better noise performance of ST plates. In other words, as
compared to resolution, noise greatly affects overall performance of the imaging systems.
When looking at Fig.9 and Fig.10(c), the performance ranking of MI values for ST and HR
IPs and those of DQE values for the two IPs are the same, i.e., the MI value and DQE value
for the ST plate are higher than those for the HR plate. The experimental results may
confirm that MI and DQE metrics are highly correlated.
6. Discussion and conclusions
In our simulation studies, we demonstrated that MI increases with the increase of contrast
and decreases with the increase of noise and blur. Therefore, it is considered that MI could
be used as a simple metric for evaluation of overall imaging performance. In this study, we
applied the MI metrology to evaluate the performance of two IPs for computed
radiography. The measured MI shown in Fig.9 is consistent with the DQE shown in
Fig.10(c), although they are described in different domains: one is in the spatial domain
scalar metric, and the other is in the spatial frequency domain metric. The results suggest
the usefulness of the proposed MI metric.
There are several advantages of using the MI metric to evaluate the performance of
imaging systems. First, computation of MI is much easier in comparison to that of spatial
frequency domain measures such as MTF, NPS and DQE. Second, the experiment setup is
simple. For example, a step-wedge or an equivalent test device is sufficient for conducting
experiments. Third, three of the most important image quality factors, i.e., contrast, noise,
and blur, can be integrated for overall evaluation. However, it should be stated clearly
that our proposed MI metric is not intended for replacing the conventionally used
metrics. The main objective of the present work is to provide a scalar metrology based on
simple image statistics for image quality evaluation.
In conclusion, we have described an information-theoretic method for quantifying overall
image quality in terms of MI. We demonstrated by way of image simulation that MI increases
with contrast, decreases with noise, and increases with resolution. We investigated the utility
of this method by applying it to evaluating the performance of two imaging detectors. We also
compared evaluation results in terms of MI against those in terms of the commonly used DQE
metric. Our simulation and experimental results demonstrate that the proposed method is
simple to implement and has potential usefulness for evaluation of overall image quality.
7. Acknowledgment
This work was supported in part by the Ministry of Education, Culture, Sports, Science and
Technology of Japan (Tokyo, Japan) through a grant-aid for Scientific Research (23602004).

Medical Imaging

212
8. References
Attneave, F. (1959). Applications of Information Theory to Psychology. Holt, Rinehart and
Winston, New York
Fetterly, K.A. & Hangiandreou, N.J. (2001). Effect of x-ray spectra on the DQE of a computed
radiography system. Medical Physics, Vol.28, pp. (241-249)
Fettery, K.A. & Schueler, B.A. (2006). Performance evaluation of a computed radiography
imaging device using a typical front side and novel dual side readout storage
phosphors. Medical Physics. Vol.33, pp. (290-296)
Flynn, M. & Samei, E. (1999). Experimental comparison of noise and resolution for 2k and 4k
storage phosphor radiography systems. Medical Physics, Vol.26, pp. (1612-1623)
Last, M.; Kandel, A. & Maimon, O. (2001). Information-theoretic algorithm for feature
selection. Pattern Recognition Letters, Vol.22, pp. (799-811)
Matsuyama, E.; Tsai, D.Y. & Lee, Y. (2009). Mutual information-based evaluation of image
quality with its preliminary application to assessment of medical imaging systems.
Journal of Electronic Imaging, Vol.18, pp. (033011-1-11)
Matsuyama, E.; Tsai, D.Y.; Lee, Y.; Sekiya, M. & Kojima, K. (2008). Physical characterization
of digital radiological images by use of transmitted information metric. Proceedings
of SPIE Medical Imaging, Vol.6913, pp. (69130V1-8)
Monnin, P.; Gutierrez, D.; Bulling, S.; Guntern, D. & Verdun, F.R. (2007). A comparison of
the performance of digital mammography systems. Medical Physics, Vol.34, pp.
(906-914)
Neitzel, U.; Gunther-Kohfahl, S.; Borasi, E. & Samei, E. (2004). Determination of the
detective quantum efficiency of a digital x-ray detector: comparison of three
evaluations using a common image data set. Medical Physics, Vol.31, pp. (2205-2211)
Pluim, J.P.M.; Maintz, J.M.A. & Viergever, M.A. (2003). Mutual-information-based
registration of medical images: A survey. IEEE Transactions on Medical Imaging,
Vol.22, pp. (986-1004)
Samei, E. & Flynn, M.J. (2002). An experimental comparison of detector performance for
computed radiography systems. Medical Physics, Vol.29, pp. (447-459)
Samei, E.; Flynn, J. & Reimann, D.A. (1998). A method for measuring the presampled MTF
of digital radiographic systems using an edge test device. Medical Physics, Vol.25,
pp. (102-113)
Saunders, Jr R.S. & Samei, E. (2003). A method for modifying the image quality parameters
of digital radiographic images. Medical Physics, Vol.30, pp. (3006-3017)
Spahn, M. (2005). Flat detectors and their clinical applications. European Radiology, Vol.15,
pp(1934-1947)
Tourassi, G.D.; Harrawood, B.; Singh, S. & Lo, J.Y. (2007). Information-theoretic CAD system
in mammography: Entropy-based indexing for computational efficiency and robust
performance. Medical Physics, Vol.34, pp. (3193-3204)
Tsai, D.Y,; Lee, Y. & Matsuyama, E. (2008). Information-entropy measure for evaluation of
image. Journal of Digital Imaging, Vol.21, pp. (338-347)
Part 3
Applications in Clinical Settings

10
Diagnostic Imaging in Oral
and Maxillofacial Pathology
Hasan Ayberk Altug
1
and Aydin Ozkan
2

1
Gulhane Military Medical Academy
Department of Oral and Maxillofacial Surgery
2
Diyarbakir Military Hospital, Dental Service
Turkey
1. Introduction
During the diagnosis of oral and maxillofacial diseases, clinical and radiological data play a
major role. In this region, only a good clinical diagnosis along with a radiological examination
may lead to a successful diagnosis. A successful diagnosis and evaluation of clinical
examination are generally up to a profound knowledge of the normal anatomy of the region.
2. Radiographic anatomy of oral and maxillofacial region
X-rays (invisible rays) were discovered by W. Conrad Roentgen in 1895. They are a form of
electromagnetic radiation with high energy and are part of electromagnetic spectrum. In
order to create X-ray, a target tissue is bombardized with energized electrons and then they
are suddenly brought to rest. The entire process takes place in a small evacuated glass
envelope which is called X-ray tube (Whaites, 2002).
2.1 Periapical radiography
Periapical radiography is a projection of radiographs including interoral radiographs which
depict 3-4 teeth and the tissue around them (Whaites, 2002). There are two projection
techniques for periapical radiography:
The paralleling technique (Long-cone technique): The periapical film is stood parallel to
the long axis of the teeth and the central is aimed at the right angles of the teeth and the
film (Fig. 1A).
The bisecting-angle technique: The periapical film is stood as close as possible to the
palatal/lingual surface of the teeth. The film and the teeth form an angle with its apex
at the point where the film is in contact with the teeth. Central ray is directed at apex of
the teeth (Fig. 1B) (White & Pharoah, 2004).
In order to create a high-quality radiograph, the central ray beam must pass through root
apex or alveolar crest. Radiolucent/radioopaque images which were obtained with
periapical radiography may not only depict pathological conditions which require treatment
but also normal anatomic variations. Therefore, achieving a good differential diagnosis has

Medical Imaging

216
an utmost importance. 10 periapical radiographs, 5 of them for the upper jaw and 5 of them
for the lower jaw, are applied for kids, whereas 14 periapical radiographs, 7 of them for the
upper jaw and 7 for the lower jaw, are applied for the adolescents, performed using the
paralleling technique (Fig. 2). It needs to different projection angle to capture third molar.
During the creation of periapical radiographs, film holders might be used in order to
comply with standardization. However, free-handed positioning may also be preferred
(Wood et al. 1997; Pasler, 1993).

(a) (b)

Fig. 1. (a) The paralleling technique; (b) The bisecting-angle technique.

Fig. 2. Periapical radiographic survey for adolescents.
Indications:
Evaluation of periapical and periodontal tissue health
Before, during and/or after surgical and endodontic treatments
Assessment of the teeth and adjacent tissue after trauma
Evaluation of apical pathology within the alveolar bone
To clarify of the presence/absence of unerupted teeth (Whaites, 2002).

Diagnostic Imaging in Oral and Maxillofacial Pathology

217
2.1.1 Anatomic landmarks of periapical radiograph
2.1.1.1 Maxilla
Maxillary anterior region
Cervical dentin of the anterior teeth is penetrated in its lateral aspects by the X-ray beam. It
is seen in the radiograph as a radiolucent image which is known as burn-out effect. The
anterior portions of the nose and the median suture can also be seen clearly in the
radiographs taken from maxillary anterior region (Fig.3A).
Maxillary canine region
This projection exhibits a nasal process of the maxilla and the nasal soft tissues.
Nasopalatine canal, incisive foramen and anterior lobe of the maxillary sinus can also be
visible in this projection (Fig.3B).
Maxillary premolar region
The radiographs which were taken from premolar region exhibit the floor of the nasal cavity
and maxillary sinus, usually separated from septum above the root tip of the second
premolar (Fig.3C).
Maxillary molar region
The radiographs which were taken from premolar region exhibit maxillary sinus, maxillary
tuberosity, and usually the body of the zygoma. Sometimes the process of the palatal bone,
the pterygoid process and coronoid process of the mandible, so-called radix relicta appear
in the radiograph (Fig.3D) (Pasler,1993; Pasler&Visser, 2003).

Fig. 3. A: Periapical view and schematic drawing of maxillary anterior region B: Periapical
view and schematic drawing of maxillary canine region C: Periapical view and schematic
drawing of maxillary premolar region D: Periapical view and schematic drawing of
maxillary molar region

Medical Imaging

218
2.1.1.2 Mandible
Mandibular anterior region
The radiographs which were taken from the anterior region exhibit 4 mandibular incisor
teeth, mental fovea which shows a radiolucent outfit, vascular canals and the chin
prominence. Burn-out effect may also be observed in this radiograph just like in the
radiographs which were taken from the maxillary region (Fig. 4A).
Mandibular canine region
The radiographs which were taken from this region do not exhibit any important anatomic
formation. Depending on the radiographic angle, mental foramen and enostosis
surrounding it can be seen (Fig.4B).
Mandibular premolar region
The radiographs which were taken from premolar region exhibit mental foramen between
the roots of the premolar, course of mandibular canal and sublingual fovea. Depending on
the radiographic projection angle, mental foramen may lead to diagnostic problem. It may
be seen as a periapical lesion (Fig.4C).
Mandibular molar region
The radiographs which were taken from molar region exhibit mandibular canal, mylohyoid
line, external and internal oblique line (Fig.4D) (Pasler,1993; Pasler&Visser, 2003).

Fig. 4. A: Periapical view and schematic drawing of mandibular anterior region B: Periapical
view and schematic drawing of mandibular canine region C: Periapical view and schematic
drawing of mandibular premolar region D: Periapical view and schematic drawing of
mandibular molar region


219
Legends for Figure 3-4:
1. Alveolar bone
2. Anterior nasal spine
3. Burn-out effect
4. Coronoid process
5. Genial tubercle
6. Incisive foramen
7. Infeior nasal concha
8. Lateral fossa
9. Lingual foramen
10. Mandibular canal
11. Mandibular tooth 1, central incisor
12. Mandibular tooth 2, lateral incisor
13. Mandibular tooth 3, canine
14. Mandibular tooth 4, first premolar
15. Mandibular tooth 5, second premolar
16. Mandibular tooth 6, first molar
17. Mandibular tooth 7, second molar
18. Mandibular tooth 8, third molar
19. Maxillary sinus
20. Maxillary sinus recession
21. Maxillary sinus septum
22. Maxillary tooth 1, central incisor
23. Maxillary tooth 2, lateral incisor
24. Maxillary tooth 3, canine
25. Maxillary tooth 4, first premolar
26. Maxillary tooth 5, second premolar
27. Maxillary tooth 6, first molar
28. Maxillary tooth 7, second molar
29. Maxillary tooth 8, third molar
30. Maxillary tuberosity
31. Mental foramen
32. Mental fossa
33. Mental ridge
34. Middle suture of hard palate
35. Mylohyoid ridge
36. Nasal septum
37. Nasal cavity
38. Processus hamularis
39. Soft tissue of nose
40. Submandibular fossa
41. Zygomatic arch
2.2 Panoramic radiography
Panoramic radiography, also known as an orthopantomogram, is a panoramic scanning
dental X-ray of the two-dimensional view of the jaws and their supporting structures from
ear to ear. It is obtained with patient, whose head stands between X-ray generator and the

Medical Imaging

220
film. The main advantage of panoramic radiography is the fact that it is clinically useful for
diagnostic problems associated with maxilla and mandible. One of the disadvantages of it is
that the images do not exhibit a fine anatomically detailed outfit gained from periapical
radiographs. Another problem related to orthopantomogram includes unequal
magnification (Lurie, 2004).
Indications of panoramic radiographies are included in the following cases:
Detection of the presence/absence of unerupted teeth
Evaluation of relationship of the upper posterior teeth with maxillary sinus
Evaluation of relationship of the lower posterior teeth with canalis alveolaris inferior
Suspicion of asymptomatic swellings
Radiographic examination of temporomandibular joint disturbances
Examination of odontogenic, nonodontogenic cysts and tumors
Evaluation of alveolar crest for insertion dental implants
Evaluation of maxillomandibular region following trauma
Examination of maxillary/mandibular surgical interventions
2.2.1 Anatomic landmarks of panoramic radiograph
While evaluating panoramic radiographs, first of all, normal anatomic structure of the
region must be known well. Complicated structure of the regions, superposition of these
structures and variations of the projection orientations may lead to problems during the
evaluation process.
There are four diagnostic regions in the panoramic radiography:
Dentoalveolar Region
Maxillary Region
Mandibular Region
Temporomandibular, Retromaxillary and Cervical Region.
Dentoalveolar region
It is surrounded by maxillary sinus and inferior border of the nasal cavity from above and
mandibular canal from below. Frontal side of ramus takes place on its left and its right.
The teeth which are located in the upper and lower jaws and alveolus supporting them
are seen in this region. Caries, fillings and prostheses are evaluated for the teeth whereas
periodontal problems and intraalveolar pathologies related to the teeth are evaluated for
alveolus (Fig.5).
Maxillary region
It is surrounded by orbita from above and maxillary sinus and the inferior border of the
nasal cavity from below. Coronoid processus of the mandible and zygoma take place on
its left and its right. Maxillary sinuses, zygomatic complex, nasal cavity and conchae,
sphenoid, ethmoid, palate, frontal bones and pterygomaxillary fissure can be observed in
this region. Lefort fractures and maxillary sinus pathologies are evaluated in this region
(Fig.5).


221
Mandibular region
It is comprised of the mandibular teeth and mandibula rather than alveolus. Condylar and
coronoid processes, ramus, body and angle and symphysis take place in this region.
Mandibular canal, mental foramen, submandibular fossa, superimposed shadow of cervical
vertebrae, external oblique ridge, posterior surface of tongue, soft palate and uvula, floor of
nasopharynx and hyoid bone can also be observed in this region. Internal bone lesions and
fractures are evaluated (Fig.5).
Temporomandibular, retromaxillary and cervical region
It is surrounded by temporal bone from above, and hyoid bone from below. Anterior of the
ramus of the mandible takes place in its anterior. Cervical vertebra takes place in its
posterior. The most important anatomic formation in this region is temporomandibular joint
(TMJ). TMJ is comprised of glenoid fossa, articular eminence and articular process of
mandibular condyle. Cervial vertebra, ear lobe, soft palate and uvula, posterior pharyngeal
airway, floow of nasopharynx, zygomatic arch, styloid process of temporal bone,
pterygomaxillary fissure and maxillary tuberosity can be observed in this region. Fractures
in this region are evaluated (Fig.5) (Lurie, 2004).

Fig. 5. Panoramic radiograph with marked anatomic structures
Legends for Figure 5:
2. Articular tubercle of the temporal bone
3. Cervical vertebra
4. Coronoid process
5. Dorsum of tongue (Shadow)
6. Ear lobe
7. Epipharynx

Medical Imaging

222
8. External auditory canal
9. External oblique ridge
10. Hard palate
11. Hyoid bone
12. Incisive foramen
13. Inferior nasal concha
14. Inferior border of mandible
15. Infraorbital canal
16. Mandibular angle
18. Mandibular condyle
19. Maxillary sinus
21. Mental foramen
22. Nasal septum
23. Nasal cavity
24. Nasopalatine canal
25. Orbital rim
26. Pterygoid process of sphenoid bone
27. Pterygopalatine fossa
28. Sigmoid notch
29. Soft palate
30. Styloid process
31. Submandibular fossa
32. Zygoma
33. Zygomatic arch
2.3 Dental computed tomography
Computed tomography was discovered by Hounsfield in 1974. After improvements,
nowadays, dental computed tomography is performed for diagnosis of oral and
maxillofacial pathology in most patients. Its advantage over 2D radiography is the fact that
it can eliminate the superimposition of images of adjacent tissues. Since it provides bone
images at the highest quality, it is the most widely used imaging technique (Curtain et al.,
1998; Karjodkar,2006). Tomographic images are taken as trans-axial cross sections. These
images are stored on the computer and then recreated from the cross sections passing
through the surfaces which are desired to be observed. This is called multiplanar
reformation. This way, axial, sagittal and coronal planes of the material that was imaged can
be obtained. When these planes are combined by means of a software application, a 3D
image may also be obtained. The images are obtained with the patient supine and during
quite respiration. Contrast agent injection may be needed to evaluate soft tissues. When
taking a computed tomography of oral and maxillofacial region, images are acquired from
the top of the frontal sinus to the sub mental region (Hermans et al., 2006). Computerized
tomography is used in maxillofacial surgery, reconstructive surgery, orthognatic surgery,
dental implant applications, and detection of lesions like cyst/tumor, trauma and
temporomandibular joint diseases.


223
Dental computed tomography has a number of advantages over other conventional
radiography:
Undesired superimposition of other tissues in the region is eliminated.
Thanks to the high-resolution of computerized tomography, differences between the
tissues with different physical densities can be distinguished better.
It is possible to obtain images of the tissues which are located on axial, coronal
sagittal planes.
It is especially a very useful tool for the planning of dental implant insertion.
It has no magnification and no distortion.
In the presence of formations like cysts/tumors, it can be determined whether this
formation has a solid or a liquid structure by means of density measurements
(Frederiksen, 2004).
Dental computed tomography has also disadvantages over other conventional radiography:
Administration of contrast agent is necessary for imaging soft tissue
More radiation exposure
Degradation of image quality by metallic objects, like as dental crown, fillings
2.3.1 Anatomic landmarks of dental tomography
This chapter presents the imaging of normal anatomic structures by dental tomography on
axial, coronal, sagittal planes and in 3D view (Fig.6,7,8).

Fig. 6. 3D CT anatomy of the facial skeleton

Medical Imaging

224

Fig. 7. (A,B). Axial CT anatomy of the facial skeleton; (C,D). Axial CT anatomy of the facial
skeleton


225

Fig. 8. (A,B). Coronal CT anatomy of the facial skeleton; (C,D). Coronal CT anatomy of the
facial skeleton; (E,F). Coronal CT anatomy of the facial skeleton
Figure legends for figure 6,7,8:
2. Carotid canal
3. Concha bullosa
4. Coronoid process
5. Crista galli
6. Dens axis
7. Ethmoid sinus
8. External auditory canal
9. Foramen ovale
10. Foramen rotundum
11. Frontal bone
12. Frontal sinus
13. Genial tubercle of mandible
14. Glenoid fossa
15. Greater palatine canal

Medical Imaging

226
16. Greater wing of sphenoid bone
17. Hamulus of medial pterygoid plate
18. Hard plate
19. Hyoid bone
20. Incisive canal
21. Incisive foramen
22. Inferior meatus
23. Inferior orbital fissure
24. Infraorbital canal
25. Infratemporal fossa
26. Lacrimal bone
27. Lateral pterygoid plate
28. Lesser palatin canal
29. Mandible
31. Mandibular condyle
32. Mandibular foramen
33. Mandibular notch
34. Mandibular ramus
35. Maxilla
36. Maxillary sinus
38. Medial pterygoid plate
39. Mental foramen
40. Middle meatus
41. Middle suture of hard palate
42. Nasal bone
43. Nasal cavity airway
44. Nasal septum
45. Nasofrontal suture
46. Nasolacrimal canal
47. Nasopharynx
48. Orbit
49. Oropharynx
50. Parapharyngeal space
51. Perpendicular plate of ethmoid bone
52. Pterygoid fossa
53. Pterygoid process of sphenoid bone
54. Pterygopalatine fossa
55. Sphenoid bone
56. Sphenoid sinus
57. Sphenoid sinus septum
58. Sphenozygomatic suture
59. Styloid process
60. Submandibular space
61. Submandibular gland
62. Tongue
63. Trigomun retromolare
64. Uvula


227
65. Vomer
66. Zygoma
67. Zygomatic arch
3. Radiographic description of oral and maxillofacial pathology
3.1 Radiolucent/radiopaque lesions of the jaws
Odontogenic cysts and tumors present problems of diagnosis, radiology and
histopathology. In general, their differential diagnosis requires radiographic clinical data,
since many of them possess similar histological characteristics. Radiologic appearance of
jaw cysts and odontogenic tumors varies considerably. The common lack of physical
findings and the development of most of these lesions within the confines of the bone
make radiologic investigation and interpretation uniquely important. Radiographs are
also important in treatment planning for surgical removal. They can evaluate
encroachment on vital structures, extent into soft tissue, size of the lesion, and
requirements for reconstruction. Radiography allows for creation of a radiologic
differential diagnosis. (Escobar et al.,2007)
3.1.1 Radiolucent lesions of the jaws
Dental granuloma
Radicular cyst
Dentigerous cyst
Keratocystic odontogenic tumor
Ameloblastoma
Incisive canal cyst
Simple bone cyst
Central giant cell granuloma
Odontogenic myxoma
3.1.2 Radiopaque lesions of the jaws
Odontoma
Torus
Osteoma
Osteochondroma
Cementoblastoma
Fibrous dysplasia (late stage)
3.1.3 Mixed radiolucent/radiopaque lesions of the jaws
Fibrous dysplasia (early stage)
Ossifying fibroma
Cemento-osseous dysplasia
Chronic osteomyelitis
Osteosarcoma
Metastasis

Medical Imaging

228
Dentigerous cyst
Dentigerous cysts are the second most common odontogenic cysts after radicular cysts. It
surrounds the crown of an impacted tooth, caused by fluid accumulation between the
reduced enamel epithelium and the enamel surface, resulting in a cyst in which the crown
is located within the lumen and roots outside. It is usually asymptomatic but produces
some swelling or pain when become large or inflamed. It is associated clinically with
impacted tooth most commonly an unerupted 3
rd
molar (mandibular- more than
maxillary), then maxillary canines, rarely involve deciduous teeth. Radiographically, the
dentigerous cyst appears as a unilocular radiolucency of variable size with well-defined
sclerotic borders, associated with the crown of an unerupted tooth. In an infected cyst the
borders may be ill-defined. The radiographic appearance of such a cyst, though quite
typical, is not diagnostic (Daley&Wysocki, 1995). The treatment of dentigerous cysts is
determined by the size of the lesion. Small lesions should be removed by surgery; larger
cysts are treated by marsupialization or decompression. The possible complications of the
dentigerous cysts are the permanent bony deformation from its expansive destruction of
bone, loss of essential permanent dentition or its innervation of the mandibular nerve.
Dentigerous cysts with long evolution can present areas with keratin or dysplasic changes
of its epithelial revetment with development of an ameloblastoma or an epidermoid
carcinoma (Fig.9) (Weber, 1993).

Fig. 9. Panoramic radiograph shows a well-defined expansile radiolucent lesion in the right
mandible and unerupted right canine.
Radicular cyst
Radicular cysts are the most common cyst of the jaws. They are most frequent between the
ages of 20 and 60 years. Radicular cysts may cause slowly progressive painless swellings,
with no symptoms until they become expansion of the cortical plates. If the infection enters,
the tooth and swelling develop all the painful symptoms of an abscess. Initially, the swelling


229
is rounded and hard. Later they are caused the demolition of cortical plate, than the swelling
is rubbery and fluctuant because of the cyst fluid. Large cysts may involve a complete
quadrant with some of the teeth occasionally mobile and some additional pulps nonvital.
Radiographically, radicular cyst appears well defined radiolucent area. Infection of a cyst
causes resorption of the surrounding tissue. If the cyst extends slowly, a condensed
radiopaque periphery is present. Enucleation is usual method for the treatment of radicular
cyst. Larger cysts are treated by marsupialization (Fig.10) (Cawson&Odell, 2002; Wood et
al.,1997; Sahin et al., 2009).

Fig. 10. Radiographic appearance of the radicular cyst (maxillary right second molar region)
on the panoramic radiograph
Keratocystic odontogenic tumor
The most recent classification of the World Health Organization (WHO) reallocated
keratocyst (keratinized primordial cyst) within the classification of odontogenic tumours
under the term keratocystic odontogenic tumor. It has specific histopathological
features and clinical behavior, it makes up to 10- 20% of all developmental odontogenic
cyst (Barnes et al.,2005). Keratocystic odontogenic tumor occurs more in mandible 80%
(posterior body and ascending ramus). It grows in anterior-posterior direction within
medullary cavity of the bone without causing obvious bone expansion. This is useful to
differentiate clinical and radiographic dentigerous and radicular cyst of similar size which
produce bone expansion. Multiple keratocystic odontogenic tumor is seen in nevoid basal
cell carcinoma (gorlin syndrome). Radiographically, it is well defined radiolucent area
with smooth corticated margin. Large lesions in posterior body and ascending ramus of
the mandible appear as multilocular radiolucency. An unerupted tooth is involved in the
lesion in 25-40 % of cases. The treatment of keratocystic odontogenic tumor is complete

Medical Imaging

230
removed by enucleation and curettage it has tendency to recurred because of friable
nature of cyst wall that result in fragment or due to formation of new cysts from dental
lamina (daughter cysts). Marsupialization has been effective in reducing the size of the
cyst (Fig.11) (Mendes RA et al., 2010)

Fig. 11. Panoramic radiograph shows an ellipsoid, expansile, well-corticated, radiolucent
lesion in left mandibular body and ascending ramus.
Lateral periodontal cyst
Lateral periodontal cyst is rare asymptomatic lesion, mainly in mandible in canine-premolar
region. It is usually seen by chance in routine radiographs. Radiographically, it appears as a
well circumscribed radiolucent area located laterally to the roots of vital tooth. Occasionally
this cyst appears as multilocular (poly cystic) named botryoid odontogenic cyst. The
radiographical picture is not diagnostic. Lateral periodontal cyst should be enucleated. If the
affected tooth is healthy, it can be retained (Fig.11) (Cawson&Odell, 2002).

Fig. 12. Periapical view of lateral periodontal cyst A. Pre-treatment B. Post-treatment


231
Ameloblastoma
Ameloblastoma are most common tumors of jaws. The majority of ameloblastoma are
benign, with less than 1% showing malignant behavior. The most common site of
ameloblastoma is the ascending ramus and proximal body of the mandible (80%). Based
on radiological appearance, ameloblastomas are divided into two subtypes. Multicystic
ameloblastomas account for approximately 85% of all ameloblastomas and occur in the
third to seventh decades of life. On radiographs, it is typically form rounded, cyst-like,
radiolucent area appear multilocular. There is marked buccolingual cortical expansion
with internal osseous septae, giving rise to a soap bubble appearance. Tooth
displacement or root resorption may occur. Unicystic ameloblastomas occur in a younger
age group and tend to be non-invasive. They present as a well-circumscribed, unicystic,
radiolucent lesion, mostly in the region of the mandibular third molar (DelBalso, 1998).
Treatment is by wide excision, preferably taking up to 2 cm of apparently surrounding
normal tissue. Unicystic ameloblastomas can be enucleated with low risk of recurrence
(Cawson&Odell, 2002).

Fig. 13. Panoramic radiographic view of ameloblastoma of mandibular ramus.
Central giant cell granuloma
Central giant cell granulomas (CGCG) typically occur in patients younger than 30 years of
age, more often in females. The lesion is more common in the anterior part of mandible with
a tendency to cross the midline. In the early stage, the lesion manifests as a small unilocular
lucent lesion. However, with development, it appears multilocular with fine trabeculae.
They may cause a variable degree of bony expansion, divergence of roots and root
resorption. Brown tumour of hyperparathyroidism can mimic CGCGs radiologically as well
as pathologically; however, the patients age, radiological changes in other bones, and
biochemical findings help in differentiation (Altug et al., 2011b).

Medical Imaging

232

Fig. 14. Panoramic radiographic view of CGCG of mandibula.
Odontoma
Odontomas are considered to be a hamartomatous lesion rather than a neoplasm. Most
cases are diagnosed in the second decade of life, and are usually associated with an
impacted tooth. Radiologically, it is seen as a radiopaque mass surrounded by thin
radiolucent space. The compound odontomas are composed of multiple well-formed teeth
whereas the complex odontomas appear as an irregular calcified tissue. A related but very
rare lesion is ameloblastic fibro-odontoma. Most cases occur in young males involving
posterior jaws and may expand into the ramus. The amount of radiolucent internal structure
exceeds the odontomas component (Weber et al., 1993; Altug et al, 2010b).

Fig. 15. Panoramic radiographic view of odontoma of mandibular canine region.


233
Ossifying fibroma
Ossifying fibroma is rare. It is slow expansile growth and it can expand the cortices and displace
adjacent structures. True benign tumors of mesenchyme are with strong predilection for tooth-
bearing sections of jaw. Patients are usually females in 20-40s. Radiographically, early lesion is
radiolucent with varying degrees of calcification and has well circumscribed margins. If the
lesion is more opaque shows that the lesion is mature. The patterns of calcifications have no
effect on the lesions behavior. Small lesions should be removed by enucleation and curettage.
However, large lesions (5mm margin) which have distorted the jaw require local resection.
Recurrence is rare (Fig. 16) (Cawson & Odell, 2002; Ortakolu et al., 2006).

Fig. 16. Axial CT view of maxillary ossifying fibroma
Cementoblastoma
Cementoblastoma is a benign neoplasm and forms a mass of cementum and cementum-like
tissue on roots of teeth. It is usually affect of posterior region of mandible, especially
mandibular first molar. Clinically, cementoblastoma mainly affects young adults,
particularly males. It is slow-growing and usually asymptomatic; but pain and swelling
have been reported in a number of cases. Radiographically, there is typically a radiopaque
mass with a thin radiolucent margin with the root of the tooth. The mass may be rounded or
irregular in shape. Resorption of related tooths roots is common, but the tooth remains vital
(Fig.17) (Sumer M et al., 2006).

Medical Imaging

234

Fig. 17. Panoramic radiographic view of cementoblastoma related to mandibular premolar
tooth.
Focal osseous-dysplasia
The current classification of fibro-osseous lesions, released in 2005 by the World Health
Organization (Barnes et al.,2005), is based on age, sex and histopathologic, radiographic and
clinical characteristics, as well as location of the lesion. Focal osseous dysplasia is an
asymptomatic benign malformation and belonging to the group of fibro-osseous lesions.
Although the etiology and pathogenesis of focal osseous-dysplasia are unknown,
histogenetically it is believed that it originated from periodontal ligament. It usually appears in
dentate and/or edentulous posterior mandibular region. Radiographically, the lesion is well
defined by radiolucent borders and an unilocular dense radiopaque appearance. No treatment
is necessary for focal osseous dysplasia and follow-up is essential to confirm the diagnosis.
Some authors have suggested that the transformation into florid cemento-osseous dysplasia is
possible and should be considered at recall visits (Fig.18) (Summerlin &Tomich, 1994).

Fig. 18. Periapical radiographic view of focal osseous dysplasia in mandible


235
Odontogenic myxoma
Odontogenic myxoma derives from dental mesenchyme and generally affects the young
persons (of average age 15 years). It usually affects in the mandible, and typically appears
in the mandibular angles. It has benign behaves but can infiltrate widely. Due to the
inadequate surgical excision, it may recur. Radiographically, the odontogenic myxoma
may produce several types: unicystic, multilocular, pericoronal and radiolucent-
radiopaque and it is close resemblance to soap bubble-like picture of ameloblastoma. The
lesion gives rise to fusiform swelling and radiolucent area with scalloped margins. The
treatment of odontogenic myxoma is required wide excision but some cases have been seen
over 30 years after the first intervention (Fig.19) (Cawson & Odell, 2002; Wood et al., 1997;
Altug et al., 2011a).

Fig. 19. Axial CT view of maxillary odontogenic myxoma.

Medical Imaging

236
Solitary eosinophilic granuloma
Solitary eosinophilic granuloma of the jaws belongs to the group of langerhans cell
histiocytosis and causes restricted bone destruction with swelling and pain. It generally
occurs in child and young adults. Eosinophilic granuloma can affect almost any bone but,
when the jaws are affected mandible is commonly. Radiographically, the lesion has rounded
radiolucency and an appearance of teeth floating air. The treatment of eosinophilic
granuloma curettage is sufficient and some localize cases spontaneous regression are
possible (Fig.20) (Cawson & Odell , 2002; Altu et al.,2010a)

Fig. 20. Coronal CT view of mandibular eosinophilic granuloma.
4. Conclusion
Although oral radiology is the precious member of oral diagnosis procedures, only one
imaging modality can provide us to wrong diagnosis in maxillofacial region. Especially in
cysts/tumors differential diagnosis, it is recommended that combination of different
imaging modalities.


237
5. Acknowledgment
We are grateful to Mrs. Muge Unlukoc, has provided invaluable help throughout including
providing us with periapical radiographic illustrations. We would like to thank Mr. Hakan
Yamanyar, for English proof-reading.
6. References
Altu, HA, Almerolu, M, ahin, S, enimen, M, Doan, N. & Krshnan, DG. (2010a).
Incidental discovery asymptomatic radiolucent lesion of the posterior mandible. J
Oral Maxillofac Surg, Vol.68. pp:845-848 ISSN 0278-2391
Altug, HA, Altug, H, Sari, E, Sencimen, M. & Altun C. (2010b). Diagnosis and surgically
management of supernumerary teeth in both the primary and the permenant
dentitions. The Journal of Gazi University Faculty of Dentistry. Vol.27. pp:77-82 ISSN
1300-3100
Altu, HA, Glses, A & enimen M. (2011a). Clinico-radiographic Examination of
Odontogenic Myxoma with Displacement of Unerupted Upper Third Molar:
Review of the Literature. Int J Morphol 29(3):930-933 ISSN 0717 9367
Altu, HA, enimen, M, Altun, C & Guven G. (2011b). Computed tomography and
magnetic resonance imaging of giant cell granuloma of the mandible. Int J Comp
Asist Radiol Surg. 6:s384-s385 ISSN 1861 6410
Barnes, L, Eveson, JW, Reichart, P. & Sidransky, D. (2005). Classification of Tumors In:
Pathology & Genetics Head and Neck Tumours. IARCPress, ISBN 92 832 2417 5, Lyon,
France
Cawson, RA. & Odell, EW. (2002). Hard Tissue Pathology, In: Cawsons Essentials of Oral
Pathology and Oral Medicine. Seventh edition, 102-158, ISBN 0 443 071055, Spain
Curtin, HD., Ishwaran, H., Mancuso, AA., Dalley, RW., Caudry, DJ. & McNeil, BJ. (1998).
Comparison of CT and MR imaging in staging of neck metastases. Radiology,
Vol.207(1):123130, ISSN 0033-8419
Daley, TD. & Wysocki, GP. (1995). The small dentigerous cyst. A diagnostic dilemma. Oral
Surg Oral Med Oral Pathol Oral Radiol Endod, Vol.79, pp:77-81,101-120 ISSN 1079-
2104
DelBalso, AM. (1998). An approach to the diagnostic imaging of jaw lesions, dental
implants, and the temporomandibular joint. Radiol Clin North Am. Vol. 36, pp:855-
90 ISSN 0033 8389
Escobar, E, Godoy, L. & Peafiel, C. (2007). Odontogenic Cysts: Analysis of 2.944 cases in
Chile Germn Ochsenius. Med Oral Pathol Oral Cir Bucal, Vol.12,pp:E85-91, eISSN
1698-6946
Frederiksen, NL. (2004). Specialized Radiographic Techniques, In: Oral Radiology: Principles
and Interpratation, White, SC.&Pharoah, MJ, (Ed), 245-250, Mosby, ISBN 978-0-323-
02001-5, China
Hermans, R., De Keyzer, F.& Vandecaveye V. (2006). Imaging Tecqnique, In: Medical
Radiology, Diagnostic Imaging, A.L. Baert, K.Sartor, (Ed),31-43, Springer-Verlag,
ISBN 3-540-22027-5, Germany
Karjodkar, FR. (2006). History of Radiology, In: Textbook of Dental and Maxillofacial
Radiology,1-11, Jaypee Brothers Medical Publishers, ISBN 81-8061-854-4, India

Medical Imaging

238
Lurie, AG. (2004). Panoramic Imaging, In: Oral Radiology: Principles and Interpratation, White,
SC.&Pharoah, MJ, (Ed), 191-195, Mosby, ISBN 978-0-323-02001-5, China
Mendes, RA, Carvalho, JF. &Van der Waal, I. (2010). Characterization and management of
the keratocystic odontogenic tumor in relation to its histopathological and
biological features. Oral Oncol. Vol.46, No.4, pp:219-225, ISSN 1368 8375
Ortakolu, K, Aydntu, YS, Altug, HA, Oku, KM. & Gnhan . (2006). Benign
fibroosseous lesions. The Turkish J Dentistry Vol.65. pp:132-136 ISSN 1304-6071
Pasler, FA& Visser, H. (2003). Radiographic Anatomy in Intraoral Radiographs, In :
Pocket Atlas of Dental Radiology, 70-80, Thieme, 978-3-13-139801-7, USA
Pasler, FA. (1993). Radiographic Anatomy of Special Region, In: Color Atlas of Dental
Medicine.Radiology, K.H. Rateitschak, H.F. Wolf, (ed),71-80,Thieme, ISBN 3-13-
78901-6,USA.
ahin, S, Saygun, NI, anak, CF, ngr, . & Altug HA. (2009). Root canal treatment
failure mediated lateral radicular cyst: Case report. T Klin J Dental Sciences, Vol.15.
pp: 214-219 ISSN 1300-7734
Scholl, RJ, Kellett, HM, Neumann, DP. & Lurie, AG. (1999). Cysts and cystic lesions of the
mandible: clinical and radiologic-histopathologic review. Radiographics Vol.19.
pp:1107-24 ISSN 0271 5333
Sumer, M, Gunduz, K, Sumer, AP. & Gunhan O. (2006). Benign cementoblastom A case
report. Med Oral Patol Oral Cir Bucal. Vol. 11. pp:e483-4855. ISSN 1698-6946
Summerlin, DJ. & Tomich, CE. (1994). Focal cemento-osseous dysplasia: a clinico-pathologic
study of 221 cases. Oral Surg Oral Med Oral Patho Oral Radiol Endod, Vol. 78. pp:611
20, ISSN 1079-2104
Weber, AL. (1993). Imaging of the cyst and odontogenic tumors of the jaw. Definition and
classification. Radiol Clin North Am, Vol.31,pp:101-120, ISSN 0033 8389
Whaites, E. (2002). Periapical Radiograph, In: Essential of Dental Radiograhy and Radiology, 75-
100, Harcourt Publihers Limited, ISBN 0443-07027-X, China
White, SC.& Pharoah, MJ. (2004). Intraoral Radiographic Examinations, In :Oral Radiology :
Principles and Interpratation, 121-126, Mosby, ISBN 978-0-323-02001-5, China
Wood, NK., Goaz PW., Jacobs MC. (1997). Periapical Radiolucencies, In : Differential
Diagnosis of Oral and Maxillofacial Lesions, N.K. Wood, P.W. Goaz, (Ed), 252-
279,Mosby, ISBN 0-8151 9432-3, St Louis, USA
11
Fast MRI Methods for the Clinical
Evaluation of Skeletal Disorders
Renato Toffanin
1
, Giuseppe Guglielmi
2,3
and Maria A. Cova
4

1
Advanced Research Centre for Health, Environment and Space (ARCHES)
2
Dept. of Radiology, Casa Sollievo della Sofferenza, IRCCS
3
Dept. of Radiology, University of Foggia
4
Dept. of Radiology, University of Trieste
Italy
1. Introduction
Evaluation of specific magnetic resonance (MR) parameters of the skeletal system holds
great potential for the accurate clinical assessment of degenerative changes occurring in
bone and soft tissues at different anatomical sites. MR imaging of the water in a joint can
provide anatomical information about all the soft tissues within the synovial sac (articular
cartilage, meniscus, ligaments, synovial fluid) and surrounding it (muscles, tendons,
vascular structures). Importantly too, MRI of the water-plus-fat can provide information
pertaining both to the bone density and trabecular architecture. Recent developments
have led to combinations of scan protocols and image-measurement software such that
MRI can be used to evaluate the spatial distribution of specific relaxation parameters and
thus detect, assess, and quantify the many pathologic processes affecting the skeletal
tissues. In the articular cartilage of knee, for example, gradual deterioration of the
chondral tissue leads to progressive increases in the transverse relaxation time (T2) of the
water protons (David-Vaudey, 2004; Dunn et al., 2004; Apprich et al., 2010). Similarly,
bone loss in the calcaneus of patients with varying degrees of osteopenia and osteoporosis
causes a prolongation of the effective transverse relaxation time (T2*) of the bone marrow
protons (Wehrli et al., 1995; Damilakis et al, 2004). Nonetheless, standard scan protocols
for quantitative MRI are relatively slow and, therefore, not suitable for routine clinical
applications. Faster methods would highly enhance their applicability in the clinical
evaluation of skeletal disorders.
The purpose of this chapter is to provide a perspective on fast MRI methods for the non-
invasive assessment of the skeletal status and their relevance to the diagnosis of
osteoporosis and osteoarthritis, two major public health burdens (Hannan et al., 2001;
Theis et al., 2007). The emphasis lies on echo-planar imaging (EPI)-based sequences (Tsao,
2010) for the accurate evaluation of pathologic processes affecting bone marrow and
cartilage at different anatomical locations. A multi-shot EPI sequence is proposed for the
fast T2* mapping of the lumbar bone marrow while a gradient- and spin-echo (GRASE)
sequence is suggested for the fast T2 mapping of patellar articular cartilage. The
description of these fast acquisition techniques is followed by a presentation of two in

Medical Imaging

240
vivo feasibility studies on a clinical 1.5 T MRI scanner. These investigations demonstrate
that the proposed MRI methods can produce relaxation maps of specific skeletal sites in
just a few minutes, and with mean values comparable with data obtained using
conventional sequences. Their potential application in the clinical evaluation of
osteoporosis and osteoarthritis is also discussed.
2. Fast MRI techniques
Transverse relaxation is the result of random interactions at the atomic and molecular
levels (Abragam, 1961). This physical phenomenon is primarily related to the intrinsic
field caused by adjacent protons (spins) and hence is called spin-spin relaxation.
Transverse relaxation causes irreversible dephasing of the transverse magnetization.
There is also a reversible bulk field dephasing effect caused by local field
inhomogeneities, and its characteristic time is referred to as T2* relaxation. These
additional dephasing fields come from the main magnetic field inhomogeneity, the
differences in magnetic susceptibility among various tissues or materials, chemical shift,
and gradients applied for spatial encoding (Mugler, 2006). This dephasing can be
eliminated by using a 180 pulse, as in a spin-echo sequence. Therefore, in a spin-echo
sequence, only the true T2 relaxation is seen. In gradient-echo sequences, there is no 180
refocusing pulse, and these dephasing effects are not eliminated. Hence, transverse
relaxation in gradient-echo sequences (i.e., T2* relaxation) is a combination of true T2
relaxation and relaxation caused by magnetic field inhomogeneities.

Fig. 1. A diagram showing the T2 and T2* relaxation decay curves
In order to obtain an accurate estimate of the transverse relaxation decay curves several
images obtained at different echo times are generally required. The gold standard for T2
acquisition is likely to be a single slice single echo sequence (i.e. spin-echo sequence),
repeated at several echo times, with long TR. A major improvement of the spin-echo
technique is represented by the turbo spin-echo (TSE) sequence, of which a simplified
diagram is depicted in Fig. 2.

Fast MRI Methods for the Clinical Evaluation of Skeletal Disorders

241

Fig. 2. A simplified diagram for the turbo spin-echo pulse sequence. RF: radiofrequency
pulse, SS: slice selection gradient, PE: phase encoding gradient, FE: frequency encoding
gradient
This sequence is based on multi-echo multi-shot (MEMS) (Mehlkopf et al., 1984) and rapid
acquisition with relaxation enhancement (RARE) (Hennig et al., 1986) sequences and
provides T2-weighted images at fractions of the acquisition time of the conventional spin-
echo images. By applying multiple refocusing 180 RF pulses after the first echo, additional
spin echoes can be generated. Between each successive echo, the phase-encoding gradients
can be used to prepare the spins for different lines in k-space. Thus, multiple lines in k-space
can be sampled per excitation. Each echo in the readout train is progressively weaker, as
defined by the T2 decay.
Another method of decreasing image acquisition time is by echo-planar imaging (EPI)
(Mansfield, 1977). In EPI multiple lines of k-space are acquired through a multiple-echo
readout. However, in EPI signals are produced by rapid switching of gradient polarity in
place of the slower selective 180 RF pulses. In this way, EPI can produce an image in less
than 100 ms. However, in EPI sequences, since the multiple echoes are refocused by
gradients and not by 180 pulses, there is more effect of T2* decay and other artefacts.
Therefore, both spin-echo and gradient-echo EPI sequences may be applied for the fast
evaluation of the T2* relaxation.

Fig. 3. A simplified diagram for the GRASE (gradient- and spin-echo) pulse sequence

Medical Imaging

242
By combining the TSE and EPI methods, the GRASE (gradient- and spin-echo) sequence
(Fig. 3) uses a train of refocusing 180 pulses, but for each spin-echo of the readout, there are
additional gradient recall echoes (Feinberg & Oshio, 1991; Oshio & Feinberg, 1991). In this
sequence, each successive spin-echo is progressively weaker, as defined by T2 decay
whereas the strength of the gradient recalled echoes surrounding the spin-echo is defined by
the T2* decay envelope. By combining spin-echoes and short gradient-echo trains, the
GRASE technique overcomes several potential problems of EPI, including large chemical
shift, image distortions and signal loss from field inhomogeneity.
3. MRI of trabecular bone
Even though bone cannot be evaluated with most of the available MRI techniques in that
they are unable to generate sufficient signal, new quantitative MRI approaches are used to
study trabecular bone density and structure (Wehrli et al., 2006; Majumdar, 2008). MRI
can be used to evaluate trabecular bone in a number of skeletal sites, indirectly via the
protons of the bone marrow. Indeed, the presence of the trabecular bone matrix affects the
signal intensity of bone marrow, an effect that is particularly pronounced with certain
MRI sequences. With respect to gradient-echo acquisitions, static magnetic field
inhomogeneities produced by the difference between trabecular bone and neighbouring
bone marrow cause a more rapid decay of the MRI signal, which can be quantified by
measuring T2*. Pioneering studies have shown that T2* is correlated with trabecular bone
density (Davis et al., 1986; Rosenthal et al., 1990), and therefore, the effective transverse
relaxation (T2*) is shorter in normal trabecular bone than in the less dense trabecular
structures of osteoporotic bone tissue. It has also been shown that bone marrow T2*
reflects the orientation of the trabeculae and correlates with their mechanical strength
(Chung et al., 1993; Jergas et al., 1995). These characteristics make MRI a fundamental tool
in evaluating the quality of spongy bone and increase the ability of the technique not only
in identifying occult fractures but also in making possible a more accurate prediction of
fracture risk.
T2* relaxometry has been conducted at several sites of both axial and peripheral skeleton
(Funke et al., 1994; Grampp et al., 1995; Link et al., 1998). The preferred site for quantitative
MRI studies is the calcaneus in that it is mostly composed of spongy bone (95%). Therefore
quantitative MRI of the calcaneus is extremely sensitive in identifying changes in bone
quality that are not revealed by bone mineral densitometry. In one MRI study at 1.5 T that
examined 68 women with different degrees of vertebral deformity (Wehrli et al., 2002), it
was demonstrated that of the various areas of the calcaneus examined, the subtalar region
was best able to discriminate patients with fracture from those without. The authors of this
study also demonstrated that the R2* (1/T2*) is sensitive to changes in bone quality that
were not identified with BMD.
Trabecular bone is also prominent in the vertebral body (up to 90%). The spine certainly
represents the most critical skeletal site for quantitative MRI since vertebral fractures are the
most common type of osteoporotic fractures (Wasnisch, 1999). In one MRI investigation
done at 1.5 T on a group of 54 postmenopausal women, T2* mapping of the lumbar spine
was shown to be capable of differentiating between healthy subjects and subjects with low
energy fractures (Damilakis et al., 2004).


243
The use of T2* relaxometry can certainly promote the application of quantitative MRI in
the diagnosis of osteoporosis. Nonetheless, MRI protocols commonly applied to estimate
T2* in bone marrow are relatively slow and, therefore, not suitable for routine clinical
application. In one recent study on the calcaneus of six healthy volunteers, Toffanin et al.
(2006) demonstrated the possibility of ultrafast T2* mapping of the bone marrow both at 1.5
and 3 T. To obtain an accurate estimate of T2* at 3.0 T or higher magnetic fields, corrective
measures may be required during postprocessing to minimise local field variations (B0)
responsible for signal loss and consequent overestimation of the R2* relaxation rate
(1/T2*). In the method proposed by Dahnke and Schaeffter (2005), the main field
heterogeneity is derived from T2* calculated on more than one slice and is used as an
initial value for interactive optimisation, with which the relaxation signal is corrected for
each voxel.
3.1 Fast T2* mapping of the lumbar bone marrow
The feasibility of a multi-shot gradient-echo EPI sequence for the fast T2* mapping of the
lumbar bone marrow was evaluated by our research team on a commercial clinical 1.5 T
MRI scanner located in the Department of Radiology of the Cattinara Hospital at the
University of Trieste. The MRI trial was performed on 21 subjects (8 males and 13 females)
referred to the hospital for low back pain. Five slices were acquired to image the lumbar
spine in the sagittal plane and the L2 vertebral body in the axial plane. In both cases, a fast
field-echo (FFE) multi-shot EPI sequence was applied with removed blip gradients in order
to apply the same phase encoding to all gradient echoes The overall examination time was
approximately 5 minutes. The main acquisition parameters are summarised in Table 1.

Sagittal plane Axial plane
TR 400 ms 400 ms
TEmin 2.0 ms 2.0 ms
TEmax 15.8 ms 15.8 ms
EPI factor 25 25
Flip angle 30 30
FOV 300 mm 300 mm 200 mm 200 mm
Matrix 320 320 224 224
Slice thickness 5 mm 5 mm
No. of slices 5 5
Table 1. Acquisition parameters of the fast field-echo multi-shot EPI sequence used for fast
T2* mapping of the lumbar bone marrow at 1.5 T

Medical Imaging

244
Estimation of the T2* relaxation time was performed on one manually-defined region drawn
on the entire L2 vertebral body as shown in Fig. 4. The multi-shot EPI sequence produced
T2* maps with mean values comparable with previous data obtained with conventional
sequences (Fig. 5). The mean T2* measured in the sagittal plane (14.2 3.9 ms) was slightly
lower than that measured in the axial plane (14.7 3.9 ms) but no statistically significant
difference was observed (P < 0.05).

(a)

(b)
Fig. 4. T2* maps of the central slice of the L2 vertebra generated from sagittal (a) and axial
(b) images by means of a monoexponential fitting algorithm as described by Dahnke &
Schaeffter (2005). The mean T2* was measured over the entire vertebral body excluding the
cortical bone


245

Fig. 5. Overview of the T2* data previously obtained with conventional sequences
These results indicate that fast T2* mapping of the lumbar bone marrow is feasible on a 1.5 T
scanner. However, further studies are required to investigate the full potential of the
proposed approach in the clinical evaluation of osteoporosis.
4. MRI of articular cartilage
Articular cartilage, is one of the types of hyaline cartilage that persists throughout adult life.
Basically, it comprises chondrocytes incorporated in an extracellular matrix composed
mainly of water, collagen II fibrils and proteoglycans (Seibel et al., 2004). Despite its simple
appearance, this tissue hides various modifications in respect of the original cartilage that
make it a singular structure. The articular cartilage is, in fact, stratified and classically, four
distinct layers are described from the surface to the interior: tangential, transitional, radial
and calcified, respectively (Fig. 6). Both morphological and biochemical information can be
obtained by MRI, which is probably the most accurate imaging modality in evaluating the
state of hyaline cartilage (Disler et al., 2000; Cova & Toffanin, 2002). Apart from clinical MRI
protocols that depict cartilage morphology, there is a growing interest in developing
quantitative MRI approaches that are sensitive to its early structural changes (Burstein et al.,
2000; Mosher & Dardzinski, 2004).
Over the past years, quantification of the human articular cartilage has been performed
using T1, T1 and T2 relaxation time constants as well as the magnetization transfer ratio
(Toffanin et al., 2001; Menezes et al., 2004; Wheaton et al., 2005). One of the magnetic
parameter that is currently evaluated for studying cartilage damage is the transverse
relaxation time (T2), whose relaxation mechanism results dominated by the dipolar
interaction between water molecules and collagen (Mlynrik et al., 2004) In this regard,
dual-echo or multi-echo sequences are typically employed for quantitative T2 mapping.
Nonetheless, faster quantitative MRI techniques are required in order to allow the
introduction of T2 mapping in routine clinical protocols.

Medical Imaging

246

Fig. 6. Histological zones in hyaline cartilage. Collagen fibres are parallel to the surface in
the superficial (tangential) zone, curved in the intermediate (transitional) zone and
perpendicular to subchondral bone in the deep (radial) zone (From Cova & Toffanin, 2002.
Reprinted with permission).
4.1 Fast T2 mapping of the patellar articular cartilage
Recently, we have devoted particular attention to optimising specific quantitative MRI
protocols for the fast T2 mapping of knee cartilage. The focus was on the gradient- and spin-
echo (GRASE) sequence able to produce a set of T2-weighted images in less than 2 minutes.
Also this research study was conducted on a commercial clinical 1.5 T MRI scanner located
in the Department of Radiology of the Cattinara Hospital at the University of Trieste. The
feasibility of the proposed approach was assessed on 35 patients (21 males and 14 females)
with moderate degree of patellar osteoarthritis. (Quaia et al., 2008).
For each patient, transverse GRASE and TSE images of patellar cartilage were acquired
using the scan protocols summarised in Table 2.

GRASE TSE
TR 3,000 ms 3,000 ms
TE
min
15 ms 15 ms
TE
max
120 ms 120 ms
EPI factor 3 -
Turbo factor 8 8
FOV 80 mm 80 mm 80 mm 80 mm
Matrix 128 128 128 128
Slice thickness 3 mm 3 mm
No. of slices 10 10
Total scan time 1 min 51 s 5 min 52 s
Table 2. Acquisition parameters of the GRASE and TSE sequences used for fast T2 mapping
of the patellar articular cartilage at 1.5 T


247

Fig. 7. ah. Axial GRASE images of patella cartilage (TR/TE
min
-TE
max
: 3,000/15120 ms); a:
15 ms, b: 30 ms, c: 45 ms, d: 60 ms, e: 75 ms, f: 90 ms, g: 105 ms, h: 120 ms. There is a clear
decay in the signal intensity of the patellar articular cartilage at longer TE, which can be
visually observed. The T2 relaxation time constant was calculated from a linear least-square
fit to the logarithm of the image intensity data (From Quaia et al., 2008. Reprinted with
permission).

Medical Imaging

248
Estimation of T2 was performed on one manually-defined region drawn on the entire
patella cartilage as shown in Fig. 8.

Fig. 8. Axial GRASE image of patella cartilage (TR/TE: 3,000/15 ms). A manually defined
ROI is drawn on the entire patellar articular cartilage for the quantification of the global T2
relaxation time. The ROI includes the entire cartilage, encompassing both the deep cartilage
close to the subchondral bone and the superficial cartilage, and the edges of the joint surface
(From Quaia et al., 2008. Reprinted with permission).

Patient No. GRASE TSE Arthroscopic grading
1 44.010.0 43.58.0 2B
2 48.38.0 47.26.0 2B
3 25.59.0 26.37.5 0
4 38.17.0 38.46.0 1
7 45.26.0 46.37.0 3A
12 30.58.0 31.06.0 2A
18 35.58.0 37.28.0 1
20 39.27.0 38.56.0 2A
27 28.56.0 29.09.0 0
28 58.26.0 57.211.0 3A
35 61.211.0 64.310.0 3B
Table 3. Mean T2 values for patellar articular cartilage obtained from the GRASE and TSE
images of the patellar articular cartilage of selected patients together with the corresponding
arthroscopic grading
1
.

1
Osteoarthritis from anteroposterior radiographs was graded according to the KellgrenLawrence
scoring system (0=no osteoarthritic features; 1=minute osteophytes of doubtful importance; 2=definite
osteophytes without reduction of the joint space; 3=reduction of the joint space; 4=greatly reduced joint
space and sclerosis of the subchondral bone) (Kellgren & Lawrence, 1957).


249
In our series the GRASE sequence provided T2 values slightly lower than those obtained by
the TSE sequence in most patients. This may relate to T2* decay present in the GRASE
sequence. Indeed, a potential drawback of the GRASE sequence is the possible presence of
artefacts due to the echo-planar imaging readout module that may lead to pronounced
chemical shift, image distortion, and signal loss from magnetic field inhomogeneity.
Nonetheless, differences in T2 values within deep and superficial cartilage were in line with
those observed with other MR sequences (Van Breuseghem et al., 2004). They are related to
cartilage T2 anisotropy determined by the different direction of the collagen fibres in the
superficial and deep cartilage with respect to the static magnetic field (Mosher et al., 2001).
5. Conclusion
The above MRI methods can drastically reduce the scan time for measuring the spatial
distribution of specific relaxation parameters in bone and cartilage. They may become more
widely adopted, being applied either with other imaging techniques or in isolation, to better
evaluate skeletal disorders, identify early tissue degeneration, tailor therapeutic
interventions and follow treatment response.
6. Acknowledgment
This work was supported in part by a grant from the Italian Ministry for Education,
University and Research (5 per mille 2009).
7. References
Abragam, A. (1983). The Principles of Nuclear Magnetism (2
nd
edition), Oxford University
Press, ISBN 978-0-19-852014-6, New York
Apprich, S.; Welsch, G.H.; Mamisch, T.C.; Szomolnyi, P.; Mayerhoefer, M.; Pinker, K. &
Trattnig, S. (2010). Detection of degenerative cartilage disease: comparison of high-
resolution morphological MR and quantitative T2 mapping at 3.0 Tesla.
Osteoarthritis and Cartilage, Vol.18, No.9, pp. 1211-1217, ISSN 1063-4584
Burstein, D.; Bashir, A. & Gray, M.L. (2000). MRI techniques in early stages of cartilage
disease. Investigative Radiology, Vol.35, No.10, pp. 622-638, ISSN 0020-9996
Chung, H.; Wehrli, F.W.; Williams, J.L. & Kugelmass, S.D. (1993). Relationship between
NMR transverse relaxation, trabecular bone architecture, and strength. Proceedings
of the National Academy of Sciences U.S.A., Vol.90, No.21, pp. 1025010254, ISSN
0027-8424
Cova, M. & Toffanin, R. (2002). MR microscopy of hyaline cartilage: current status. European
Radiology, Vol.12, No.4, pp. 814-823, ISSN 0938-7994
Dahnke, H. & Schaeffter, T. (2005). Limits of detection of SPIO at 3.0 T using T2*
relaxometry. Magnetic Resonance in Medicine, Vol.53, No.5, pp 1202-1206, ISSN
0740-3194
Damilakis, J.; Maris, T.; Papadokostakis, G.; Sideri, L. & Gourtsoyiannis, N. (2004).
Discriminatory ability of magnetic resonance T2* measurements in a sample of
postmenopausal women with low-energy fractures. A comparison with phalangeal
speed of sound and dual X-ray absorptiometry. Investigative Radiology, Vol.39,
No.11, pp. 706-712, ISSN 0020-9996

Medical Imaging

250
David-Vaudey, E.; Ghosh, S.; Ries, M. & Majumdar, S. (2004). T2 relaxation time
measurements in osteoarthritis. Magnetic Resonance Imaging, Vol.22, No.5, pp. 673
682, ISSN 0730-725X
Davis, C.A.; Genant, H.K. & Dunham, J.S. (1986). The effects of bone on proton NMR
relaxation times surrounding liquids. Investigative Radiology, Vol.21, No.6, pp. 472
477, ISSN 0033-8419
Disler, D.G.; Recht, M.P. & McCauley, T.R. (2000). MR imaging of articular cartilage. Skeletal
Radiology, Vol.29, No.7, pp. 367-377, ISSN 0364-2348
Dunn, T.C.; Lu, Y.; Jin, H.; Ries, M.D. & Majumdar, S. (2004). T2 relaxation time of cartilage
at MR imaging: comparison with severity of knee osteoarthritis. Radiology, Vol.232,
No.2, pp. 592598, ISSN 0033-8419
Feinberg, D.A. & Oshio, K. (1991). GRASE (Gradient- and spin-echo) MR imaging: a new fast
clinical imaging technique. Radiology, Vol.181, No.2, pp. 597-602, ISSN 0033-8419
Funke, M.; Bruhn, H.; Vosshenrich, R.; Rudolph, O. & Grabbe, E. (1994). The determination
of the T2* relaxation time for characterizing trabecular bone. Rofo-Fortschritte auf
dem Gebiet der Rntgenstrahlen und der Bildgebenden Verfahren, Vol.161, No.1, pp. 58-
63, ISSN 1438-9029
Grampp, S.; Majumdar, S.; Jergas, M. Lang P, Gies, A. & Genant, H.K. (1995). MRI of bone
marrow in the distal radius: in vivo precision of effective transverse relaxation
times. European Radiology, Vol.5, No.1, pp. 43-48, ISSN 0938-7994
Guglielmi, G.; Biccari, N.; Mangano, F. & Toffanin, R. (2010). 3 T magnetic resonance
imaging of the musculoskeletal system. La Radiologia Medica, Vol.115, No.4, pp. 571-
584, ISSN: 0033-8362
Hannan, H.L.; Magaziner, J.; Wang, J.J.; Eastwood, E.A.; Silberzweig, S.B.; Gilbert, M.;
Morrison, R.S.; McLaughlin, M.A.; Orosz, G.M. & Siu, A.L. (2001). Mortality and
locomotion 6 months after hospitalization for hip fracture: risk factors and risk-
adjusted hospital outcomes. JAMA, Vol.285, No.21, pp. 2736-2742, ISSN 00987484
Hennig, J.; Nauerth, A. & Friedburg, H. (1986). RARE imaging: a fast imaging method for
clinical MR. Magnetic Resonance in Medicine, Vol.3, No.6, pp. 823-833, ISSN 0740-
3194
Jergas, M.D.; Majumdar, S.; Keyak, J.H.; Lee, I.Y.; Newitt, D.C.; Grampp, S.; Skinner, H.B.
& Genant, H.K. (1995). Relationship between young modulus of elasticity, ash
density, and MRI derived effective transverse relaxation T2* in tibial specimens.
Journal of Computed Assisted Tomography, Vol.19, No.3, pp. 472479, ISSN 0363-8715
Kellgren, J. & Lawrence, J. (1957). Radiological assessment of osteoarthritis. Annals of the
Rheumatic Diseases Vol.16, No.4, pp. 494502, ISSN 00034967
Link, T.; Majumdar, S.; Augat, P.; Lin, J.C.; Newitt, D.; Lane, N.E. & Genant, H.K. (1998).
Proximal femur: assessment for osteoporosis with T2* decay characteristics at MR
imaging. Radiology, Vol.209, No.2, pp. 531-536, ISSN 0033-8419
Majumdar, S. (2008). Magnetic resonance imaging for osteoporosis. Skeletal Radiology, Vol.37,
No.2, pp. 9597, ISSN 0364-2348
Mansfield, P. (1977). Multi-planar imaging formation using NMR spin-echo. Journal of
Physics C: Solid State Physics, Vol.10, pp. L55-L58, ISSN 0022-3719
Mehlkopf, A.F.; van der Meulen, P. & Smidt, J. (1984). A multiple-echo and multiple-shot
sequence for fast NMR-Fourier imaging. Magnetic Resonance in Medicine, Vol.1,
No.2, pp. 295-297, ISSN 0740-3194


251
Menezes, N.M.; Grey, M.L.; Hartke, J.R. & Burstein, D. (2004). T2 and T1(rho) MRI in
articular cartilage systems. Magnetic Resonance in Medicine, Vol.51, No.3, pp. 503
509, ISSN 0740-3194
Mlynrik, V.; Szomolnyi, P.; Toffanin, R.; Vittur, F. & Trattnig, S. (2004). Transverse
relaxation mechanisms in articular cartilage. Journal of Magnetic Resonance, Vol.169,
No.2, pp. 300-307, ISSN 1090-7807
Mosher, T.J.; Smith, H.; Dardzinski, B.J.; Schmithorst, V.J. & Smith, M.B. (2001). MR imaging
and T2 mapping of femoral cartilage: in vivo determination of the magic angle effect.
AJR American Journal of Roentgenology, Vol.177, No.3, pp. 665669, ISSN 0361-803X
Mosher, T.J. & Dardzinski, B.J. (2004). Cartilage MRI T2 relaxation time mapping: overview
and applications. Seminars in Musculoskeletal Radiology, Vol.8, No.4, pp. 355-368,
ISSN 1089-7860
Mugler, J.P. III. (2006). Basic principles. In: Clinical Magnetic Resonance Imaging, Edelman,
R.R.; Hesselink, J.R.; Zlatkin, M.B.; Crues, J.V., Editors, pp. 2357, Elsevier, ISBN
978-0721603063, Philadelphia
Oshio, K. & Feinberg, D.A. (1991). GRASE (Gradient- and Spin-Echo) imaging: a novel fast
MRI technique. Magnetic Resonance in Medicine, Vol.20, No.2, pp. 344-349, ISSN
0740-3194
Quaia, E.; Toffanin, R.; Guglielmi, G.; Ukmar, M.; Rossi, A.; Martinelli, B. & Cova, M.A. (2008).
T2 mapping of the patellar articular cartilage with gradient and spin-echo magnetic
resonance imaging at 1.5 T: validation and initial clinical experience in patients with
osteoarthritis. Skeletal Radiology, Vol.37, No.6, pp. 511-517, ISSN 0364-2348
Rosenthal, H.; Thulborn, K.R.; Rosenthal, D.I.; Kim, S.H. & Rosen, B.R. (1990). Magnetic
susceptibility effects of trabecular bone on magnetic resonance bone marrow
imaging. Investigative Radiology, Vol.25; No.2, pp. 173178, ISSN 0020-9996
Seibel, M.J.; Robins, S.P. & Bilezikian, J.P. (Eds.). (2004). Dynamics of Bone and Cartilage
Metabolism, Academic Press, ISBN 978-0-12-088562-6, San Diego, USA
Theis, K.A.; Murphy, L.; Hootman, J.M.; Helmick, C.G. & Yelin, E. (2007). Prevalence and
correlates of arthritis-attributable work limitation in the US population among
persons ages 18-64: 2002 National Health Interview Survey Data. Arthritis &
Rheumatism, Vol.57, No.3, pp. 355-363, ISSN 1529-0131
Toffanin, R.; Mlynrik, V.; Russo, S.; Szomolnyi, P.; Piras, A. & Vittur, F. (2001).
Proteoglycan depletion and MR parameters of articular cartilage. Archives of
Biochemistry and Biophysics, Vol.390, No.2, pp. 235242, ISSN 0003-9861
Toffanin, R.; Cadioli, M.; Scotti, G. & Cova, M.A. (2006). Ultrafast T2* mapping of bone
marrow at 1.5 Tesla and 3.0 Tesla. Proceedings of the 14th Annual Meeting of the
International Society for Magnetic Resonance in Medicine, p. 3631, Seattle, Washington,
USA, May 6-12, 2006, ISSN 1545-4428
Tsao, J. (2010). Ultrafast imaging: principles, pitfalls, solutions, and applications. Journal of
Magnetic Resonance Imaging, Vol.32, No.2, pp. 252-266, ISSN 1053-1807
Van Breuseghem, I.; Bosmans, H.T.; Elst, L.V.; Maes, F.; Pans, S.D.; Brys, P.P.; Geusens, E.A.
& Marchal, G.J. (2004). T2 mapping of human femorotibial cartilage with turbo
mixed MR imaging at 1.5 T: feasibility. Radiology, Vol.233, No.2, pp. 609614, ISSN
0033-8419
Wasnich, R.D. (1999). Incidence Rates for Vertebral, Wrist & Hip Fractures in Women after
Age 50. In: Primer on the Metabolic Bone Diseases and Disorders of Mineral Metabolism.

Medical Imaging

252
(4
th
edition), Murray, F.J., Editor, pp. 257259, Lippincott, Williams & Wilkins,
ISBN 978-0781720380, Philadelphia
Wehrli, F.W.; Hilaire, L.; Fernndez-Seara, M; Gomberg, B.R.; Song, H.K.; Zemel, B.; Loh, L. &
Snyder, P.J. (2002). Quantitative magnetic resonance imaging in the calcaneus and
femur of women with varying degrees of osteopenia and vertebral deformity status.
Journal of Bone Mineral Research, Vol.17, No.12, pp. 22652273, ISSN 0884-0431
Wehrli, F.W.; Ford, J.C. & Haddad, J.G. (1995). Osteoporosis: clinical assessment wit
quantitative MR imaging in diagnosis. Radiology, Vol.196, No.3, pp. 631-641, ISSN
0033-8419
Wehrli, F.W.; Song, H.K.; Saha, P.K. & Wright, A.C. (2006). Quantitative MRI for the
assessment of bone structure and function. NMR in Biomedicine, Vol.19, No.7, pp.
731776, ISSN 0952-3480
Wheaton, A.J.; Dodge, G.R.; Borthakur, A.; Kneeland, J.B.; Schumacher, H.R. & Reddy, R.
(2005). Detection of changes in articular cartilage proteoglycan by T1(rho) magnetic
resonance imaging. Journal of Orthopaedic Research, Vol.23, No.1, pp. 102108, ISSN
0736-0266
0
Determination of Cardiac Ejection Fraction by
Electrical Impedance Tomography
Franciane C. Peters
1
, Luis Paulo da S. Barra
2
and Rodrigo W. dos Santos
2
1
Federal University of Rio de Janeiro
2
Federal University of Juiz de Fora
Brazil
1. Introduction
Cardiac ejection fraction is a clinical parameter that infers the efciency of the heart as a
pump. The ejection fraction of left ventricle (EFLV) and right ventricle (EFRV) are determined
separately. However, the clinical use of EFLV is more common and it is frequently called
ejection fraction (EF). By denition, the ejection fraction is calculated in the following way:
EF =
PV
EDV
=
EDV ESV
EDV
, (1)
where PV is the volume of blood pumped, that is given by the difference between the
end-diastolic volume (EDV) and the end-systolic volume (ESV). Cardiac ejection fraction
is a relevant parameter because it is highly correlated to the functional status of the heart.
To determine the EF, different non-invasive techniques can be used, like echocardiography,
cardiac magnetic resonance and computed tomography. However, they are not suitable
for continuous monitoring. In this work, we numerically evaluate a new method for the
continuous estimation of cardiac ejection fraction based on Electrical Impedance Tomography
(EIT).
EIT is a technique that reconstructs an image of the electrical resistivity inside a domain based
on protocols of current injection and potential measurement on the external boundary of
the domain. Detailed description about the development of this technique can be found,
for instance, in Cheney et al. (1999) and Holder (2005). The EIT has a large utilization
on geophysics and monitoring of industrial activities (Kim et al., 2004; Park et al., 2008), as
non-destructive technique to evaluate structures (Karhunen et al., 2009; Peters et al., 2010)
and on biomedical engineering (Barber & Brown, 1984; Brown et al., 1985; Mello et al., 2008;
Trigo et al., 2004). In this last area, the EIT has been developed to detect breast and other
kinds of cancer (Choi et al., 2007; Seo et al., 2004) and to monitor lung ventilation (Adler et al.,
1997; Lima et al., 2007), brain activity (Polydorides et al., 2002), heart activity (Peters et al.,
2009; Zlochiver et al., 2006), among others.
The special interest of the biomedical engineering in the development of the EIT is due to
its safety for monitoring long periods, since it is not based on ionizing radiation. On the
other hand, the EIT spacial resolution is not as high as the traditional imaging methods.
12
Nevertheless, its portability, low cost and time resolution are the main advantages of the
technique.
Mathematically, the EIT is classied as a non-linear inverse problem. Inverse because one
wants to nd the electrical resistivity of the body using measures of electrical potential on the
boundary excited by known electrical current. The forward (or direct) problem related to the
inverse one is to compute the electrical potential with known body resistivity and injected
current. The inverse problem is non-linear, what means that there is not a linear relation
between electrical resistivity and the electrical potential on the boundary. So, in general,
the solution process starts with an estimated resistivity distribution and such estimative is
iteratively adjusted in order to achieve an acceptable approximation for the actual resistivity
distribution.
Furthermore, the inverse problem is ill-posed and ill-conditioned. In general, the number
of parameters needed to represent the resistivity distribution is greater than the number of
potential measures. So, in order to treat the ill-posedness of the problem, some strategies of
regularization should be implemented. For instance, via the inclusion of previously known
information about the resistivity distribution in the solution of the inverse problem. The
problem is considered ill-conditioned because small perturbations in the measures can cause
a large change in the found resistivity distribution. So, the process of image generation by EIT
is very sensible to noise in the potential measures.
It is possible to see that many aspects are involved in the solution of the EIT problem and
some of them were discussed in previous works of the same authors. Barra et al. (2006a) and
Barra et al. (2006b) treat some computational aspects of the solution process. Peters & Barra
(2009) treats the special kind of regularization also adopted here. Peters & Barra (2010)
compares different measurement protocols and addresses the sensitivity of the process in
the presence of noisy measures. Peters et al. (2009) presents the viability analysis of a
biomedical application of the EIT. So the aimof this work is to revisit this particular biomedical
application, describing all the procedures involved in the generation of a computational
model based on cardiac magnetic ressonance images that allows the determination of the
cardiac ejection fraction by the EIT. Preliminary results are presented and the potentialities
and limitations of the proposed technique are discussed. The results suggest the proposed
technique is a promising diagnostic tool that offers continuous and non-invasive estimation
of cardiac ejection fraction.
2. Methods
Usual strategies to generate the EIT image is based on the discretization of the body in small
parts (Borsic et al., 2001). Each part has an unknown parameter, its resistivity. So, to get a good
image resolution, a great number of parameters is needed and the problembecomes ill-posed.
In this strategy, regularization techniques are applied and after solving the inverse problem,
the image obtained is modied by a limiarization process in order to get the nal image.
In this work we adopted a different strategy to generate the EIT images. In order to
monitor the cardiac ejection fraction, we assume that recent magnetic resonance images of
the patient are available. So, this previous information allows the use of a different kind of
parameterization in which the number of parameters is greatly reduced.
254 Medical Imaging
Determination of Cardiac Ejection Fraction by Electrical Impedance Tomography 3
In this section, all the methods used to solve the EIT problem will be described. First,
the parameterization based on magnetic resonance images will be explained, as well as the
resistivity model of the thorax. Second, the governing equations of the forward problem
will be introduced in addition to the numerical methods used to solve it. Third, the inverse
problemwill be formulated as a minimization problemand the adopted optimization method
will be presented. Finally, the computational experiments will be reported.
2.1 Two-dimensional models based on magnetic resonance images
2.1.1 Parameterization
From magnetic resonance (MR) images, the regions of interest, in this case the two ventricles,
were manually segmented in two different phases: end of the systole and the end of diastole.
Each curve of the segmentation was parameterized by a spline, with a minimum number of
points. The left ventricle spline has 7 control points and the right one 8 points. The external
boundary of the thorax and the boundaries of the lungs were also segmented. For simplicity,
the shape and size of the thorax and the lungs are assumed constant during the heart cycle.
Figure 1 shows a segmentation example.
Fig. 1. Manual segmentation of an MR image. The boundaries of the lungs, ventricle cavities
and thorax are represented by splines. LV and RV denotes left ventricle and right ventricle,
respectively.
The goal of our method is to recover the shape of the internal cavities of the heart,
presently considered in two dimensions, from electric potential measures. Therefore, with
two coordinates for each control point of the spline, the methods would need to estimate a
total of 30 ((7 + 8) 2) parameters. To reduce the number of parameters in half, another
parameterization is used. In this, only one parameter denes the position of each control
point.
During MRI segmentation we have used the same number of control points for the splines in
both systolic and diastolic phase. This allows us to restrict the search space forcing that each
control point i belongs to a line that connects its position at systole and diastole, as shown in
Fig. 2.
A linear interpolation, parameterized by a scalar t
i
, is used between the values of the
coordinates of each control point i. The spline relative to the end of systole can be obtained
255 Determination of Cardiac Ejection Fraction by Electrical Impedance Tomography
(a) Systole (b) Diastole
Fig. 2. The control points are represented by red crosses. The dashed lines are the search
space.
with t
i
= 0 for all i, and the one relative to the end of diastole with t
i
= 1 for all i. Doing so,
the method goal is to recover the shape of the ventricular cavities via the estimation of the 15
parameters t
i
, with i = 1, 2, ..., 15.
2.1.2 Splines
Splines are mathematical models that associate a curve with a set of points named control
points. Here we use a special type of spline called Extended Cross-Splines or Extended
X-Splines (Blanc & Schlick, 1995), for short. In this model, each control point i has the
coordinates (x
i
, y
i
) and an additional parameter s
i
[0, 1] that allows the curve C(t)
approximates (0 < s
i
1) or sharp interpolates (s
i
= 0) the control point. This feature is
very interesting because it allows the denition of smooth or sharp curves. In this work, just
smooth curves are represented.
Considering an Extended X-Spline model in which each segment depends on four control
points, a segment C(t) on the parameter range (t
k+1
, t
k+2
) is dened as follows:
C(t) = F
0
P
k
+ F
1
P
k+1
+ F
2
P
k+2
+ F
3
P
k+3
, (2)
where the blending functions F
k
are obtained by the normalization of the blending functions
A
k
(t) as follows:
F
k
(t) =
A
k
(t)
A
0
(t) + A
1
(t) + A
2
(t) + A
3
(t)
(3)
and their non null part can be divided in two parts, F
k
, dened between T
k
and T
k
, and F
+
k
,
dened between T
k
and T
+
k
.
The functions A
k
(t) are obtained by the generic quintic polinomial f
p
(u), whose coefcients
were determined in order to satisfy the constraints of the model, resulting the following
expression:
f
p
(u) = u
3

10 p + (2p 15) u + (6 p) u
2
. (4)
Hence, for the non null parts of the functions A
k
(t), we have:
A
0
(t) = f
p
k1
(u
0
) , A
1
(t) = f
p
k
(u
1
) , A
2
(t) = f
p
k+1
(u
2
) , A
3
(t) = f
p
k+2
(u
3
) (5)
256 Medical Imaging
with
u
0
=
t T
+
k
t
k
T
+
k
, u
1
=
t T
+
k+1
t
k+1
T
+
k+1
, u
2
=
t T
k+2
t
k+2
T
k+2
, u
3
=
t T
k+3
t
k+3
T
k+3
(6)
and
p
k1
= 2
t
k
T
+
k
2
, p
k
= 2
t
k+1
T
+
k+1
2
,
(7)
p
k+1
= 2
t
k+2
T
k+2
2
, p
k+2
= 2
t
k+3
T
k+3
2
.
(8)
The exibility of the model is improved by the introduction of new degree of freedom s
k
. In
each control point P
k
this new parameter affects the values T
+
k1
e T
k+1
as follows:
T
+
k1
= t
k
+ s
k
, T
k+1
= t
k
s
k
(9)
In other words, one can say that the parameter s controls the distance between the curve and
the control point, what allows the approximation or the interpolation of these points, as shown
in Fig. 3.
P
0
P
1
P
2
P
3
P
4
P
5
P
6
(a) Shape of the curve
0
0.5
1
0 1 2 3 4 5 6
F
(
t
)
t
F
0
F
1
F
2
F
3
F
4
F
5
F
6
(b) Blending functions
Fig. 3. The values of the parameters of each control point are: s
0
= 0, s
1
= 1, s
2
= 0, s
3
= 0,
s
4
= 0, s
5
= 1, s
6
= 0. These gures were inspired by the work of Blanc & Schlick (1995).
Although the expressions used to implement the Extended X-spline were rewritten above,
detailed description about this and other types of splines can be found in Blanc & Schlick
(1995).
2.1.3 Resistivity model
The proposed2Dmodel splits the domain in regions that represent different biological tissues,
heart cavities, lungs and torso, each one associated with a different electrical resistivity.
Grimnes (2008) presents the main factors that inuence the properties of biological tissues.
Although they may be classied in only four groups ( epithelium, muscle, connective tissue
and nervous tissue), the tissues can be divided in thirty kinds in accordance to their electrical
properties (Gabriel, 1996). In addition, the value of the resistivity of each tissue depends on
the frequency of the electrical current, on the temperature, on the presence of water, among
other issues.
In this work, we assume the resistivity of each tissue as known, constant and isotropic.
These are all simplied assumptions, since biological tissues are usually heterogeneous and
anisotropic. However, biological tissues are difcult to characterize, and the reported values
vary substantially in the literature. Table 1 presents some resistivity values for blood, heart
and lung found in the literature.
Tissue Resistivity (cm)
150 (Barber & Brown, 1984)
Blood 150 (Yang & Patterson, 2007)
100 (Schwan & Kay, 1956)
400 (Patterson & Zhang, 2003)
Heart 250 (Yang & Patterson, 2007)
400 - 800 (Baysal & Eyuboglu, 2000)
727 - 2363 (Barber & Brown, 1984)
Lung 1400 (Patterson & Zhang, 2003)
600 - 2000 (Baysal & Eyuboglu, 2000)
Table 1. Resistivity values of biological tissues that are found in the literature.
For the remaining tissues that compose the section of the thorax, called here torso region,
Bruder et al. (1994) proposes a mean resistivity of 500cm. The resistivity of the air is
10
20
cm, but the resistivity of the lung lled of air is difcult to determine. Rush et al. (1963)
presents a very simplied resistivity distribution model of the thorax characterized by the
presence of cavities lled of blood, surrounded by homogeneous material with resistivity ten
times greater. The same scheme, properly extended to include the lung regions, is used in this
work. Preliminarily, the resistivity of the blood is here taken as 100cm and the torso to be
1000cm. Two different values were tested for the resistivity of the lungs: 20000cm (Ratio
of Lung to Torso resistivity, RLT = 20) and 50000cm (RLT = 50).
2.2 Forward problem
The forward problem consists on calculating the electrical potential on the external boundary
of the torso that is generated by the current injection on a pair of electrodes. Figure 4 presents
the simplied model of the thorax.
Given that our 2D model has three regions with different but constant and isotropic
conductivities (heart cavities full of blood,
B
, lungs,
L
, and torso,
T
) the electrical
potential u at each point of the regions must satisfy Laplaces equation:
2
u(x) = 0 , x , (10)
the following boundary conditions:
1
T
u
n
= J
i
, x
ie
3
(11)
u
n
= 0 , x (
3
ie
3
) (12)
258 Medical Imaging
and these compatibility equations:
L
u =
T
u , x
1
(13)
B
u =
T
u , x
2
(14)
where =
L
+
B
+
T
,
1
is the interface between the lung and torso region,
2
is the
interface between the blood and the torso region,
3
is the external boundary of the thorax,
ie
3
is the part of
3
where the ith electrode is, J
i
is the electrical current injected on the i-th
electrode and
L
,
B
and
T
are the resistivities of the lung, blood and torso, respectively.
Fig. 4. The simplied thorax model. Here, the electrodes are represented in green. The
regions L represent the lungs, B the blood and T the torso.
2.2.1 Numerical solution of Laplaces equation
In order to solve Equation 10 for each subregion the Boundary Element Method (BEM) is
used. Further details about this technique can be found in Brebbia et al. (1984). The integral
equation of BEM for this problem is
c()u() +
(; x)u(; x)d(x) =
(; x)p(; x)d(x) , (15)

where is the collocation point, is the boundary of the sub-domain, u is the electrical
potential, p is its derivative, u
and p
are the fundamental solutions for the potential and

its normal derivative, respectively, and c() is a function of the boundary shape, whose value
is 0 if is outside of the domain, 1 if and /2 if . The parameter is the angle
between the left and right tangents at the collocation point .
To obtain a numerical solution for Equation 15, the boundary of the body is discretized. The
external boundary is divided in N
0
elements and each subregion boundary in N
k
elements.
In this work, the element adopted approximates the geometry linearly and the value of
the electrical potential is considered constant in each element. In this case, the parameter
= and then c() = 0.5 if . Each boundary element has two nodes for the
geometrical denition and a centered node, called functional node, where the potential and
its derivative are computed. Thus, the discretized form of Equation 15 for each subregion k
allows evaluating the potential at each functional node as follows
c(
i
)u(
i
) +
N
k
j=1
u
j
j
p
d
J
=
N
k
j=1
p
j
j
u
d
J
, (16)
where u
j
and p
j
represent the potential and its normal derivative at the j-th functional node.
The regular integrals are computed numerically by the usual Gauss Quadrature scheme and
the singular ones are computed analytically.
The application of Equation 16 for each sub-domain
k
, in addition to the boundary and
compatibility conditions (Equations 11 to 14) for the potential and its normal derivative at the
functional nodes of the interface elements at
0k
, yields a linear system of algebraic equations
that can be expressed in the matrix form as follows:
Hu = Gp , (17)
where u and p are vectors that store the values of potential and its derivative, respectively,
at the functional nodes of the boundary elements. The matrices H and G store the respective
computed coefcients.
The number of unknowns is the number of the external boundary elements, in which the
potential or the ux is unknown, in addition to the double of interface elements, in which the
potential and the ux are unknowns. After collect all of them at the vector y, Equation 17 can
be rewritten as
Ay = b , (18)
where A is the matrix of coefcients and b is the free vector of the linear system.
After determining the unknowns at the boundary, the values of the electrical potential at the
nodes in the center of the electrodes without prescribed values are collected in the vector V.
Such vector will be compared with the vector of measures

V during the process of solving the
inverse problem.
The implementation of the Boundary Elements Method to solve Laplaces equation was
written in Fortran language.
2.3 The inverse problem
As was said before, the aim of the EIT is to generate an image of the electrical resistivity from
measures of electrical potential at the external boundary. This problem can be formulated
as a minimization problem in which one wants to nd the model of electrical resistivity that
minimizes the distance between measured (

V) and computed (V) potentials. In this work, the
objective is to recover the shape of the ventricular cavities under the circumstances explained
before. Therefore, the resistivity model is obtained via the estimation of the vector t dened in
Section 2.1.1. In this case, the vector contains 15 parameters t
i
, with i = 1, 2, ..., 15 as described
260 Medical Imaging
before. In other words, the goal is to estimate the parameter vector t that minimizes f :
f =
1
2
R(t)
T
R(t) (19)
with
R(t) =

VV(t) (20)
where f is the objective function ( f : R
n
R), R(t) is the residual function (R : R
n
R
m
),
m is the number of measures and n is the number of parameters. The number of measures
depends on the adopted protocol to inject current and measure electrical potential.
Equation 19 shows that the problem leads to a non-linear least square problem. So, many
techniques can be used to solve it. The so called global convergent methods, for example,
Genetic Algorithms (Eiben & Smith, 2003; Michalewicz, 1996), has the advantage of avoid the
convergence to local minimum. However, they demand a large number of evaluations of
the objective function. On the other hand, the local minimization methods converge faster
to the minimum because they use the derivatives of the objective function with respect to
the parameters. They also demand a suitable initial approximation to converge to the global
minimum. Hybrid strategies, that combine the advantages of local and global methods can
be used with success (Hsiao et al., 2001; Peters et al., 2010). In this work, the features of the
problem allows the use of a local strategy, the called Levenberg-Marquard Method, that will
be briey described as follows.
2.3.1 Levenberg-Marquardt method
The Levenberg-Marquardt Method was adopted to solve the non-linear least square
problem represented by Equation 19. The detailed description about this method is vastly
found in the literature (Dennis & Schnabel, 1996; Fletcher, 1980; Madsen et al., 2004). The
Levenberg-Marquardt Method can be understood as the Gauss-Newton method modied by
the model trust region approach. In this method, the minimizer of the non-linear least-square
problemis obtained iteratively. Each update of t is given by the minimizer t
+
of the following
constrained linear least-square problem:
minimizes R(t
0
) +J(t
0
)(t
+
t
0
)
2
(21)
subject to t
+
t
0
2

0
. (22)
where t
0
is the current value for the vector of minimization parameters and t
+
is the updated
vector. R is the same residual function mentioned before. J R
mn
is the Jacobian matrix,
storing the derivatives of each element of the residual vector with respect to the optimization
variables (J
ij
= R
i
/t
j
).
0
is the initial value of the radius of the trust region.
The solution of this constrained minimization problem is the updated vector of variables t
+
:
t
+
= t
0
J (t
0
)
T
J (t
0
) + I
1
J (t
0
)
T
R(t
0
) , (23)
where I is the identity matrix and is the parameter that modies the Gauss-Newton method.
If
J (t
0
)
T
J (t
0
)
1
J (t
0
)
T
R(t
0
)
2

0
, = 0, otherwise, = 0.
There are some different implementations of this method with respect to the update of the
parameter . In this work, the implementation of the Levenberg-Marquardt Method is
provided by MINPACK-1, a standard package of subroutines written in Fortran language
to solve non-linear equations and non-linear least squares problems, that is available
at the Netlib repository (http://www.netlib.org/minpack). More details about the adopted
implementation can be found in Mor et al. (1980). In the numerical experiments presented in
this work, the subroutine LMDIF of MINPACK-1 was used. Such subroutine demands only
the computation of the residual function R(t). The jacobian matrix is approximated by nite
differences. So, the element J
ij
of the Jacobian matrix is computed as follows:
J
ij
=
R
i
t
j
R
i
(t + h
j
e
j
) R
i
(t)
h
j
(24)
where h
j
is a small nite perturbation at the j-th element of the original vector of optimization
variables t and e
j
is the j-th column of the identity matrix.
The value of the parameter h
j
is computed by the product

t
j
, where is a parameter
provided by the user. If the machine precision is greater than the computed h
j
, this value
is substituted by the machine precision.
2.4 Numerical experiments
2.4.1 Stimulation patterns
An important aspect of the EIT problem is the choice of the protocols of current injection and
electrical potential measurements. Since the problem is ill-conditioned, the suitable choice
can be determinant to the success of the image generation. However, a deep study of the
inuence of different protocols on the solution of the inverse problem is not the focus of this
work. More information on this topic can be found in other works, for instance, Peters & Barra
(2010). Here we are limited to compare two patterns of electrical current injection. The rst is
called diametrical and the second is called alternative. Furthermore, all the experiments were
done considering 16 electrodes equally spaced on the external boundary of the torso.
The name of the rst pattern, diametrical, comes froman analogy. If the domain were circular,
the electrodes used to inject current are diametrically opposed. Although the torso is not
circular, the name of the pattern remains. In this pattern, 8 different cases of current injection
is taken and 13 measures of potential for each case. So, the diametrical pattern yields 104
measures.
The second pattern, here called alternative, is a tentative of illuminating the region of interest
better than other regions. Therefore, in this pattern, the electrodes used to inject current, the
driven electrodes, are taken near to the heart. So, 6 cases of current injection with 13 measures
each one give a total of 78 measures. Each double arrowof Fig. 5 indicates the driven electrode
pair in each case of current injection. In both patterns, measurements on driven electrodes are
not considered.
It is important to note that, in this work, the measured electrical potential values (

V) were
also synthetically generated, i.e., also numerically obtained.
262 Medical Imaging
(a) Diametrical Pattern (b) Alternative Pattern
Fig. 5. Two stimulation patterns used in this work. Each double arrow indicates one pair of
driven electrodes.
2.4.2 Problem
From MR images taken at the end of the systole and at the end of the diastole the cardiac
ventricles were manually segmented. In this work we are considering an image of just
one transversal section of the thorax. Therefore, the problem is treated as 2D and an
approximation of the ejection fraction is needed.
Here, the section of the cavities were assumed to be proportional to their volumes, i.e. a
cylindrical approximation. So that, in accordance to Equation 1, EF is approximated by:
EF =
EDAESA
EDA
(25)
where EDA is the end-diastolic area and ESA is the end-systolic area.
Therefore, after segmentation, the EF is calculated in accordance to Equation 25. The EF of the
left ventricle is 59.24% and the EF of the right ventricle is 29, 95%. These values characterize
the initial situation of the heart cicle. From this moment, the EIT can be used to monitor the
EF.
Later, a cardiac dysfunction was synthetically generated. In this simulated dysfunction model
the end-diastolic volume is the same as in the normal cycle but the end-systolic volume is
greater than the normal one. In this pathological situation new cardiac cycle, the EF of the left
ventricle is 33.01% and the EF of the right ventricle is 16.19%. These dysfunction values are
the target values to be estimated by the here proposed method.
As mentioned in Section 2.1, we have also tested the methods considering two different 2D
models. Each with a different value for the resistivity of the lungs: 20000cm (RLT = 20) and
50000cm (RLT = 50).
As was said before, the solution depends on the initial guess provided for the local
minimization method. So that, for each of 4 optimization problems (2 stimulus patterns times
2 RLT models) we have tested the optimization method with two different initial guesses. One
guess corresponds to the shape of the ventricles at the end of the diastole of normal heart, i.e.
t
i
= 1, i and the other at the end of the systole for the normal tissue, i.e. t
i
= 0, i . The initial
guesses and the target can be compared in Fig. 6. Thus, the method was executed a total
of 8 times (2 stimulus patterns times 2 RLT models times 2 initial guesses). Each execution
computes the parameters t of the end of the systole. The areas inside the curves dened by t
are the ESA. These values together with the known EDA values, that is supposed to be the
same as the initial condition of the heart cycle, are used with Equation 25 to compute the EF.
Therefore, each execution yields 2 values: EF of the left and right ventricle.
ROI
Target
Initial Guess
(a) t
i
= 0, i
ROI
Target
Initial Guess
(b) t
i
= 1, i
Fig. 6. A typical target (pink) and the two initial guesses (green) given to the optimization
procedure. The rst one corresponds to the systole and the second one to the diastole. ROI
denotes the region of interest.
3. Results
Table 2 presents the results of our numerical experiments that aim the EF estimation of the
synthetically generated cardiac dysfunction. The columns present the results for the models
with different values for the resistivity of the lungs: 20000cm (RLT = 20) and 50000cm
(RLT = 50). Each couple of rows presents the comparison of the two stimulation patterns
implemented: diametrical and alternative. In addition, for each pair (stimulus pattern, RLT)
results are presented for two different initial guesses. The rst one corresponds to the shape
of the ventricles at the end of the systole for the normal heart, i.e. t
i
= 0, i and the other at
the end of the diastole of the normal heart, i.e. t
i
= 1, i. The last row of the table presents the
target values for comparison: 16.19% the EF of right ventricle (RV) and 33.01% the EF of left
ventricle (LV).
Table 3 shows the relative errors between the values of EF obtained in each execution and the
target values for each ventricle. These values are computed as
% = 100
|

EF EF|
EF
, (26)
where %is the error,

EF is the value achieved by the inverse problemsolution for the ejection
fraction and EF is the target value. The relative errors are used to compute the mean relative
errors used to compare patterns, initial guesses and models.
264 Medical Imaging
Ejection Fraction (%)
Initial RLT = 50 RLT = 20
Guess RV LV RV LV
Diametrical Pattern
t
i
= 0 13.00 34.41 15.32 34.22
t
i
= 1 16.09 32.21 15.80 33.04
Alternative Pattern
t
i
= 0 12.97 35.86 20.54 29.94
t
i
= 1 18.72 32.84 20.89 29.40
Target 16.19 33.01 16.19 33.01
Table 2. Values of the ejection fraction estimated for the synthetic cardiac dysfunction for two
resistivity models (RLT = 20 and RLT = 50), two different stimulation patterns (diametrical
and alternative) and two initial guesses (t
i
= 0 and t
i
= 1). The last row shows the target
values of the EF.
Relative Errors (%)
Initial RLT = 50 RLT = 20
Guess RV LV RV LV
Diametrical Pattern
t
i
= 0 19.70 4.24 5.37 3.67
t
i
= 1 0.62 2.42 2.41 0.09
Alternative Pattern
t
i
= 0 19.89 8.63 26.87 9.30
t
i
= 1 15.63 0.51 29.03 10.94
Table 3. Relative errors of the obtained EF with respect to the target values.
Figure 7 allows a geometrical comparison between the nal results and the actual (target)
curves. In order to make the visualization easier, these gures show the region of interest
dened in Fig. 6 without the curves of the lungs. It is important to emphasize that, to make
the comparison fair, the results presented in these gures were obtained with the same initial
guess, t
i
= 1, i.
The results show that, except in one case, the error of the ejection fraction of the left ventricle
is smaller than the right ventricle value. The mean relative error of the eight results of the left
ventricle results is 4.98% while the right ventricle is 14.94%.
Moreover, except in one case, the diametrical pattern provides results closer to the actual
values than the alternative pattern. The diametrical pattern provides a mean relative value of
4.82% while the mean error of the alternative pattern is 15.10%.
Target
Result
(a) RLT = 50 - Diametrical
Target
Result
(b) RLT = 20 - Diametrical
Target
Result
(c) RLT = 50 - Alternative
Target
Result
(d) RLT = 20 - Alternative
Fig. 7. Some results for the diametrical and the alternative pattern and the target.
About the initial guess, both of themprovided good results. However, in this experiments, the
best results were obtained with the guess of the original diastole curve, t
i
= 1, with a mean
relative error of 7.70% against an error of 12.21% for the other initial guess.
The geometrical results presented in Fig. 7, showing only the ventricular cavities, suggest
that they become worse in the case the lung resistivity is greater. This behavior is expected
because greater resistivities around the region of interest tend to block the electrical current to
reach this area. For instance, for the best experimented pattern, the diametrical one, the mean
relative error obtained with the greatest resistivity (RLT = 50) is 3.37% while the mean relative
error obtained with the other lung resistivity (RLT = 20) is 1.44%.
The best result can be consideredthe one obtained for RLT = 20, diametrical pattern and t
i
= 1.
The relative error in the value of the ejection fraction is 0.09% for the left ventricle and 2.41%
for the right ventricle. Figure 7(b) shows this result. In this case it is very difcult to see the
difference between the result and the target.
266 Medical Imaging
4. Discussions and conclusions
The presented results suggest that the proposed methodology allows a suitable indication
of the cardiac ejection fraction. We have observed that the error in the ejection fraction
predictions for the right ventricle are greater than those found for the left ventricle and this is
in agreement with other techniques, such as with echocardiography.
Concerning the different patterns for current injection tested in this work, the errors obtained
with the diametrical pattern are smaller than those using the alternative pattern, in general.
However this fact does not discard the use of the alternative pattern, as it presents good results
and spends around 19 min. in a Pentium 4, 3.00 GHz, for a complete solution, 25% less then
the diametrical.
Comparing the results obtained with different lung resistivities we may conclude that the
inverse problem becomes more difcult to be solved as the RLT increases. Therefore, the
results suggest the current injection should be triggered during the expiratory phase, when
the air volume and the corresponding lung resistivity are smaller.
The preliminary results presented in this work suggest the proposed technique is a promising
diagnostic tool that may offer continuous and non-invasive estimation of cardiac ejection
fraction. However, the use of the EIT in real applications demands further improvements.
The model could be improved, for instance, by the implementation of the complete electrode
model (Cheney et al., 1999).
Another point is that, in this work, we assume the resistivities of the tissue known. Future
works should include the resistivities of the tissue as parameters of the inverse problem, as
well as deeper studies about the electrical properties of biological tissues. Furthermore, the
behavior of the proposed method when subjected to real data should be evaluated.
5. Acknowledgements
This work was partially supported by FAPEMIG, UFJF, CAPES, CNPq and FINEP. In
particular, the rst author would like to thank CAPES for the Masters scholarship at Federal
University of Juiz de Fora and CNPq for the Doctoral scholarship at Federal University of Rio
de Janeiro.
6. References
Adler, A., Amyot, R., Guardo, R., Bates, J. H. T. &Berthiaume, Y. (1997). Monitoring changes in
lung air and liquid volumes with electrical impedance tomography, Journal of Applied
Physiology 83(5): 17621767.
Anderson, E., Bai, Z., Bischof, C., Blackford, L. S., Demmel, J., Dongarra, J. J., Du Croz, J.,
Hammarling, S., Greenbaum, A., McKenney, A. & Sorensen, D. (1999). LAPACK
Users guide (third ed.), Society for Industrial and Applied Mathematics, Philadelphia,
PA, USA.
Barber, D. C. & Brown, B. H. (1984). Applied potential tomography, Journal of Physics
E-Scientic Instruments 17(9): 723733.
Barra, L. P. S., Peters, F., Martins, C. P. & Barbosa, H. (2006). Computational experiments
in electrical impedance tomography, XXVII Iberian Latin American Congress on
Computational Methods in Engineering, Belm, Brasil.
Barra, L., Santos, R., Peters, F., Santos, E. & Barbosa, H. (2006). Parallel computational
experiments in electrical impedance tomography, 18th Symposium on Computer
Architecture and High Performance Computing, Vol. 1, Sociedade Brasileira de
Computao, High Perfomance Computing in the Life Sciences, Ouro Preto, Brasil,
pp. 713.
Baysal, U. & Eyuboglu, B. M. (2000). Tissue resistivity estimation in the presence of positional
and geometrical uncertainties, Physics in Medicine and Biology 45(8): 23732388.
Blanc, C. & Schlick, C. (1995). X-splines: A spline model designed for the end-user,
SIGGRAPH, ACM - Association for Computing Machinery, Los Angeles, EUA,
pp. 377386.
Borsic, A., McLeod, C., Lionheart, W. & Kerrouche, N. (2001). Realistic 2d human thorax
modelling for eit, Physiological Measurement 22(1): 7783.
Brebbia, C., Telles, J. C. F. & Wrobel, L. C. (1984). Boundary Elements Techniques: Theory and
Applications in Engineering, Springer-Verlag.
Brown, B. H., Barber, D. C. & Seagar, A. D. (1985). Applied potential tomography - possible
clinical applications, Clinical Physics and Physiological Measurement 6(2): 109121.
Bruder, H., Scholz, B. & Abrahamfuchs, K. (1994). The inuence of inhomogeneous
volume conductor models on the ecg and the mcg, Physics in Medicine and Biology
39(11): 19491968.
Cheney, M., Isaacson, D. & Newell, J. C. (1999). Electrical impedance tomography, SIAM
Review 41(1): 85101.
Choi, M. H., Kao, T. J., Isaacson, D., Saulnier, G. J. & Newell, J. C. (2007). A reconstruction
algorithm for breast cancer imaging with electrical impedance tomography in
mammography geometry, IEEE Transactions on Biomedical Engineering 54(4): 700710.
Dennis, J. & Schnabel, R. B. (1996). Numerical Methods for Unconstrained Optimization and
Nonlinear Equations, SIAM.
Eiben, A. E. & Smith, J. E. (2003). Introduction to Evolutionary Computing, Springer.
Fletcher, R. (1980). Practical methods of optimization, Wiley-Interscience, New York, NY, USA.
Gabriel, C. (1996). The dielectric properties of biological tissue: I. literature survey, Phys. Med. Biol.
Grimnes, S. (2008). Bioimpedance and Bioelectricity Basics, Academic Press.
Holder, D. (2005). Electrical Impedance Tomography: Methods, History and Applications,
Medical Physics 32: 2731.
Hsiao, C. T., Chahine, G. & Gumerov, N. (2001). Application of a hybrid genetic/powell
algorithm and a boundary element method to electrical impedance tomography,
Journal of Computational Physics 173: 433454.
Karhunen, K., Seppanen, A., Lehikoinen, A., Monteiro, P. J. M. & Kaipio, J. P. (2009).
Electrical resistance tomography imaging of concrete, Cement and Concrete Research
40(1): 137145.
Kim, M., Kim, K. & Kim, S. (2004). Phase boundary estimation in two-phase ows with
electrical impedance tomography, Int. Comm. Heat Transfer 31(8): 11051114.
268 Medical Imaging
Lima, C. R., Mello, L. A. M., Lima, R. G. & Silva, E. C. N. (2007). Electrical impedance
tomography through constrained sequential linear programming: a topology
optimization approach, Measurement Science and Technology 18(9): 28472858.
Madsen, K., Nielsen, H. & Tingleff, O. (2004). Methods for non-linear least squares problems.
Mello, L. A. M., de Lima, C. R., Amato, M. B. P., Lima, R. G. & Silva, E. C. N. (2008).
Three-dimensional electrical impedance tomography: A topology optimization
approach, Biomedical Engineering, IEEE Transactions on 55(2): 531540.
Michalewicz, Z. (1996). Genetic Algorithms + Data Structures = Evolution Programs, Springer
Verlag.
Mor, J. J., Garbow, B. S. &Hillstrom, K. E. (1980). User guide for MINPACK-1, Technical Report
ANL-80-74, Argonne National Laboratory, Argonne, IL, USA.
Park, B. G., Moon, J. H., Lee, B. S. & Kim, S. (2008). An electrical resistance tomography
technique for the monitoring of a radioactive waste separation process, International
Communications in Heat and Mass Transfer 35(10): 13071310.
Patterson, R. P. & Zhang, J. (2003). Evaluation of an eit reconstruction algorithm using
nite difference human thorax models as phantoms, Physiological Measurement
24(2): 467475.
Peters, F. & Barra, L. (2009). A strategy for parametrization renement in the solution of
a geometric inverse problem, Computational Modeling (MCSUL), 2009 Third Southern
Conference on, pp. 136 142.
Peters, F., Barra, L. & Santos, R. (2009). Determination of cardiac ejection fraction by
electrical impedance tomography - numerical experiments and viability analysis,
Computational Science - ICCS 2009, Vol. 5544/2009 of Lecture Notes in Computer Science,
Springer Berlin / Heidelberg, pp. 819828.
Peters, F. C. & Barra, L. P. S. (2010). Some numerical results on the inuence of measurement
strategies and load patterns in the eit inverse problem, Journal of Physics: Conference
Series 224(1): 012145.
Peters, F. C., Barra, L. P. S. & Lemonge, A. C. C. (2010). Application of a hybrid optimization
method for identication of steel reinforcement in concrete by electrical impedance
tomography, 2nd International Conference on Engineering Optimization.
Polydorides, N., Lionheart, W. & McCann, H. (2002). Krylov subspace iterative thechniques:
On the brain activity with electrical impedance tomography, IEEE Transactions on
Medical Imaging 21(6): 596603.
Rush, S., McFee, R. & Abildskov, J. A. (1963). Resistivity of body tissues at low frequencies,
Circulation Research 12(1): 4050.
Schwan, H. P. & Kay, C. F. (1956). Specic resistance of body tissues, Circulation Research
4(6): 664670.
Seo, J., Kwon, O., Ammari, H. & Woo, E. (2004). A mathematical model for breast cancer
lesion estimation: Electrical impedance technique using ts2000 commercial system,
IEEE Transactions on Biomedical Engineering 51(11): 18981906.
Trigo, F., Lima, R. & Amato, M. (2004). Electrical impedance tomography using extended
kalman lter, IEEE Transactions on Biomedical Engineering 51(1): 7281.
Yang, F. & Patterson, R. P. (2007). The contribution of the lungs to thoracic impedance
measurements: a simulation study based on a high resolution nite difference model,
Physiological Measurement 28(7): S153S161.
Zlochiver, S., Freimark, D., Arad, M., Adunsky, A. & Abboud, S. (2006). Parametric eit for
monitoring cardiac stroke volume, Vol. 27, Iop Publishing Ltd, pp. S139S146.
270 Medical Imaging
13
Assessment of Human Skin Microcirculation
and Its Endothelial Function Using Laser
Doppler Flowmetry
Helena Lenasi
Institute of Physiology, Medical Faculty, University of Ljubljana
Slovenia
1. Introduction
Human skin is the largest organ of the body: it accounts for approximately 5% of the total
body weight, extends over about 1.8 m
2
and has an average thickness of 1-2 mm. Besides
providing the mechanical barrier to protect the surface of the body and to prevent water loss
it has other important functions. It is engaged in sensory perception and vitamin D
metabolism, in inflammation, hemostasis and wound healing. Its role in human
thermoregulation is essential (Roddie, 1983; Rowell, 1993).
The major role of blood flow to the skin is therefore concerned with temperature regulation.
Accordingly, it is not surprisingly that blood flow to the skin goes far beyond its nutritive
demands. The nutritive blood flow is estimated to comprise only 20% of the skin blood flow
(SkBF), whereas the rest represents the functional blood flow. On the whole, the SkBF has
been estimated to amount to 0.3 l/min at rest in thermoneutral conditions. During exposure
to cold it can be reduced to almost zero, whereas it can increase up to 8 l/min during
strenous exercise in a hot environment. It is thus obvious that it plays an important role in
hemodynamics, since during strenous exercise and in severe heat stress it can comprise over
50% of the cardiac output, as compared to only 5% in resting thermoneutral conditions
(Johnson, 1996; Rowell, 1993). The range of flows is therefore wide, and, in extreme
conditions, high SkBF also represents a burden on the working heart. Indeed, many persons
with borderline cardiac failure develop severe failure in a hot weather because of the extra
load on the heart and then revert from failure in cool weather. As one of the major functions
of the skin is to eliminate excessive heat, its temperature is generally under 37C.
In recent years, the cutaneous microcirculation has gained increasing interest. Its easy and
noninvasive accessibility renders skin microcirculation an ideal site for measuring.
Moreover, as a dynamic structure it may serve as a model for generalized microvascular
function as studies have shown a correlation of vascular reactivity between different
vascular beds over the body (coronary arteries, brachial artery and skin microcirculation to
list a few of them) in health and disease, at least with regard to endothelial function
(Holowatz et al., 2008).
There is a constant competition between thermoregulatory and non-thermoregulatory
challenges. Many different mechanisms, ranging from systemic to local factors, play in

Medical Imaging

272
concert to balance these two demands and to eliminate redundant heat. This duality also
explains the complex ultrastructure of the skin (Braverman, 1997). SkBF is regulated by
centrally mediated neural mechanisms and by local humoral factors. Of local factors, the
endothelium plays a pivotal role in the regulation of vascular tone. In spite of immense
effort that was put in the investigation of endothelial function of skin microcirculation
there are still many mysteries to be resolved. However, the term endothelial function is
mostly used to refer to the ability of the endothelium to release substances that induce
vasodilation by directly causing relaxation of the underlaying smooth muscle cells (SMC)
(Crakowski et al., 2006).
Elucidating the role of endothelial vasodilators and their interactions as well as assessing
endothelial function in human skin microcirculation is also important from the clinical point
of view as early detectable endothelial dysfunction might precede clinical manifestation of
the disease states.
One of the measures to improve endothelial dysfunction due to reduced nitric oxide (NO)
bioavailability is regular physical activity. Nevertheless, the results on the impact of
endurance training to the endothelium-dependent vasodilation, particularly in skin
microcirculation of healthy persons, are controversial and the mechanisms induced by
training remain to be elucidated.
Over the last years, new advanced techniques and strategies emerged in order to explain the
function, the cell-to-cell communication and the complex interaction between
pharmacological mediators in cutaneous microvasculature. Although different methods
exist that tend to estimate skin blood flow none is optimal. A gold standard to evaluate skin
microcirculation and its reactivity to various stimuli is laser Doppler flowmetry (LDF).
2. Skin microcirculation
2.1 Anatomy and physiology of skin microcirculation
The skin microcirculation is an anastomotic network of vessels with many crucial functions
for the human as a whole. Namely, with its large capacity, it has to maintain nutrition of the
epidermis and its adnexa, it is essential for human thermoregulation, in takes a major part in
inflammation and wound healing, and, last but not least, it plays an important role in the
determination of peripheral resistance. From all the above, it is obvious that the blood flow
must be finely regulated and tuned in order to fulfill all the demands of the organism.
Because of its dimension, skin vasculature belongs to microcirculation only. It comprises
arterioles, capillaries and venules (Braverman, 1997). The microvessels in the papillary
dermis vary in diameter from 10 to 35 m, but most range from 17 to 22 m.
The cutaneous arteries arise from the subcutaneous tissue and enter the dermis to form the
cutaneous arterial plexuses. The arterioles and venules of the cutaneous microcirculation
form two important horizontal plexuses parallel to the skin surface: an upper horizontal
plexus in the papillary dermis (subepidermal or subpapillary plexus) from which the
nutritive capillary loops arise and a lower horizontal plexus at the dermal-subcutaneous
border. These two plexuses communicate with each other with arterioles and venules and
represent the physiologically important areas in the skin (Braverman, 1997).
and Its Endothelial Function Using Laser Doppler Flowmetry

273
Capillaries connect the arteriolar subpapillary plexus with the venous subpapillary plexus,
and single capillary loops ascend to each papilla. They have a mean length of approximately
0.2 mm to 0.4 mm and each supplies on average 0.04 to 0.27 mm
2
of the skin surface.
Special feature of human skin microcirculation are arteriovenous anastomoses (AVAs).
These are direct connections between arterioles and venules of the deep epidermal plexus
and have thick muscular and richly innervated walls. They provide a low resistance short
circuit for the passage of blood from arterioles to venules at high flow rates, as in response
to body heating (Braverman, 1997; Roddie, 1983). They are found mainly in the apical
regions of the skin (hands, feet, nose, lips and ears) and are most numerous in the nail beds,
tips of digits, and palmar surfaces of digits, palms and soles, but they are almost absent
from the dorsum of these areas.
The microvasculature of the skin varies significantly in different regions of the body, with
respect to density and architecture of the vascular network as well as regarding blood flow.
With respect to their structure and physiologic role, different skin parts could be divided
into glabrous and nonglabrous areas: nonglabrous or hairy areas are areas over the skin of
the trunk and extremities, whereas glabrous or nonhairy skin areas are found in palms, soles
and lips. These regions differ with regard to the vascular control mechanisms. Nevertheless,
there is not a discernible function that can be ascribed to one or another region.
2.2 Control of skin blood flow
As an important part of the cardiovascular system, SkBF undergoes periodic demands for
increased and decreased blood flow in order to achieve cardiovascular adjustments during
thermoregulatory challenges, during orthostasis and exercise.
Since SkBF takes part in the above-mentioned systemic reflexes its blood flow is mainly
regulated by neural mechanisms rather than by local factors. However, local factors, in the
first line factors released from the vascular endothelium and various substances co-
transmitted from local nerve endings, can strongly modulate this general response to
systemic challenges.
2.2.1 Neural control mechanisms
With respect to the anatomical organisation of skin microcirculation, there are principally
two types of neural control mechanisms (Johnson & Kellogg, 2010; Kellog, 2006).
I. All skin areas (glabrous and nonglabrous) throughout the body are supplied with
sympathetic vasoconstrictor fibers that secrete norepinephrine at their nerve endings and,
under resting conditions, exhibit a tonic vasoconstrictor influence on skin blood vessels. The
resting tone is maintained by a continuous nerve discharge of 1-3 impulses per second and
can be varied in both directions to produce vasoconstriction and, without need for any other
specialised vasodilatory fibers, also vasodilation. Indeed, under exposure to cold, there is an
increased discharge from this part of the sympathetic system, causing further
vasoconstriction. On the other hand, under heat stress, the tonic outflow through
sympathetic vasoconstrictor system is reduced. This constrictor system is extremely
powerful in glabrous areas with abundant AVAs areas that are most frequently exposed to
severe cold. Thermoregulatory as well as other reflexes in these regions are mediated by

Medical Imaging

274
changes in noradrenergic vasoconstrictor tone as well as direct effects of local temperatures
on the skin (Johnson & Kellogg, 2010; Kellog, 2006).
II. In addition, nonglabrous skin areas also receive the so-called active (cholinergic)
vasodilatory nerve fibers, most probably from the sympathetic origin too. Over the last
years, many studies have been conducted to elucidate this cholinergic vasodilation
mecanism, but the results are not conclusive. It is postulated that acetylcholine (ACh) is the
most probable neurotrasmitter, possibly linked to sweating. The current hypothesis
proposes a co-transmitter released by the nerve endings innervating either blood vessels
directly or sweat glands and exerting its relaxant action on the vascular SMC. A putative co-
transmitter has not been identified yet: the candidates include vasoactive intestinal peptide,
substance P, histamine from mast cells acting on histamine H
1
receptors, prostanoids, and
NO (Johnson & Kellogg, 2010; Kellog, 2006).
It should be mentioned that there are modulators of the neuraly mediated mechanisms
involved in cutaneous vasoconstriction and vasodilation. In women, skin vascular reactivity
depends on the phase of the menstrual cycle: progesteron resets the treshold for active
cutaneous vasodilation (Charkoudian & Johnson, 2000). Furthermore, SkBF and the
cutaneous microvascular reactivity are subject to diurnal rhytms (Aoki et al., 2003). Another
modulator that resets the threshold for active cutaneous vasodilation is physical exercise
(Johnson, 1996; Rowell, 1993) that is covered in more detail elsewhere in the chapter.
2.2.2 Local control mechanisms
Apart from neural mechanisms, local factors also impact SkBF. The standard meaning of the
term applies to the mechanisms independent of nerves and hormones, by which organs and
tissues alter their own arteriolar resistance. Included are classical vasodilators, such as
decreased oxygen partial pressure, hydrogen ion, and locally released metabolites as adenosin,
and increased concentrations of potassium. Their effect on the local vascular tone is more
pronounced when inducing short-term hypoxia, such as a transient occlusion of the proximal
artery. All things considered, little is know about these local mechanisms in controlling human
skin microcirculation although their role seems to be of minor importance.
Much more important seems to be the role of vascular endothelium as a local regulator of
vascular tone, also with respect to the skin microcirculation (Johnson, & Kellog, 2010; Kellog,
2006). The role of endothelium as a local regulator of vascular tone (in glabrous as well as in
nonglabrous parts) has been assessed and confirmed by different independent studies.
Last but not least, local thermal factors, i.e. local heating and cooling of the skin, also play an
important role in modulating vascular tone (Johnson, & Kellog, 2010; Kellog, 06).
Besides the above mentioned neural and local control mechanisms, skin microvessels also
exhibit local and autonomous variations of blood diameter that cause corresponding flow
variations and have been termed vasomotion. The contractions have a low frequency that
can be obtained by transforming the signal of the laser Doppler flux (1-2 cycles/min) using
spectral analysis. Vasomotion is suggested to facilitate blood flow through microvessels and
can be amplified under conditions of circulatory and metabolic stress, as for examle by
inducing a transient ischemia (Kvandal et al., 2006; Rossi et al., 2008). The origin of
vasomotion remains to be fully elucidated: it has been proposed that the spontaneous
muscular contractions are due to the intrinsic myogenic activity of vascular SMC.

275
2.3 Endothelium as a local regulator of vascular tone
In the last decades, a number of studies conducted in animals as well as in humans
extensively expanded the knowledge about the pivotal role of endothelium in the regulation
of vascular tone (Vanhoutte, 1989). In response to physical forces and neurohumoral
mediators, endothelial cells secrete several vasoactive substances that affect vascular tone by
inducing contraction and relaxation of the underlying smooth muscle. The most important
endothelial vasodilators are NO, prostacyclin (PGI
2
) and endothelium-derived
hyperpolarizing factor (EDHF), whereas vasoconstrictors include endothelin (ET), platelet
activating factor (PAF), etc. We will focus on the role of vasodilators.
Physical stimuli are shear stress exerted on the vascular wall by the flowing blood, and
pulsatility of the vessel wall, whereas pharmacological stimuli include a variety of
endothelial agonists, such as ACh, bradykinin, thrombin, serotonin, estrogens etc. In
general, endothelial function in vivo can be estimated by applying pharmacological agonists
to induce endothelium-dependent vasodilation. When estimating endothelial function, the
terms NO-dependent and nonNO-dependent vasodilation (Higashi & Joshizumi, 2003) are
often used to discern between different mediators involved.
NO is constitutively secreted by endothelial cells upon the action of endothelial nitric oxide
synthase (eNOS) that is constitutively expressed in all endothelial cells. eNOS can further be
activated by increases of calcium ions or by different agonists that induce changes in the
phosphorylation of specific aminoacid residues of the enzyme (Fleming, 2010). Its action
depends on the bioavailability of substrats (such as L-arginine) and cofactors, such as
tetrahydrobiopterin (BH
4
). NO induces vasodilation by activating soluble guanilat cyclase in
SMC that further catalyses cGMP, causing vasorelaxation of SMC.
As for the role of PGI
2
and other prostanoids, their importance in skin microcirculation is
unequivocal. They are synthesised by the family of cyclooxygenase (COX) enzymes. PGI
2

acts principally to modulate the function of vascular SMC and platelets; it is suggested to
play only minor role in the regulation of vascular tone in health. On the contrary, it may
play a compensatory role in the settings of decreased NO bioavailability (McCord et al.,
2006; Szerafin et al., 2006).
The vasodilatation that remains after the combined eNOS and COX blockade has been
ascribed to a yet unidentified mechanism. As it induces hyperpolarization of endothelial
and vascular SMC by activating different families of K
+
channels, it has been termed EDHF
(Feletou & Vanhoutte, 2009). The potential mechanisms leading to the hyperpolarization of
endothelial cells and SMC include arachidonic acid metabolites derived from
cyclooxygenases, lipooxygenases and cytochrome P450 (CYP) pathways, as well as H
2
O
2
,
CO and H
2
S. Another putative mechanism associated with the hyperpolarization of both
endothelial and vascular SMC is suggested to be electrical coupling through myoendothelial
gap junctions following small and intermediate conductance Ca
2+
activated K
+
channels
(IK
Ca
and SK
Ca
, respectively.).
The contributions of NO, EDHF and PGI
2
to the endothelium-dependent vasodilation are
difficult to define, as their importance may vary depending on the vessel type and size as
well as on the agonist used to stimulate the endothelium (Feletou & Vanhoutte, 2009;
Schrage et al., 2005). There is a redundancy in endothelial vasodilator mechanisms and in

Medical Imaging

276
the settings of compromised endothelial function, one vasodilator may overcome the
deficiency of another one and in this way restore normal endothelial function. The same
holds true when one or more pathways of vasodilator synthesis are inhibited, i.e. blocking
eNOS and/or COX causes an increased production of the nonblocked substance(s). Also, a
complex interaction and crosstalk between different mechanisms has been suggested
(Feletou & Vanhoutte, 2009; Nishikawa et al., 2000; Osanai et al., 2000). It has been
established that NO is involved predominantly in the control of vascular tone in larger
conductance vessels, whereas in microcirculation, the role of EDHF seems to be more
important (Pohl & deWit, 1999; Urakami-Harasawa et al., 1997). Most observations are
based predominantly on animal studies and little is known about the impact of EDHF in
humans in vivo (Yang et al., 2007). In spite of several investigations, the exact interaction of
these mechanisms is still not well defined, specially not in human skin microcirculation.
3. Methods for assessing skin microcirculation
A number of methodologies have been used to provide indexes of SkBF. As SkBF is
dynamically changing the technique employed must be able to detect these dynamic changes.
Furthermore, the method has to be safe, easy to apply, accurate and reproducible, and should
not affect SkBF. Usually it is not the basal flow that is of interest but rather vascular reactivity
that can give insight into the vascular function. Thus, the measurement techniques are usually
coupled to some provocation tests which can estimate vascular reactivity. The latter is usually
tested to investigate the mechanisms involved in the regulation of SkBF, or to detect functional
changes associated with the development of various diseases, and to evaluate the progression
of a disease and efficiency of the disease treatment.
In general, there is no optimal method to assess SkBF; none of them is able to give accurate
quantitative values of SkBF. Rather, the flow is presented in relative units. All of the
methods have their advantages and disadvantages. Moreover, as they employ different
principles, the direct comparison of the data obtained by different techniques is often not
relevant, if at all possible.
Most frequently used methods to study the dynamics of the skin microcirculation include
optical methods such as intravital dynamic capillaroscopy, photoplethysmography, venous
occlusion plethysmography, laser Doppler flowmetry (LDF) and laser Doppler imaging (Serup
et al., 2004). Alternative methods are based on thermographic assessment of blood flow or on
measuring transcutaneous pO2 or on
133
xenon clearance. The most recently developed
methods include orthogonal polarization spectral imaging technique, sidestream dark field
imaging technique (Treu et al., 2011), and laser speckle contrast imaging (Turner et al., 2008).
3.1 Laser Doppler flowmetry
LDF is a noninvasive method that has been developed and accomplished over the past 20
years. It enables a sensitive, continuous, noninvasive and real-time assessment of blood
flow, being uninfluenced by the underlying skeletal muscle blood flow (Saumet et al., 1988).
In research, it has been used to study reflex control of cutaneous blood flow, and, in
conjunction with other techniques (iontophoresis, microdialysis, and local warming and
heating) to address the involvement of the endothelium in the local control of SkBF.
Furthermore, the responses to drugs can be evaluated with respect to their local effect on

277
skin microcirculation. Its use goes far beyond research purposes as it has currently also been
used in clinical practice either as part of the diagnostic procedures or to follow the
effectiveness of treatment.
3.1.1 Principles of laser Doppler flowmetry
The fundamental principles of LDF applied to the measurement of SkBF have been
described in detail in several publications (Serup et al., 2006; Shepherd & Oberg, 1990).
The method is based on the Doppler shift of the emitted laser light when it travels through
tissue and is reflected off the moving objects, such as the moving red blood cells. The
Doppler effect is a physical phenomenon that occurs between two objects, a wave source
and a wave receiver. The relative motion of the wave source and the wave detector causes a
change in the wave frequency and wavelength. The difference between the frequencies of
the emitted and received waves is called Doppler frequency shift.

Fig. 1. A schematic diagram showing the detection of a red cell flux by laser Doppler
flowmetry. Laser light is conducted to the skin via fiber optics. In the skin, a small fraction
of the light is reflected by moving red cells with a shifting frequency (Doppler effect),
whereas the rest is reflected by the same frequency. Both reflected beams are transmitted to
the receiving optical fiber.
Prerequisites for LD technique are the characteristics of the laser beam, emitted to the tissue,
such as monochromaticity and spatial and temporal coherence. The classical single-point
probe consists of an emitting optical fiber with the source of laser light and usually, of two
separate optical fibers that serve as receivers. The emitted light is guided through an optical
fiber to the tissue where it is scattered, absorbed and only a small fraction is reflected (Fig.1).
Stationary tissue reflects the light with the same frequency whereas moving cells reflect the
light with a shifting frequency (optical Doppler effect). The frequency shift of the emitted
light is proportional to the velocity of the moving erythrocytes. The reflected light is
transmitted through one or more receiving optical fibers, and the nonshifted reference and
Doppler-shifted beam are mixed on a photodetector and processed. The signal is then
converted into electrical output that is fed through high-pass filters and amplifiers to finally
obtain the signal as fluctuating voltage expressed in millivolts. The LDF signal is a stochastic

Medical Imaging

278
representation of the number of erythrocytes in the sample volume multiplied by their
velocity and is referred to as flux; rather than flowmetry, LDF is termed LD fluxmetry. Since
the red blood cell flux is linearly correlated with SkBF it is taken as an estimation of blood
flow. That is why the flux signal obtained by LDF is usually expressed in arbitrary perfusion
units (PU). Due to the complex structure and the random orientation of the cutaneous
microcirculation the obtained measurements are only semiquantitative and relative.
Due to the scattering of the emitted light in the tissue, only a small portion of the light incident
on a tissue surface will penetrate very deeply and return to the surface. This is one of the
reasons why LDF is ideally suited for measuring superficial tissues such as skin and mucosa.
The penetration depth of laser light depends on its wavelength (shorter wavelengths
penetrate superficially, whereas longer penetrate more deeply) and on the distance between
the transmitting and receiving fibers. It has been estimated that at small fiber separations,
the signal is obtained from more superficial layers, such as capillary loops whereas at
greater separations the contribution of deeper venous plexuses may dominate. In brief,
different fiber separations or the use of lasers with lights of different wavelengths may
enable the inspection of different depths of the cutaneous microcirculation.
The depth of measurement by LDF varies among tissues: for skin, it is estimated to be 0.5-1
mm. Considering the anatomical organisation of skin microvessels, the LD devices capture
signals predominantly from the deeper subpapillary layer the site of subpapillary arterial
and venous plexuses. Thus, it records the thermoregulatory blood flow rather than the
nutritive blood flow from papillary loops and superficial plexuses (Yvonne-Tee et al., 2006).
From the above, it is obvious that the variables which influence the measurement in one
individual are the structure of skin surface and skin thickness as well as hematocrit.
Many commercial devices from different manufacturers are available today (Fig.2). Most of
the current devices use a two- or five mW-powered helium-neon lasers with a wavelength
of 632,8 nm and small probes. The surface area captured by such devices is estimated to be
about 1 mm
2
, whereas the penetration depth is 1 mm, resulting in a theoretical total
measured tissue volume of 1 mm
3
. Other available instruments nowadays use infrared light
with a wavelength of 780 nm, applied by a diode laser. It is a challenge to develop devices
using lasers with wavelengths below 600 nm, which will be able to record signals from the
epidermis-dermis border. This would enable a more selective distinction between the
deeper lying thermoregulatory component of blood flow and the more superficial
nutritional component of the blood flow (Yvonne-Tee et al., 2006).
3.1.2 Application in humans: Calibration, validation and precautions
Before any measurement, the device has to be calibrated to instrumental zero (standard or
instrumental calibration) and later on, the biological zero has to be determined. Instrumental
zero calibration is obtained by placing the probe against a white surface. To test the
sensitivity and to calibrate the device the probe tip is placed into a suspension of latex
particles undergoing Brownian motion and the output of the flowmeter is tested.
The term biological zero applies to the LD flux that remains after an occlusion of a proximal
artery. Before any measurement, biological zero can be assessed. The relative ratio between
the normal resting flux and the biological zero value varies from region to region: biological

279
zero LD flux can amount up to 20% of the resting flux (Tonneson, 2006, as cited in Serup et
al., 2006). The sources of the biological zero seem to be blood cells that remained in the
peripheral vessels during arterial occlusion and are still moving randomly and producing
minor LD components recorded by the instrument as well as the Brownian movement of the
interstitium (Kernick et al., 1999).
Before the measurement, the subject should rest for at least 20 minutes to acclimatize and
calm down (Fig.2). Namely, a wide range of factors strongly influence the LDF signal and
should be kept at minimum for a more relevant interpretation of the data. Such factors are
subject-related: anatomical position of the probe, the subjects position, physical and mental
activity, previous consumption of food, beverages containing caffeine or alcohol, smoking,
taking drugs, menstrual cycle and temporal variations (circadian rhythms, inter-day
variability, seasonal variations). Environment-related factors are first of all ambient
temperature, air humidity and movement of the adjacent air. Factors that cannot be
influenced and can have an impact on LDF are age, gender and, possibly, ethnic
background. All these aspects must be taken into consideration and effort should be made
to eliminate them or at least minimize to get the best possible results.

Fig. 2. A typical equipment for assessing skin microcirculation: Laser Doppler flowmeter
(LDF, Perimed), the device to register blood pressure in the finger artery (Finapress,
Ohmeda), and the transducer for a simultaneous transmission of the digital signals to the
computer (left-hand). A typical position of the LDF probes: a standard LD probe (placed on
the finger pulp, i.e. representative of glabrous site) and an iontophoretical delivery LD
probe (placed on the volar forearm, i.e., representative of nonglabrous area) combined with
the device for the application of iontophoresis (right-hand).
Debatable is also the temperature of the skin: while some authors recommend preheating
the skin to a constant temperature and then performing a perturbation, others use the
natural skin without preheating.
Another problem are movement artifacts. To avoid it, person must lie completely still.
Further, the probe must be in close contact with the measured skin; to do so, special two-
sided adhesive tapes to fix the probe holder are available.
Due to the small measuring area and the anatomical architecture of the dermal
microcirculation, there are considerable variations in blood flow (Braverman et al., 1990). To
minimize the problem of spatial variability, several subsequent measurements have to be

Medical Imaging

280
repeated and the data obtained averaged. Furthermore, newer devices include multiple point
probes that have been designed to overcome the problem of immense spatial variability.
When performing LDF measurement, usually multi-channel devices are used that enable a
continuous measurement of LDF in many sites simultaneously (Fig.2). Usually, the
measurements are performed in the representative sites of glabrous and nonglabrous areas.
Mostly measured glabrous sites in the upper extremity are palms, finger pulps and nailfolds,
whereas nonglabrous sites include volar forearm and dorsal aspect of the hand and fingers.
3.1.3 Interpretation of the results
Many methodological issues must be taken into consideration when interpreting the results.
First, the strong site- and time-dependence of the obtained LDF signal and a great inter-
individual variability must be taken into account. An accurate measurement should include
the assessment of reproducibility in terms of the intraindividual variability coefficient that,
based on some reports, could come up to 40%.
It is also important to realize that SkBF is very dynamic and apart from known factors that
influence the variations (vasomotion, neural and humoral activity, heart beat and
respiration related changes, respectively) there are also factors that cannot be ascribed to
any known factor and thus explained. As a consequence, a wandering baseline is usually
obtained. The seemingly random variation is much lesser when SkBF is perturbed, e.g.
increased by heating or decreased by cooling. This is the basis for the provocation tests to be
applied. Accordingly, more information is obtained in assessing vascular reactivity as a
response to these perturbations. Rather than in absolute values, changes in LDF after such a
perturbation are usually expressed as a percent of the baseline value.
Furthermore, it is advised to apply perturbation to induce maximal vasodilation (i.e. heating
the skin to 44
o
C or applying an NO donor in excess) and thus achieve the highest (peak)
flow and then express the quantity of the measured LDF relative to the peak flow. This
would allow a comparison with the values assessed either by other devices or in other
experimental conditions. Indeed, in many studies, the flow is expressed as a percent of the
maximal flow (Crakowski et al., 2006; Turner et al., 2008).
Another recommendation is to express SkBF in terms of cutaneous vascular conductance
(CVC) rather than in terms of flow. CVC is obtained by dividing the LD flux by mean arterial
pressure. It is worth mentioning that all the measurements of LDF are usually performed with
concomitant continual recordings of the blood pressure and the skin temperature.
The mostly exposed problem remains the standardization of the method as well as
interpretation of the results that would enable a more accurate and relevant comparisons
among studies using different devices and protocols. There is an urge to standardize the
method as well as the protocols used (Crakowski et al., 2006; Turner et al., 2008; Yvonne-
Tee et al., 2006).
3.1.4 Advantages and disadvantages
LDF enables a semiquantitative assessment of skin blood flow and is thus expressed in
arbitrary PU. One of its disadvantages is thus inability to determine absolute blood flow.

281
Another disadvantage is great spatial and temporal variability. This must be taken into
consideration when designing the measurements and interpreting the results. On the other
hand, LDF has many advantages over some other methods: it is noninvasive, it can
determine dynamic changes in skin microcirculation being uninfluenced by the underlying
muscle blood flow (Saumet et al., 1988), it is easy applicable, reproducible, and, compared to
many other methods, is relatively attractive regarding its price. As such, it remains the gold
standard for assessing skin microcirculation and its reactivity in research purposes as well
as in clinical practice.
3.2 How to study the endothelial function of the skin microcirculation
As with other techniques, there is no uniform technique that would separately evaluate only
the endothelial function of skin microcirculation. Namely, the endothelium is not an entity
per se but it is strongly associated with other structures, in terms of anatomy as well as
physiology. There is a complex interaction between neural and other humoral mechanisms
and the endothelium, so the methods used for evaluating separate mechanisms show
considerable overlapping too. Therefore there is no single method that could give an insight
into endothelial function but many different methods and tests have to be performed in
order to get the clearest possible information.
Several methods to study endothelial function of the cutaneous microcirculation have
been developed in recent years and are currently being used, each with their own
advantages and disadvantages. By these methods, it is actually the vascular reactivity or
vasodilator capacity that is being estimated: a part of this response also encompasses the
endothelial component. So it is obvious that the results must be interpreted carefully
when determining endothelial function.
The usual approach to test endothelial function is to induce either a local or a systemic
perturbation that will change the LDF. The stimuli used are pharmacological and physical
ones. Of locally applied physical stimuli, flow mediated vasodilatation is important as
well as locally induced changes in temperature, i.e. local heating and cooling (Johnson &
Kellog, 2010; Kellogg, 2006). Flow mediated vasodilation is usually simulated by releasing
a temporal occlusion of the proximal artery, and has thus been termed postocclusive
reactive hyperemia (PRH).
The pharmacological stimuli are various agonists and antagonists that act directly on the
endothelium or on the vascular SMC to induce changes in LDF. Often, they are coupled with
the application of different blockers to inhibit the synthesis or release of the substance under
investigation, as for example blockade of eNOS or COX or different adrenergic and cholinergic
receptors as well as channels on the endothelial cells and SMC. There are various technique of
application of these substances directly to the skin: their collective pitfall is that the exact
concentration of the substance in the very tissue could not be determined accurately.
Let us briefly mention also commonly used systemic challenges that are often applied when
studying skin microcirculation. These include systemic heating and cooling (more often the
term whole body heating/cooling has been used), changing the oxygen partial pressure to
induce hypo- or hyperoxia, and changes to perturb the baroreflexes: classic orthostatic test
with a tilt-table or negative low body pressure and different mental tests. Namely, all the
systemic changes also impact skin microcirculation and have been used to estimate the

Medical Imaging

282
involvement of skin in reflexes associated with thermoregulation, exercise and other
cardiovascular adjustments.
3.2.1 Pharmacological approach: Application of vasoactive substances
Endothelial function can be assessed by applying vasoactive substances locally and
directly to the skin and recording the LDF response. In conjunction with LDF, this is
performed most commonly by using iontophoresis and alternative methods, intradermal
microinjection and microdialysis.
Iontophoresis
This is a method that enables application of soluble charged substances into the skin by
means of an externaly applied direct electrical current (Kalia et al., 2004). Special
iontophoretic drug-delivery electrodes have been designed that could be attached to the
direct laser-Doppler probe (Figs.2,3). The quantity of the drug that penetrates into the skin
is proportional to the magnitude of electrical current used. Usually, the quantity applied
is expressed in milli Coulombs (mC), which are obtained as a product of the magnitude of
electrical current (usually in order of A) and the duration of the electrical current
application. There are different modes of current application depending on the substance
used and the protocol chosen: either protocol with a continuous application of current is
used, or, more often, intermittent (interval) application of a constant or increasing current
is accomplished. It must be taken into consideration that also curent alone causes
vasodilation, referred to as galvanic response (Abou-Elenin et al., 2002; Khan et al., 2004;
Morris & Shore, 1996; Noon et al., 1998). The magnitude of the current-induced
vasodilation depends on the vehicle used: a range of preparations have been used,
including deionized water, tap water, sodium chloride and mannitol solutions, as well as
different cellulose gels. The proposed mechanisms for the current-induced vasodilation
are induction of an axon reflex (Noon et al., 1998), competition between ions of active
substance and the vehicle (Khan et al., 2004), etc.
Another aspect that might influence drug delivery to the skin is skin resistance (Ramsay et
al., 2002). It depends on the skin site, on the hydration status of the skin and also shows
great inter-subject variability. To minimize the problem of skin resistance, skin should be
cleaned with alcohol and gently rubbed mechanically to strip off the epidermis.
Nevertheless, iontophoresis has advantages over some other techniques. Compared to
microinjection and microdialysis, it does not induce trauma affecting SkBF (Crakowski et al.,
2007; Leslie et al., 2003). The quantity of drug is minimal and causes no systemic effects. Its
disadvantage is the current-induced hyperemia, which can be minimalized by using an
appropriate vehicle solution or by applying topical anesthesia (Crakowski et al., 2007;
Morris & Shore, 1996); yet, the latter can impact the results in other way.
Microinjection and microdialysis
As alternatives to iontophoresis, microinjection or microdialysis have been used for the
application of various pharmacological agents into the skin. By these techniques, either a
solution containing a single substance or a mixture of different compounds can be applied.
By microinjection, very small amounts (up to 10 l) of agents of low concentrations are

283
applied intradermally (Leslie et al., 2003). The advantage of microdialysis over
microinjection is that the skin can be continuously perfused with the solution containing the
active substance and in this way enables a constant stimulus or blockade of a certain
compound (Crakowski et al., 2007; Turner et al., 2008). Moreover, using microdialysis, it is
also possible to remove effluent fluid from the tissue and to determine certain substances of
interest in the dialysate. The major disadvantage of microdialysis is its invasiveness
(Crakowski et al., 2007).

Fig. 3. A representative laser Doppler flux (LDF) response to pulsed iontophoretical
application of acetylcholine, an endothelium-dependent agonist (left-hand) and sodium
nitroprusside, an endothelium-independent agonist (right-hand) in the volar forearm of one
subject. Arrows denote electrical pulses. PU, perfusion units.
Conventionally, ACh has been used as a standard drug to assess endothelial function (Fig.3,
left-hand). It has been established, though, that apart from acting on muscarinic receptors
expressed in endothelial cells it also affects other parts of the vessel wall: its potential action
on the local nerve endings has also been proposed (Berghoff et al., 2002; Morris & Shore,
1996). In light of such observations, the data must be interpreted cautiously. Further, its
action on the endothelium is debatable; multiple mechanisms have been proposed,
including the involvement of NO, prostaglandins and EDHF (Holowatz et al., 2005; Kellogg
et al., 2005; Khan et al., 1997; Morris & Shore, 1996; Noon et al., 1998; Turner et al., 2008).
Other commonly used drugs are metacholine, bradykinin, histamine, and substance P (Kalia
et al., 2004). To test the reactivity of vascular SMC, sodium nitroprusside has been most
commonly applied, usually by iontophoresis (Fig.3, right-hand).
3.2.2 Postocclusive reactive hyperemia
PRH refers to an increase in SkBF over the baseline following the release of a temporal
occlusion of a proximal artery and has been used as an index of endothelial function
(Fig.4), for research purposes as well as in clinical practice. It is characterised by an initial
peak flux reached in a few seconds after the release that depends on the duration of the
occlusion (Yvonne-Tee et al., 2008), and a subsequent return of the LDF to the baseline.
However, PRH is a complex phenomenon that comprises metabolic and endothelial
vasodilators, as well as a myogenic response and a sensory component (Lorenzo et al.,
2007; Yvonne-Tee et al., 2008).

Medical Imaging

284
The endothelial component of the PRH response is believed to result from vasodilators (NO,
PGI
2
and EDHF) released from the endothelium in response to augmented shear stress due
to an increase in blood flow following the release of an occlusion. Yet, the results on the
contribution of different endotehlial vasodilators are unequivocal (Bingelli et al., 2003; Dalle-
Ave et al., 2004; Durand et al., 2004; ; Medow et al., 2007; Wong et al., 2003).
In the upper extremity, PRH has been performed by a compression of either brachial artery
or diferent finger arteries, respectively to suprasystolic blood pressure. The most commonly
used occlusion time is 3 or 5 min, but it can range from 1 and up to 15 minutes. Many
parameters can be deduced from PRH; unfortunately, they are not standardised (Crakowski
et al., 2006; Yvonne-Tee et al., 2008). Indices that are assessed most commonly are: the
maximal (peak) LDF response after occlusion release (LDF
peak
), the time to reach the peak
response (t
peak
), the half time in which the LDF returns to 50% of the baseline, the duration
of hyperemia (time to recovery, t
rec
) and the area under the curve (AUC) (Fig.4). The
response of the microvasculature strongly dependes on the occlusion time (Wong et al.,
2003; Yvonne-Tee et al., 2008).

Fig. 4. A representative laser Doppler flux (LDF) response of postocclusive reactive
hyperemia (PRH) obtained in the finger pulp after the release of a 3-min occlusion of the
brachial artery. Typical indices of the PRH are indicated. PU, perfusion units.
PRH has also been used to induce or amplify the vasomotion in skin microcirculation (Rossi
et al., 2008). It is proposed that amplified LDF oscillations are due to the local synchronicity
of oscillatory blood flow in a group of cutaneous capillaries during ischemia. It is
conceivable that these oscillations represent the local myogenic component of the PRH
(Rossi et al., 2008).
3.2.3 Locally induced thermal changes: Heating and cooling
They are used to investigate the mechanisms involved in the local cooling and heating
(Johnson & Kellog, 2010; Minson, 2010), as well as in clinical practice to determine
microvascular (dys)function in different diseases.

285
Local heating is usually achieved by applying LD probes with a built-in heater; maximal
vasodilation is obtained at a temperature between 42-44
o
C. As noted previously, maximal
vasodilation is sometimes assessed to obtain the maximal vasodilating capacity; all other
measurements are referred to this maximal vasodilation.
As for the cooling either LD probes with a built-in cooler or cold water of different degrees
is used or flexible cold-gel packages are applied over the skin (Fig.5). Cooling by cold water
or flexible packs also evoke a systemic response that can be tracked in other parts of the
skin, as for example in the contralateral arm (Fig.5).

Fig. 5. A representative laser Doppler flux (LDF) response to local cooling obtained in the
finger pulp of the ipsilateral hand (direct response, upper panel) and the contralateral hand
(indirect response, lower panel), respectively. PU, perfusion units.
Both, heating and cooling, show typical patterns: an immediate response to the thermal
stimulus as well as a later, sustained response. Part of the response is mediated by axon
reflex, whereas there is also impact of NO and the endothelium. Cooling probably acts to
inhibit NOS whereas heating causes activation of NOS (Johnson & Kellogg, 2010). Apart
from eNOS, also involvement of neuronal NOS is proposed (Kellogg et al., 2009), which
renders the interpretation of the results difficult. Additional studies are needed to clarify the
exact contribution of different NOS isoforms (Kellogg et al., 2009).
3.2.4 Spectral analysis of the LDF signal
By means of spectral analysis, the LDF signal can be decomposed into components with
different frequencies. It has been postulated that low frequencies around 0.01 Hz might
reflect endothelial function, the frequencies at arround 0.04 Hz might reflect the
neurogenic influence on the vessel wall, the ones around 0.1 Hz myogenic activity,
whereas oscillations of higher frequencies at around 0.3 Hz and 1 Hz reffer to respiratory
and heart beat activity, respectively (Kvandal et al., 2006). This is one of the tools used to
assess endothelial function in physiologic (Kvernmo et al., 2003) and pathologic (Rossi et
al., 2011) conditions. The most widely used spectral methods are fast Fourier transform,
autoregressive modeling and wavelet analysis.

Medical Imaging

286
4. Human skin microcirculation and the endothelium: Impact of NO,
prostaglandins and EDHF
The role of endothelium as a local regulator of vascular tone in skin microcirculation has
been investigated in many studies. Furthermore, many studies have established an
impairment of endothelium-dependent vasodilation in human skin microcirculation with
aging (Black et al., 2008; Holowatz et al., 2007) and in disease (Colberg et al., 2005; Rossi et
al., 2011; Taddei et al., 2006). Nevertheless, the exact contribution of various endothelial
vasodilators to endothelium-mediated vasodilation has to be elucidated as the results are
discrepant. The discrepancy may partly be due to different methods and agonists used to
test endothelial function. It might be due to the use of different inhibitors of endothelial
vasodilators or different mode of their application (oral, iontophoretical application,
perfusion of the brachial artery). A mechanistic explanation would propose that inhibiting
one pathway may lead to an upregulation of another one and thus blur the actual effect of
the substance under investigation (Feletou & Vanhoutte, 2009; Nishikawa et al., 2000; Osanai
et al., 2000). Furthermore, different endothelial vasodilators show a strong site-dependency
(Schrage et al., 2005). All above must be taken into account when interpreting the results. It
has not been extensively studied to what extent the endothelium contributes to the
regulation of vascular tone and vasodilation in glabrous and nonglabrous skin area.
Although it is generally accepted that ACh mediates a NO-dependent response of
endothelium (Kellogg et al., 2005; Turner et al., 2008), there are studies that did not confirm an
involvement of NO (Hollowatz et al., 2005; Khan et al., 1997; Noon et al,1998). It has also been
claimed that the contribution of NO to the ACh-induced vasodilation may be site-dependent:
Noon et al (Noon et al.,1996) suggested that NO may be more important in the regions rich in
AVAs (glabrous areas) than in areas with predominantly nutritive blood flow (dorsum of the
hand). Even more debatable remains the role of prostaglandins in the ACh-mediated response
in human skin microcirculation. While some studies have confirmed their role (Durand et al.,
2004; Holowatz et al., 2005, Kellogg et al., 2005; Khan, 1997; Noon et al., 1998), others failed to
do so (Abou-Elenin et al., 2002; Dalle-Ave et al., 2004; Morris& Shore, 1996).
A compelling study was performed by Hendry & Marshall: contrary to most other studies
they showed that the inhibition of COX actually augmented the ACh- mediated increase in
LDF (Hendry & Marshall, 2004). They speculate that in this setting, the vasoconstrictor
products released by the endothelium (such as thromboxane A
2
and prostaglandin H
2
) may
limit the response to ACh, and not until the COX is blocked, could the maximal response to
ACh be detected. In fact, it has been shown by other authors that thromboxane A
2
exerts its
inhibitory effects on bioavailable NO by increasing oxidative stress (Tang et al., 2005) and
inhibiting NOS (Yamada et al., 2003). Furthermore, Higashi et al. (2003) have shown that
apart from NO, potassium-ATP channels, and cytochrome P-450, but not prostaglandins,
may play a role in the ACh-induced vasodilation. Yet, their study was performed using a
strain-gauge plethysmography (Higashi et al., 2003). In other studies, a portion of the ACh-
mediated response, however, was attributed to an axon reflex (Berghoff et al., 2002;).
As for the role of NO and prostaglandins in the PRH, the results are controversial too (Dalle-
Ave et al., 2004; Medow et al., 2007; Wong et al., 2003,). The role of NO seems less important
than it was predicted. Namely, Wong et al. showed no effect of NOS inhibition on the PRH
in the forearm skin (Wong et al., 2003). The results were substantiated by the study of Zhao
et al., who assessed the concentration of NO in the microdyalisate using a NO selective

287
amperometric electrode (Zhao et al., 2004). Other studies have confirmed an involvement of
NO in the PRH response (Bingelli et al., 2003; Medow et al., 2007), yet it seems only of minor
importance compared to other mechanisms. A conceivable speculation has been made by
Medow et al: they propose that the COX inhibition unmasks the NO dependence of PRH in
human skin (Medow et al., 2007). The results could strengthen the hypothesis on the
crosstalks between various endothelial vasodilators in human skin. The role of the cross-talk
between angiotensin II and NO has also been proposed (Loot et al., 2009).
As part of the ACh- and PRH induced vasodilation in human skin microcirulation persists
after simultaneous eNOS and COX blockade, it is speculated that other endothelial
mediators might also be involved. There are also cross-talks between endothelial and neural
mechanisms: NO has been shown to attenuate cutaneous responsiveness to norepinephrine
via postsynaptic mechanisms (Shibasaki et al., 2008); furthermore, it is the
2
-adrenergic
receptor on the endothelium to mediate the NO-induced as well as prostaglandin-mediated
vasodilation (Hermann et al., 2005). NO released from neuronal nNOS might be even more
important than the one from eNOS (Kellogg et al., 2009). From above observations it is clear
that different mechanisms contribute to the local regulation of vascular tone but their exact
role has yet to be elucidated.
In the light of the aforementioned studies, we aimed at addressing the role of the nonNO-
nonPGI
2
-dependent mechanism in human skin in different measuring sites as well as
using different provocation tests to stimulate the endothelium. We simultaneously
applied eNOS and COX inhibitors to the volar forearm using intradermal micro-injection
and performed iontophoresis of ACh. Our results have shown no effect of the combined
inhibition on the baseline LDF. Furthermore, about 80% of the ACh-induced vasodilation
persisted after combined NOS and COX blockade. We may only speculate about the
mechanisms involved but according to other observations, it may be attributable to an
endothelial mechanism other than NO and PGI
2
, such as a putative EDHF. Other
mechanisms, such as axon reflex, might also be involved; by applying an anesthetic cream
to the measuring site we would be able to eliminate the potential axon reflex. This is one
of the pitfalls of our study. Another pitfall is the microinjection tehnique; as mentioned
previously, the quantity of a substance applied could not be determined. It might be that
we were not able to achieve an adequate concentrations of the drugs in situ. In this regard,
microdialysis would seem to be more suitable. In another set of experiments, we induced
PRH after blockade of eNOS and COX to assess the role of nonNO-nonPGI
2
-dependent
mechanisms involved. After having obtained indices of PRH, the results have shown no
differences in the L-NMMA and diclofenac treated sites as compared to the control sites,
but there were differences in the AUC. Namely, AUC was slightly smaller in the treated
sites (unpublished observation) which again points to only a minor role of NO and PGI
2

in the LDF response induced by PRH.
With respect to the role of EDHF in human skin in vivo, the studies are scarce. The existence
of endothelium-dependent, NO/PGI
2
-independent relaxations, potentially attributable to an
EDHF, has been confirmed in humans in vivo. Based on studies in patients who exhibited a
NO/PGI
2
-independent response to an endothelial challenge, it has been suggested that
EDHF might serve as a backup vasodilator in the settings of compromised endothelial
function (Fichtlscherrer et al., 2004; Fischer et al., 2007; Taddei et al., 2006; Yang, 2007).
Different putative mechanisms are proposed to mediate the EDHF-mediated vasodilation:

Medical Imaging

288
in humans, the most probable candidate is a CYP derived metabolite (Feletou & Vanhoutte,
2009). Indeed, a mechanism sensitive to CYP epoxygenase inhibition was confirmed (Bellien
et al., 2005; Fischer et al., 2007; Fischtlscherer et al., 2004; Hillig et al., 2003; Taddei et al.,
2006). Moreover, the CYP2C9 isoform was already confirmed on the endothelium of some
human arteries where also a functional role of a CYP-derived EDHF was shown (Hillig et
al., 2003; Larsen et al., 2006).
Based on the results from previous in vivo studies in humans our aim was to further
characterize the nonNO-nonPGI
2
-dependent vasodilation in the skin microcirculation in the
forearm. Therefore, we assessed the LDF response to ACh and PRH in the presence and
absence of the selective CYP 2C9 inhibitor, sulfaphenazole, respectively. Contrary to most
other studies that mainly applied the CYP inhibitors by perfusion of the brachial artery, we
applied it by an intradermal microinjection. Sulfaphenazole had no effect either on the baseline
LDF or on the vasodilation induced by ACh (Lenasi, 2009) or PRH (unpublished results). This
is in agreement with some studies using venous plethysmography performed in healthy
humans (Passauer et al., 2003, 2005), whereas at odds with most studies performed in patients
(Fischer et al., 2007; Fischtlscherer et al., 2003; Taddei et al., 2006). We may explain the
discrepancy with the studies performed in patients by the fact that the expression of CYP
might be upregulated in patients due to a decrased bioavailability of NO. Again,
methodological limitations must be taken into consideration, such as microinjection and the
selection of a proper CYP inhibitor. Also, to evaluate the impact of EDHF in the control of skin
microcirculation, it would be suitable to apply another endothelial agonist besides ACh.
Namely, it has been proposed that the nonNO-nonPGI-dependent vasodilation might be more
sensitive to bradykinin than to ACh (Schrage et al., 2005). Furthermore, the assumption that
the NO/PGI
2
-independent relaxation is indeed due to an EDHF would be strengthened had
we applied an inhibitor of Ca
2+
-activated K-channels or a depolarizing agent to at least
indirectly prove an endothelium-dependent hyperpolarization (Feletou and Vanhoutte, 2009).
Taken together, different endothelial vasodilators are involved in the control of vascular
reactivity of the skin microcirculation. The exact mechanisms involved and the role of a
putative EDHF remain to be established.
5. Exercise augments the endothelium dependent vasodilation in skin
microcirculation
Beneficial effects of exercise on the vascular reactivity of human skin microcirculation have
been reported in many studies (Hodges et al., 2010). In general, the collective finding is that
regular physical exercise enhances endothelium-dependent vasodilation. Moreover, it has
been shown that exercise also improves the decline of endothelium-dependent vasodilation
of cutaneous microcirculation in age (Black et al., 2008) and disease (Colberg et al., 2005;
Rossi et al., 2011; Taddei et al., 2006).
The exact mechanisms of exercise-induced adaptations in human skin microcircultion
remain unresolved. Various mechanisms have been proposed, the most likely one seems to
be an increased production of NO by increased eNOS activity, probably due to repetitively
increased shear stress on the endothelium (Green et al., 2004). Apart from endothelial
adaptations, changes in the sensitivity of vascular SMC to endothelial vasodilators or
alterations in neural vascular control may play a role. By inducing local heating as a

289
vasodilating stimulus, the study of Tew et al. has shown that aerobic fitness affects the
contribution of noradrenergic sympathetic fibers to the heating-induced LDF response (Tew
et al., 2011). Indeed, it has been shown that endurance trained athletes have a diminished
temperature threshold for active vasodilation in skin compared to matched sedentary
controls (Fritzsche & Coyle, 2000). It is thus obvious that apart from local, possibly
endothelial mechanisms, neural mechanisms also are subject to adaptations.
We aimed at assessing vascular reactivity in highly endurance-trained athletes with a high
maximal aerobic capacity (V
O2max
65 ml/min/kg) and comparing it with age-matched,
sedentary controls (V
O2max
41 ml/min/kg). Vascular reactivity was assessed at two
representative sites with different control mechanisms: glabrous and nonglabrous area,
respectively. Namely, it has been shown that the hemodynamic changes associated with acute
exercise differ between these two sites (Yamazaki & Sone, 2006). The amplitude of changes,
expressed as CVC as well as the time in which the changes occur during exercise differ
between these two sites which might point to different control mechanisms. We hypothesised
that similar differences might also be observed regarding adaptation to chronic exercise. We
determined vascular reactivity by an iontophoretical application of ACh and SNP as well as by
inducing PRH. Unfortunately, we were able to aplly ACh and SNP only to nonglabrous skin
sites. As for the glabrous areas, we assessed the response to a 4 min brachial artery occlusion
on the finger pulp and compared it with the response in the volar forearm. As expected, we
found an increased responsiveness, as assessed by peak response to ACh and the indices of
PRH, respectively, in the trained. Yet, the differences in indices of PRH were observed only in
the glabrous sites and not in the nonglabrous, respectively (Lenasi & trucl, 2010). This may
prove that glabrous parts with abundant AVAs are indeed more strongly involved in
thermoregulatory adaptations to exercise. The adaptation of skin microcirculation to
exercise might be explained by a direct effect of exercise-induced shear stress on the vessels,
as well as by adaptive changes attending thermoregulatory challenges. The discrimination
between these two mechanisms is not possible as they overlap each other. Nevertheless,
Lorenzo & Minson have shown that skin microcirculation also undergoes thermoadaptative
changes as a part of acclimation to heat without eliciting reflexes engaged in exercise
(Lorenzo & Minson, 2010). To do so, they trained subjects only at 50% of their V
O2max
,
intensity that assumingly does not induce changes in the reflexes affecting exercise. Another
interesting approach was by Rossi et al.: by performing spectral analysis of the LDF signal,
they showed an augmented amplitude of the low frequency band presumably reflecting
endothelial function in the trained (Rossi et al., 2006). Similar observations were found in the
study performed by Kvernmo et al. (Kvernmo et al., 2003).
Exercise training probably induces adaptations in all age groups. Namely, a recent study
performed in adolescents showed an enhanced endothelium-dependent vasodilation in
nonglabroous skin in the group of trained subjects (Roche et al., 2010). Also, in the group of
postmenopausal women who underwent a 48-week regime of aerobic exercise training the
endothelium-dependent and independent vasodilation was enhanced as well as the
response to local heating (Hodges et al., 2010). Black et al. have shown that age-related
decline in endothelial function can be improved already after 24 weeks aerobic training
(Black et al., 2008). Contrary to the observation by Hodges et al (Hodges et al., 2010) on an
increased endothelium-independent vasodilation (induced by an application of SNP) we
found a decreased response to SNP in the trained. The result is at odds with most other
studies that either showed no differences in the SNP-induced vasodilation between the

Medical Imaging

290
sedentary and the trained or an increased response in the trained. Our observation is
difficult to interpret; we suggested a decreased sensitivity of vascular SMC to a NO donor
(Gori & Parker, 2002) which seems in certain respect illogical. The data on the vasodilator
response to heating show that the LDF increase to the application of ACh occurs prior to the
increase induced by SNP (Kellog et al., 2005). This might suggest that the endothelium is
modified by exercise sooner then the vascular SMC. The observation would support the
results of the studies that found no differences in endothelium-independent vasodilation
between the trained and the sedentary.
Interestingly, most studies found no differences in the baseline LDF between the trained and
the sedentary. The same holds true also for longitudinal studies that were performed in a
cohort of subjects who underwent training and thus are more reliable (Black et al., 2008;
Hodges et al., 2010). This seems logical as under resting conditions the nutritional needs of
skin microcirculation are similar for the trained and the sedentary. When performing
exercise, the trained exhibit a greater vasodilator capacity, probably of endothelial origin
that may fulfil the increased demands for heat elimination in the trained.
It has been accepted that augmented endothelium vasodilator capacity induced by training are
due to increased bioavailability of NO (Green et al., 2004). The speculative functional studies
were confirmed by Wang, who has shown an increased response to ACh as well as an
increased level of plasma NO metabolites after 8 weeks of training in healthy subjects. (Wang,
2005). Also, Vassalle et al. have shown an enhanced release of NO in the trained combined to
an increased LDF response induced by PRH (Vassalle et al., 2003). Increased NO might be due
to an upregulation of eNOS or its enhanced activity. Other mechanisms are proposed, such as
an increase in plasma antioxidant activity (Franzoni et al., 2004) that prevent the scavenging of
NO by free radicals and increase the bioavailability of its precursors (Doutreleau et al., 2010)
and cofactors and thus increase its bioavailability. Apart from changes in the bioavailability of
NO, an increase of other vasodilators, such as prostanoids (McCord et al., 2006) and EDHF
(Taddei et al., 2006), may also be involved. It has been shown that prostanoids contribute to
endothelium-mediated vasodilation in response to acute exercise (Duffy et al., 1999). Studies
performed in animals have shown an increase in EDHF-mediated vasodilation induced by
training (Minami et al., 2002). The same mechanisms are speculated to also play a role in
humans. Additional studies in humans are needed to clarify this issue.
6. Conclusion
Skin microcirculation and the mechanisms controlling it remain an interesting area of
research. As a dynamic structure, it is engaged in various thermoregulatory and non-
thermoregulatory reflexes. Apart from neural control, the endothelium has been suggested
to play an important role in the regulation of vascular tone and reactivity in skin
microcirculation. However, glabrous and nonglabrous skin areas differ with respect to the
control of vascular tone, also regarding the endothelium.
Laser Doppler flowmetry (LDF) represents an easy-to-apply, noninvasive and reproducible
method to investigate skin microcirculation, yet with some disadvantages, such as great
spatial and temporal variability. Coupled to LDF, many methods have been developed over
the last 20 years in order to assess the endothelial function of human skin. Most frequently
used are iontophoresis of ACh, postocclusive reactive hyperemia, and locally induced
thermal changes. Apart from research purposes, they are often applied in clinical practice to

291
assess the endothelial (dys)function and its response to therapy. They are suggested to be
applicable indices of global microvascular function. Nevertheless, none of them is able to
detect specifically endothelial function: rather, they provide integrated indices of vascular
reactivity. This must be taken into consideration when interpreting the results obtained.
As skin microcirculation is often compromised in disease, also due to endothelial
dysfunction, strategies on how to improve it are being sought. One of the measures to
improve endothelial function seems to be participation in regular aerobic exercise, as it has
been shown to be associated with enhanced mirovascular reactivity, in particular
endothelium-dependen vasodilation, in glabrous and nonglabrous areas over the body.
The mechanisms involved in adaptations to exercise training as well as the exact contribution
of various endothelial vasodilators (NO, PGI
2
and EDHF) to the regulation of vascular tone in
human skin microcirculation, remain to be resolved. Nevertheless, they represent putative
therapeutical targets and are thus important also from the clinical point of view.
7. Acknowledgement
The author wish to thank Olga Shrestha for language editing.
8. References
Abou-Elenin, K., Xydakis, A., Hamdy, O., Economides, P.A., Horton, E.S., & Veves, A.
(2002). The effect of aspirin and various iontophoresis solution vehicles on skin
microvascular reactivity. Microvasc Res, 63, 1, 91-5.
Aoki, K., Stephens, D.P., Saad, A.R., & Johnson, J.M. (2003). Cutaneous vasoconstrictor
response to whole body skin cooling is altered by time of day. J Appl Physiol, 94,
930-934.
Bellien, J., Joannides, R., Iacob, M., Arnaud, P., & Thuillez, C. (2005). Evidence for a basal
release of a cytochrome-related endothelium-derived hyperpolarizing factor in the
radial artery in humans. Am J Physiol Heart Circ Physiol, 290, 4, H1347-52.
Berghoff, M., Kathpal, M., Kilo, S., Hilz, M.J., & Freeman, R. (2002) Vascular and neural
mechanisms of ACh-mediated vasodilation in the forearm cutaneous
microcirculation. J Appl Physiol, 92, 780-788.
Binggeli, C., Spieker, L.E., Corti, R., Sudano, I., Stojanovic, V., Hayoz, D., Lscher, T.F., &
Noll, G. (2003). Statins enhance postischemic hyperemia in the skin circulation of
hypercholesterolemic patients: a monitoring test of endothelial dysfunction for
clinical practice? J Am Coll Cardiol, 42, 1, 71-7.
Black, M.A., Green, D.J., & Cable, N.T. (2008). Exercise prevents age-related decline in nitric-
oxide-mediated vasodilator function in cutaneous microvessels. J Physiol, 586, 14,
3511-24.
Braverman, I., Keh, A., & Goldminz, D. (1990). Correlation of laser Doppler wave patterns
with underlying microvascular anatomy. J Invest Dermatol, 95, 283-286.
Braverman, I.M. (1997). The cutaneous microcirculation: Ultrastructure and Microanatomical
Organization. Microcirc, 4, 3, 329-340.
Charkoudian, N., & Johnson, J.M. (2000). Female reproductive hormones and
thermoregulatory control of skin blood flow. Exerc Sport Sci Rev, 28, 108-112.

Medical Imaging

292
Colberg, S.R., Parson, H.K., Nunnold, T., Holton, D.R., Swain, D.P., & Vinik, A.I. (2005).
Change in cutaneous perfusion following 10 weeks of aerobic training in type 2
diabetes. J Diabetes Complications, 19, 276-283.
Crakowski, J-L., Minson, C.T., Salvat-Melis, M., & Halliwill, J.R. (2006). Methodological
issues in the assessment of skin microvascular endothelial function in humans.
TRENDS Pharmacol Sci, 27, 9, 503-508.
Crakowski J.-L., Lorenzo, S., & Minson, C.T. (2007). Effects of local anaesthesia on
subdermal needle insertion pain and subsequent tests of microvascular function in
human. Eur J Pharmacol, 559, 150-154.
Dalle-Ave, A., Kubli, S., Golay, S., Delachaux, A., Liaudet, L., Waeber, B., & Feihl, F. (2004).
Acetylcholine-induced vasodilation and reactive hyperemia are not affected by
acute cyclo-oxygenase inhibition in human skin. Microcirculation, 11, 4, 327-36.
Doutreleau, S., Rouyer, O., Di Marco, P., Lonsdorfer, E., Richard, R., Piquard, F., & Geny, B.
(2010). L-arginine supplementation improves exercise capacity after a heart
transplant. Am J Clin Nutr, 91, 5, 1261-7.
Duffy, S.J., New, G., Tran, B.T., Harper, R.W., & Meredith, I. (1999). Relative contribution of
vasodilator prostanoids and NO to metabolic vasodilation in the human forearm.
Am J Physiol, 276, H663-H670.
Durand, S., Tartas, M., Bouye, P., Koitka, A., Saumet, J.L., & Abraham, P. (2004).
Prostaglandins participate in the late phase of the vascular response to
acetylcholine iontophoresis in human. J Physiol, 561, 3, 811-819.
Feletou, M., & Vanhoutte, P.M. (2009). EDHF: an update, Clin Sci, 117, 139-155.
Fichtlscherer, S., Dimmeler, S., Breuer, S., Busse, R., Zeiher, A.M., & Fleming, I. (2004).
Inhibition of cytochrome P450 2C9 improves endothelium-dependent, nitric oxide-
mediated vasodilatation in patients with coronary artery disease. Circulation, 109, 2,
178-83.
Fischer, D., Landmesser, U., Spiekermann, S., Hilfiker-Kleiner, D., Hospely, M., Mller, M.,
Busse, R., Fleming, I., & Drexler, H. (2007). Cytochrome P450 2C9 is involved in
flow-dependent vasodilation of peripheral conduit arteries in healthy subjects and
in patients with chronic heart failure. Eur J Heart Fail, 9, 8, 770-5.
Fleming, I. (2010). Molecular mechanisms underlying the activation of eNOS. Pflugers Arch
Eur J Physiol, 459, 6, 793-806.
Franzoni, F., Plantinga, Y., Femia, F.R., Bartolomucci, F., Gudio, C., Regoli, F., Carpi, A.,
Santoro, G., & Galetta, F. (2004). Plasma antioxidant activity and cutaneneous
microvascular endothelila function in athletes and sedentary controls. Biomed
Pharmacotherap, 58, 432-436.
Fritzsche, R.G., & Coyle, E.F. (2000). Cutaneous blood flow during exercise is higher in
endurance-trained humans. J Appl Physiol, 88, 2, 738-44.
Gori, T., Parker, J.D. (2002). Nitrate Tolerance A Unifying Hypothesis. Circulation, 106: 2510-
2513.
Green, D.J., Maiorana, A., ODriscoll, G., & Taylor, R. (2004). Effect of exercise training on
endothelium-derived nitric oxide function in humans. J Physiol, 561, 1, 1-25
Hendry, R,G,, & Marshall, J.M. (2004). Vasoconstrictor products of cyclo-oxygenase activity
limit acetylcholine-induced cutaneous vasodilatation in young men. Clin Sci (Lond),
107, 3, 323-30.
Hermann, D., Schlereth, T., Vogt, T., & Birklein, F. (2005). Clonidine induces nitric oxide-
and prostaglandin-mediated vasodilation in healthy human skin. J Appl Physiol, 99,
6, 2266-70.

293
Higashi, Y., & Yoshizumi, M. (2003). New methods to evaluate endothelial function: Method
for Assessing Endothelial Function in Humans Using a Strain-Gauge
Plethysmography: Nitric oxide-dependent and Independent Vasodilation. J
Pharmacol Sci, 93, 399-404.
Hillig, T., Krustrup, P., Fleming, I., Osada, T., Saltin, B., & Hellsten, Y. (2003). Cytochrome
P450 2C9 plays an important role in the regulation of exercise-induced skeletal
muscle blood flow and oxygen uptake in humans. J Physiol, 546(Pt 1), 307-14.
Hodges, G.J., Sharp, L., Stephenson, C., Patwala, A.Y. George, K.P., Goldspink, D.F., &
Cable, N.T. (2010). The effect of 48 weeks of aerobic exercise training on cutaneous
vasodilator function in pst-menopausal females, Eur J Appl Physiol. 108, 1259-1267.
Holowatz, L.A., Thompson, C.S., Minson, C.T., & Kenney, W.L. (2005) Mechanisms of
acetylcholine-mediated vasodilatation in young and aged human skin. J Physiol,
563, 965-973.
Holowatz, L.A., Thompson-Torgerson, C.S., & Kenney, W.L. (2007). Altered Mechanisms of
Vasodilation in Aged Human Skin. Exerc Sport Sci Rev, 35, 3, 119-25.
Holowatz, L.A., Thompson-Torgerson, C.S., & Kenney, W.L. (2008). The human cutaneous
circulation as a model of generalized microvascular function. J Appl Physiol, 105, 1,
370-2.
Johnson, J.M., & Proppe, D.W. (1996). Cardiovascular adjustments to heat stress. In:
Handbook of Physiology. Environmental Physiology, 215-243, Am. Physiol. Soc.,
Bethesda.
Johnson, J.M., & Kellogg, D.L.Jr. (2010). Thermoregulatory and thermal control in the
human cutaneous circulation. Front Biosci, S2: 825-853.
Kalia, Y.N., Naik, A., Garrison, J., & Guy, R.H. (2004). Iontophoretic drug delivery. Adv Drug
Deliv Rev, 56, 619-658.
Kellogg, D.L.Jr., Zhao, J.L., Coey, U, & Green, J.V. (2005). Acetylcholine-induced
vasodilation is mediated by nitric oxide and prostaglandins in human skin. J Appl
Physiol. 98, 629-632.
Kellogg, D.L.Jr. (2006). In vivo mechanisms of cutaneous vasodilation and vasoconstriction
in humans during thermoregulatory challenges. J Appl Physiol. 100, 1709-1718.
Kellogg, D.L.Jr., Zhao, J.L., & Wu, Y. (2009). Roles of nitric oxide synthase isoforms in
cutaneous vasodilation induced by local warming of the skin and whole body heat
stress in humans. J Appl Physiol, 107, 5, 1438-44.
Kernick, D.P., Tooke, J.E., & Shore, A.C. (1999). The biological zero signal in laser Doppler
fluximetry origins and practical implications. Pflugers arch Eur J Physiol, 437, 624-
631.
Khan, F., Davidson, N.C. , Littleford, R.C., Litchfield, S.J., Struthers, A.D., & Belch., J.J. (1997).
Cutaneous vascular responses to acetylcholine are mediated by a prostanoid-
dependent mechanism in man. Vasc Med, 2, 82-6.
Khan, F., Newton, D.J., Smyth, E.C., & Belch, J.J. (2004). Influence of vehicle resistance on
transdermal iontophoretic delivery of acetylcholine and sodium nitroprusside in
humans. J Appl Physiol, 97, 883-887.
Kvandal, P., Landsverk, S.A., Bernjak, A., Stefanovska, A., Kvernmo, H.D., & Kirkeben,
K.A. (2006). Low-frequency oscillations of the laser Doppler perfusion signal in
human skin. Microvasc Res, 72, 120-127.
Kvernmo, H.D., Stefanowska, A., & Kirkeboen, K.A. (2003). Enhanced endothelial activity
reflected in cutaneous blood flow oscillations of athletes. Eur J Appl Physiol, 90, 16-22.

Medical Imaging

294
Larsen, B.T., Miura, H., Hatoum, O.A., Campbell, W.B., Hammock, B.D., Zeldin, D.C., Falck,
J.R., & Gutterman, D.D. (2006). Epoxyeicosatrienoic and dihydroxyeicosatrienoic
acids dilate human coronary arterioles via BK(Ca) channels: implications for
soluble epoxide hydrolase inhibition. Am J Physiol Heart Circ Physiol, 290, 2, H491-9.
Lenasi, H. (2010). The role of nitric oxide- and prostacyclin-independent vasodilatation in
the human cutaneous microcirculation: effect of cytochrome P450 2C9 inhibition.
Clin Physiol Funct Imaging, 29, 4, 263-70.
Lenasi, H., & trucl, M. (2010). Regular physical activity alters the postocclusive reactive
hyperemia of the cutaneous microcirculation. Clin Hemorheol Microcirc, 45, 365-374.
Leslie, S.J., Affolter, J., Denvir, M.A., & Webb, D.J. (2003). Validation of laser Doppler
flowmetry coupled with intra-dermal injection for investigating effects of vasoactive
agents on the skin microcirculation in man. Eur J Clin Pharmacol, 59, 99-102.
Loot, A.E., Schreiber, J.G., Fisslthaler, B., & Fleming, I. (2009). Angiotensin II impairs
endothelial function via tyrosine phosphorylation of the endothelial nitric oxide
synthase. J Exp Med, 206, 13, 2889-2896.
Lorenzo, S., & Minson, C.T. (2007). Human cutaneous reactive hyperaemia: role of BK
Ca

channels and sensory nerves. J Physiol, 585, 1, 295-303.
Lorenzo, S., & Minson, C.T. (2010). Heat acclimation improves cutaneous vascular function
and sweating in trained cyclists. J Appl Physiol, 109, 6, 1736-43.
McCord, G.R., Cracowski, J-L. & Minson, C.T. (2006). Prostanoids contribute to cutaneus active
vasodilation in humans. Am J Physiol Regul Integr Comp Physiol, 291, R596-R602.
Medow, M.S., Taneja, I., & Stewart, J.M. (2007). Cyclooxygenase and nitric oxide synthase
dependence of cutaneous reactive hyperemia in humans. Am J Physiol Heart Circ
Physiol, 293, H425-H432.
Minami, A., Ishimura, N., Harada, N., Sakamoto, S., Niwa, Y., & Nakaya, Y. (2002). Exercise
training improves acetylcholine-induced endothelium-dependent
hyperpolarization in type 2 diabetic rats, Otsuka Long-Evans Tokushima fatty rats.
Atherosclerosis, 162, 1, 85-92.
Minson, C.T. (2010). Thermal provocation to evaluate microvascular reactivity in human
skin. J Appl Physiol, 109: 1239-1246.
Morris, S.J., & Shore, A.C. (1996). Skin blood flow responses to the iontophoresis of
acetylcholine and sodium nitroprusside in man: Possible mechanisms. J Physiol,
496, 531-542.
Nishikawa, Y., Stepp, D.W., & Chilian, W.M. (2000). Nitric oxide exerts feedback inhibition
on EDHF-induced coronary arteriolar dilation in vivo. Am J Physiol Heart Circ
Physiol, 279, H459H465.
Noon, J.P., Haynes, W.G., Webb, D., & Shore, A.C. (1996). Local inhibition of nitric oxide
generation in man reduces blood flow in finger pulp but not in hand dorsum skin. J
Physiol, 490, 501-8.
Noon, J.P., Walker, B.R., Hand, M.F., & Webb, D.J. (1998). Studies with iontophoretic
administration of drugs to human dermal vessels in vivo: cholinergic
vasodilatation is mediated by dilator prostanoids rather than nitric oxide. Br J Clin
Pharmacol, 45, 545-550.
Osanai, T., Fujita, N., Fujiwara, N., Nakano, T., Takahashi, K., Guan, W., & Okomura, K.
(2000). Cross talk of shear-induced production of prostacyclin and nitric oxide in
endothelial cells. Am J Physiol Heart Circ Physiol, 278, H233H238.

295
Passauer, J., Pistrosch, F., Lssig, G., Herbrig, K., Bssemaker, E., Gross, P., & Fleming, I.
(2005). Nitric oxide- and EDHF-mediated arteriolar tone in uremia is unaffected by
selective inhibition of vascular cytochrome P450 2C9. Kidney Int, 67, 5, 1907-12.
Passauer, J., Bussemaker, E., Lassig, G., Pistrosch, F., Fauler, J., Gross, P., & Fleming, I.
(2003). Baseline blood flow and bradykinin-induced vasodilator responses in the
human forearm are insensitive to the cytochrome P450 2C9 (CYP2C9) inhibitor
sulphaphenazole. Clin Sci (Lond), 105(4), 513-8.
Pohl, U., & deWit, C. (1999). A Unique Role of NO in the Control of Blood Flow. News
Physiol Sci, 14, 74-80.
Ramsay, J.E., Ferrell, W.R., Greer, I.A., & Sattar, N. (2002). Factors critical to iontophoretic
assessment of vascular reactivity: implications for clinical studies of endothelial
dysfunction. J Cardiovasc Pharmacol, 39, 9-17.
Roche, D.M., Rowland, T.W., Garrard, M., Marwood, S., & Unnithan, V.B. (2010). Skin
microvascular reactivity in trained adolescents. Eur J Appl Physiol, 108, 1201-1208.
Roddie, I.C. Circulation to skin and adipose tissue. In: Handbook of Physiology. J. T. Shepherd,
and S. R. Geiger (Eds.). Bethesda: American Physiology Society, 1983, pp: 285-309.
Rossi, M., Santoro, G., Maurizio, S., & Carpi, A. (2006). Spectral analysis of skin blood
flowmotion before and after exercise in healthy trained subjects. Int J Sports Med,
27: 540-545.
Rossi, M., Carpi, A., Galetta, F., Franzoni, F., & Santoro, G., (2008). Skin vasomotion
investigation: A useful tool for clinical evaluation of microvascular endothelial
function? Biomed Pharmacother, 62: 541-545.
Rossi, M., Bradbury, A., Magagna, A., Pesce, M., Taddei, S., & Stefanovska A. (2011).
Investigation of skin vasoreactivity and blood flow oscillations in hypertensive
patients: effect of short-term antihypertensive treatment. J Hypertens, 29, 8, 1569-76.
Rowell, LB. (1993). Human cardiovascular control. Oxford University Press, New York,
Oxford.
Saumet, J.L., Kellogg, D.L.Jr., Taylor, W.F., & Johnson, J.M. (1988). Cutaneous laser-Doppler
flowmetry: influence of underlying muscle blod flow. J Appl Physiol, 65, 478-481.
Schrage, W.G., Dietz, N.M., Eisenach, J.H., & Joyner, M.J. (2005). Agonist-dependent
variablity of contributions of nitric oxide and prostaglandins in human skeletal
muscle. J Appl Physiol, 98, 4, 1251-7.
Serup, J., Jemec, G.B.E., & Grove, G.L. (eds.) (2006). Handbook of Noninvasive Methods and the
Skin, CRC Taylor & Francis Group, ISBN 0-8493-1437-2, Boca Raton, London, New
York.
Shepherd, A.P., & Oberg, P.A. (Eds.) (1990). Laser-Doppler blood flowmetry, Kluwer Academic
Publishers, ISBN 0-7923-0508-6, Boston Dordrecht London.
Shibasaki, M., Low, D.A., Davis, S.L., & Crandall, C.G. (2008). Nitric oxide inhibits cutaneous
vasoconstriction to exogenous norepinephrine. J Appl Physiol, 105, 5, 1504-8.
Szerafin, T., Erdei, N., Fulop, T., Pasztor, E.T., Edes, I., Koller, A., & Bagi, Z. (2006).
Indcreased cyclo-oxygenase-2 expression and prostaglandin-mediated dilation in
coronary arteries of patients with diabetes mellitus. Circ Res, 99, 112-117.
Taddei, S., Versari, D., Cipriano, A., Ghiadoni, L., Galetta, F., Franzoni, F., Magagna, A.,
Virdis, A., & Salvetti, A. (2006). Identification of a cytochrome P450 2C9-derived
endothelium-derived hyperpolarizing factor in essential hypertensive patients. J
Am Coll Cardiol, 48, 508-15.

Medical Imaging

296
Tang, M., Cyrus, T., Yao, Y., Vocun, L., & Pratico, D. (2005). Involvement of thromboxane
receptor in the proatherogenic effect of isoprostane F2alphaIII: evidence from
apolipoprotein E- and LDL receptor-deficient mice. Circulation, 112: 2867-2874.
Tew, G.A., Saxton, J.M., Klonizakis, M., Moss, J., Ruddock, A.D., & Hodges, G.J. (2011).
Aging and aerobic fitness affect the contribution of noradrenergic sympathetic
nerves to the rapid cutaneous vasodilator response to local heating. J Appl Physiol,
110, 1264-1270.
Treu, C.M., Lupi, O., Bottino, D.A., & Bouskela, E. (2011). Sidestream dark field imaging: the
evolution of real-time visualization of cutaneous microcirculation and its potential
application in dermatology. Arch Dermatol Res, 303, 69-78.
Turner, J., Belch, J.J.F., & Khan, F. (2008). Current Concepts in Assessment of Microvascular
Endotehlial Function Using Laser Doppler Imaging and Iontophoresis. Trends
Cardiovasc Med, 18, 109-116.
Urakami-Harasawa, L., Shimokawa, H., Nakashima, M., Egashira K, & Takeshita A. (1997).
Importance of endothelium-derived hyperpolarizing factor in human arteries. J
Clin Invest, 100, 2793-2799.
Vanhoutte, P.M. (1989). Endothelium and control of vascular function. Hypertension, 13, 658-
667.
Vassalle, C., Lubrano, V., Domenici, C., & L'Abbate, A. (2003). Influence of chronic aerobic
exercise on microcirculatory flow and nitric oxide in humans. Int J Sports Med, 24, 1,
30-5.
Wang, J.S. (2005). Effects of exercise training and detraining on cutaneous microvascular
function in man: the regulatory role of endothelium-dependent dilation in skin
vasculature. Eur J Appl Physiol, 93, 429-34.
Wong, B.J., Wilkins, B.W., Holowatz, L.A., & Minson, C.T. (2003). Nitric oxide inhibitions
does not alter the reactive hyperemic response in the cutaneous circulation. J Appl
Physiol, 95, 504-510.
Yamada, T., Fujino, T., Yuhki, K., Hara, A., Karibe, H., Takahata, O., Okada, Y., Xiao, C.Y.,
Takayama, K., Taniguchi, T., Shiokoshi, T., Ohsaki, Y., Kikuchi, K., Narumyia, S., &
Ushikubi, F. (2003). Thromboxane A
2
regulates vascular tone via its inhibitory effect
on the expression of inducible nitric oxide synthase. Circulation, 108, 2381-2386.
Yamazaki, F., & Sone, R. (2006). Different vascular responses in glabrous and nonglabrous skin
with increasing core temperature during exercise. Eur J Appl Physiol, 97, 5, 582-90.
Yang, Q., Yim, A.P.C., & He, G.-W. (2007). The Significance of Endothelium-Derived
Hyperpolarizing Human Circulation. Curr Vasc Pharmacol, 5, 85-92.
Yvonne-Tee, G.B., Rasool, A.H.G., Halim, A.S., & Rahman, A.R.A. (2006). Noninvasive
assessment of cutaneous vascular function in vivo using capillaroscopy,
plethysmography and laser-Doppler instruments: Its strengths and weaknesses.
Clin Hemorheol Microcirc, 34, 457-473.
Yvonne-Tee, G.B., Rasool, A.H.G., Halim, A.S., Wong, A.R., & Rahman, A.R.A. (2008).
Method optimization on the use of postoclussive hyperemia model to assess
microvascular function. Clin Hemorheol Microcirc, 38, 119-133.
Zhao, J.L., Pergola, P.E., Roman, L.J., & Kellogg, D.L.Jr. (2004). Bioactive nitric oxide
concentration does not increase during reactive hyperemia in human skin. J Appl
Physiol, 96, 628-632.
14
Determination of Limb Hemodynamics
During Rhythmical Muscle Contractions
Assessed by Doppler Ultrasound
Takuya Osada
1,2
and Gran Rdegran
2,3

1
Department of Sports Medicine for Health Promotion, Tokyo Medical University, Tokyo
2
The Copenhagen Muscle Research Centre, Rigshospitalet
University of Copenhagen, Copenhagen
3
The Clinic for Heart Failure and Valvular disease, Skne University hospital
and Department of Cardiology, IKVL, Lund University, Lund
1
Japan
2
Denmark
3
Sweden
1. Introduction
Ultrasound Doppler instruments may provide valuable measurements of hemodynamic
changes with high temporal resolution. In the clinical setting, Doppler velocity profiles may
be useful for the evaluation of cardiac function, valvular disease and major conduit vascular
disease. Furthermore, arterial stiffness and limb conduit arterial vessel dilatation following
cuff-ischemia as measured by 2-dimentional echo imaging and Doppler ultrasound method
are useful as surrogate parameters in atherosclerosis and hypertension.
Lifestyle related diseases such as obesity, hypertension, hyperlipidemia, and diabetes
mellitus may be prevented by physical activity. Furthermore, a decreased amount of
physical fitness or long term bed rest may lead to reduction of exercise tolerance (maximum
oxygen consumption) due to cardiovascular-, respiratory- and muscle-dysfunction.
Circulatory changes following physical activity may also be used for studying
cardiovascular regulation. For instance, cardiac output increases with increasing exercise
intensity along with enhanced skeletal muscle vasodilatation and muscle pumping in the
exercising muscle. Perfusion in the active muscle is furthermore one indicator of oxygen
delivery to the muscles. Therefore, non-invasive Doppler measurements of blood velocity
and flow in the feeding conduit arteries to working skeletal muscle may give us valuable
information of the hemodynamic response in peripheral upper or lower limbs.
2. Exercise model
Determinations of blood flow to contractile muscles are the most important focus of the
present chapter. Precise and stable measurements in conduit arteries assessed by Doppler
ultrasound may be performed during exercise. Whole body exercise methods such as

Medical Imaging

298
walking and running on a treadmill would be even better for the investigation of
exercising leg blood flow. However, methods do not easily allow measurement of upper-
and lower-limb blood flow using Doppler ultrasound in these models as motion artifacts
are present. There is also difficulty in fixing the ultrasound Doppler probe. Local lower
limb muscle blood flow may be measured using the one-legged, dynamic knee-extensor
exercise model described by Andersen and Saltin (1985). This exercise model allows stable
measurements of femoral arterial blood velocity using Doppler ultrasound (Figure 1).
Therefore, all hemodynamic data described in this chapter are from dynamic knee-
extension exercise.

Fig. 1. Rhythmic thigh muscle contractions (one-legged dynamic knee extensor) model.
Modified figure, reprinted with permission from Osada, T. & Rdegran, G. (2005). Japanese
Journal of Physiology, Vol. 55, No. 1, pp. 19-28. by The Physiological Society of Japan.
3. Participants
All healthy male volunteers were familiarized with the above mentioned one-legged,
dynamic knee-extensor exercise model before starting the experiments (Andersen et al.,
1985; Andersen & Saltin, 1985). They trained at 30 and 60 contractions per minute at
external target work rates of 10, 20, 30 and 40 W. Their hamstring muscles were allowed
to fully relax, so that only the knee extensor musles performed the work. The subjects
thigh was positioned horizontally with the knee joint bent. The lower leg moved up to ~
60 angle from the bent knee joint. All subjects provided written informed consent to
participate in the study, which was approved by the Ethical Committees of Copenhagen
and Frederiksberg (KF-01-013/96). The study adhered to the principles of the Helsinki
Declaration.
During Rhythmical Muscle Contractions Assessed by Doppler Ultrasound

299
4. Hemodynamic measurements
4.1 Doppler instrumentation
The measurements were performed using a Doppler ultrasound instrument (Model CFM
800, Vingmed Sound, Horten, Norway) equipped with an annular phased array transducer
(Vingmed Sound) probe (11.5-mm diameter). The imaging frequency was 7.5 MHz and the
Doppler frequencies varied between 4.0 and 6.0 MHz (high-pulsed repetition frequency
mode, 4 - 36 kHz). Blood velocity was measured with the probe at the lowest possible
insonation angle and always < 60 (Gill, 1985). The mean value of the insonation angle was ~
50, which remained constant throughout the experiments for each individual. The probe
position was stable and the sample volume was precisely positioned in the center of the
vessel and adjusted to cover the diameter width of the vessel.
4.2 Measurement of blood velocity, vessel diameter and blood flow in femoral artery
The measurements of blood velocity and blood flow in the femoral artery using Doppler
ultrasound has previously been validated and shown to produce accurate absolute values
both at rest and during leg exercise such as rhythmical thigh muscle contractions (Hughson
et al., 1997; Osada, 2004; Rdegran, 1997; Shoemaker et al., 1994; Walle & Wesche, 1988).
Compared with thermodilution, the high temporal resolution of Doppler ultrasound
additionally enables continuous measurement of blood velocity throughout the kicking
cycle during exercise (Osada & Rdegran, 2002; Rdegran, 1997; Rdegran & Saltin, 1998;
Roberg et al., 1997; Shoemaker et al., 1994; Walle & Wesche, 1988).
The angle-corrected, time and space-averaged, and amplitude-weighted mean (V
mean
) and
maximum (V
max
; outer envelope) blood velocities, respectively, were measured. V
mean
was
defined by averaging the mean blood velocity trace (Osada & Rdegran, 2002; Rdegran,
1997). V
max
was defined as the maximum outer envelope (Leyk et al., 1992; Osada et al.,
1999; Osada et al., 2003). The V
max
obtained in the present study was expressed as the blood
velocity measured at the center of the vessel. Each blood velocity parameter was measured
in relation to the blood pressure curve. The site of blood velocity and vessel diameter
measurements in the femoral artery was distal to the inguinal ligament but above the
bifurcation into the branch of the superficial and deep femoral artery. This location
minimizes turbulence from the femoral bifurcation and the influence of blood flow from the
inguinal region. In addition, the arterial diameter is not affected by the contractions and
relaxations at this site located proximal to the muscle.
The blood velocity measurements were performed for approximately 2 to 3 min, when
steady-state had been reached after 3 min of one-legged, dynamic knee extensor (Osada,
2004; Rdegran & Saltin, 1998), as previously described (Rdegran, 1997). The systolic and
diastolic diameters of the femoral artery were measured on a monitor relative to the ECG
at rest. The mean vessel diameter was calculated in relation to the temporal duration of
the blood pressure curve as; [(systolic vessel diameter value 1/3) + (diastolic vessel
diameter value 2/3)] (Rdegran, 1997). The diameters were measured under
perpendicular insonation at rest before exercise. The value of the vessel diameter at rest
(pre-exercise) was used to calculate femoral arterial blood flow during rest and during
one-legged, dynamic knee extensor, since the diameter does not significantly vary
between rest and steady-state exercise (Hughson et al., 1997; Isnard et al., 1996; Leyk et al.,

Medical Imaging

300
1992; MacDonald et al., 1998; Osada et al., 1999; Rdegran, 1997). Steady-state leg blood
flow was calculated by multiplying the cross-sectional area [Area = (vessel
diameter/2)
2
] of the femoral artery, with the angle corrected, time and space-averaged,
and amplitude (signal intensity) weighted mean-blood velocity, where blood flow =
mean-blood velocity cross sectional area.
5. Validation of blood flow during incremental one-legged, dynamic knee-
extensor as measured by Doppler ultrasound
In previous reports, central and peripheral hemodynamic measurements have been
demonstrated using the thermodilution technique for cardiac output and limb blood flow
during incremental cycling ergometer exercise. However, this invasive technique has the
limitation of poor time resolution of blood flow. Several techniques have previously been
developed that enable estimates to be made of arterial inflow, venous outflow, and local
blood flow within a muscle. Whereas many of the techniques are impaired by different
methodological limitations, the indicator methods and the ultrasound Doppler method have
both been found to give repeatable measurements of the same magnitude during both rest
and dynamic knee extensor exercise (Rdegran, 1997).
The Doppler ultrasound is unique as it allows continuous blood flow measurements non-
invasively with a high temporal resolution. With continuous measurements, transitional
changes in blood flow can be characterized (Walle & Wesche, 1988; Wesche, 1986). The
techniques precision and accuracy have furthermore been improved by sampling of the
blood velocity continuously during dynamic knee extensor exercise (Rdegran, 1997).
Doppler ultrasound may therefore be suitable for the investigation of transitional changes of
limb hemodynamics in the conduit brachial, femoral and poplitial artery during rest and
upper forearm or lower limb exercise.
Temporal blood flow changes due to muscle contractions during dynamic knee extensor
exercise have been well described (Osada, 2004). Doppler ultrasound has furthermore
been used to examine the hyperaemic response at the onset of exercise (Hughson et al.,
1993; Rdegran & Saltin, 1998), steady-state (Osada & Rdegran, 2002) and recovery after
exercise (Osada et al., 2003). At the onset of exercise as well as recovery after exercise,
variations in blood flow may be influenced by changes in blood pressure, heart rate or
strength of muscle contractions.
Furthermore, femoral arterial blood velocity and blood flow increases linearly with
incremental target exercise intensities of work rate (workload) during steady-state rhythmic
thigh muscle contractions. This implies that an enhanced vasodilatation is elicited, in
relation to the increased average muscle force exerted at higher workloads, to meet the
elevated metabolic activity.
Figure 2 shows the relationship between limb femoral arterial blood flow and target
workload (10, 20, 30 and 40 W) in relation to rhythmic thigh muscle contractions at 60
contractions per minute. In addition, thermodilution blood flow measurements obtained
under similar experimental conditions by Andersen and Saltin (1985) are closely related to
those obtained by Doppler ultrasound. Thus, blood flow measured by Doppler ultrasound
is valid not only at rest but also during incremental one-legged dynamic knee extensor
exercise.

301

Fig. 2. Blood flow during incremental one-legged dynamic knee extensor exercise, estimated
from mean blood velocity and averaged on a beat-by-beat basis or continuously measured
and analyzed in relation to muscle contraction cycle. Reprinted with permission from
Rdegran, G. (1997), Journal of Applied Physiology, Vol. 83, No. 4, pp. 1383-1388, by The
American Physiological Society.
6. Variation of blood velocity and blood flow modified by muscle contraction-
relaxation cycles
It is well known that conduit arterial blood velocity and blood flow is markedly influenced
by intramuscular pressure, as well as the superimposed influence of the arterial blood
pressure. The high intramuscular pressure during muscle contractions may consequently
temporarily reduce or even reverse the blood velocity, depending on the relationship
between the intramuscular- and arterial blood pressure. The major extent of the blood
velocity and flow consequently occurs during the muscle relaxation phase. Figure 3 shows
that mean blood velocity increased to its highest value at the systolic blood pressure phase
during muscle relaxation, and significantly decreased to its lowest value at the diastolic
blood pressure phase during muscle contraction. Mean blood velocity showed an
intermediate value at the systolic blood pressure phase during muscle contraction and at the
diastolic blood pressure phase during muscle relaxation, respectively. The blood velocity
curve was furthermore retrograde in the diastolic blood pressure phase during muscle
contraction. The figure below shows the contribution to the magnitude of the physiological
variability in blood velocity and flow by the contraction-relaxation-induced variations in

Medical Imaging

302
muscle force, and consequently the intramuscular pressure variations, along with the
superimposed influence of the blood pressure as well as the tonic influence of the state of
vasodilatation. Moreover, the exercise blood velocity and blood flow fluctuations due to
muscle contractions and relaxations, with the superimposed influence of the cardiac cycle,
are described in Figure 4.

Fig. 3. Continuous recording of mean blood velocity, blood pressure and muscle force
during steady-state one-legged dynamic knee extensor at 20 W and 60 contractions per
minute in one subject: sys, systole; dia, diastole; cont, muscle contraction; relax, muscle
relaxation. Modified figure, reprinted with permission from Osada, T. & Rdegran, G.
(2006). Journal of Sports Medicine and Physical Fitness, Vol. 46 No. 4, pp. 590-597, by Edizioni
Minerva Medica.

303

Fig. 4. Flow profile in the blood velocity (V
max
and V
mean
) in the limb femoral artery for the
cardiosystolic and cardiodiastolic phases during the muscle contraction and muscle
relaxation phases of one-legged dynamic knee-extensor exercise at 20 W and 60 contractions
per minute. Reprinted with permission from Osada, T. & Rdegran, G. (2006), Journal of
Physiological Science, Vol. 56, No. 3, pp. 195-203, by The Physiological Society of Japan.
As seen in Figure 4, blood velocities (V
max
and V
mean
) during the cardiosystolic and
cardiodiastolic phases were measured continuously in parallel with the blood pressure
curve during the muscle contraction and muscle relaxation phases determined from the
electromyography (EMG) and the muscle force curve. Blood velocity fluctuated in relation
to the state of vasodilatation and the muscle contraction-relaxation duty cycles, indicated by
the oscillations in muscle force.

Medical Imaging

304
Four variations (AD) in the coupling between the blood pressure curve and the state of
contraction and relaxation were indicated; (A) the cardiosystolic phase during muscle
contraction, (B) the cardiodiastolic phase during muscle contraction, (C) the cardiosystolic
phase during muscle relaxation, and (D) the cardiodiastolic phase during muscle
relaxation. V
max
and V
mean
were determined as the average transient maximum (outer
envelope) and mean (amplitude-weighted, time-and-spatial averaged) angle-corrected
blood velocity values, respectively, for the cardiosystolic and cardiodiastolic phases. The
formation of the blood velocity profile and flow was modified by the intramuscular
pressure, as indicated by the muscle force curve, and the superimposed influence of the
blood pressure in relation to the cardiosystolic and cardiodiastolic phases. The arrow
down () and up () indicates the influence on the blood velocity and flow, depending on
the magnitude of, and temporal relation between, the intramuscular pressure and the
arterial blood pressure, respectively.

Fig. 5. Blood flow for the cardiosystolic and cardiodiastolic phases during the muscle
contraction and muscle relaxation phases of one-legged dynamic knee-extensor exercise at
60 contractions per minute, *P < 0.05. Reprinted with permission from Osada, T. &
Rdegran, G. (2006), Journal of Physiological Science, Vol. 56, No. 3, pp. 195-203, by The
Physiological Society of Japan.
There is furthermore information available regarding differences in blood flow variations
induced by muscle contractions and muscle relaxations at incremental exercise intensities at
5, 10, 20, 30 and 40 W (Fig. 5). These data clearly demonstrated that blood flow in the
femoral artery increases during muscle relaxation, rather than muscle contraction. Also,
there is a large difference in blood flow magnitude between cardiosystole and cardiodiastole
at incremental exercise intensities. In consequence, the average exercise blood flow value
may be over- or under-estimated if the time for measurement is restricted to only a single
phase among the muscle contraction and muscle relaxation phases and the cardiosystolic
and cardiodiastolic phases. This evidence of blood flow response due to muscle mechanical
factor and cardiac pumping cycle is a clear physiological phenomenon, but the blood flow
response in relation to the variations in the force of muscle contractions ( intramuscular
pressure, workload) should be further clarified.

305
Of further interest is whether sudden changes in workload and rhythm may exert influence
upon the magnitude of blood flow. Such information may be useful for the temporal
evaluation of exercising blood flow during consecutive rhythmic muscle contractions with
large variation due to poor exercise performance.
7. Changes in blood flow due to spontaneous changes of workload
Limb femoral arterial blood flow during steady-state rhythmic thigh muscle contractions
increases linearly with incremental target workloads (work rates) (Hughson et al., 1996;
1997; Osada & Rdegran, 2002; Rdegran, 1997; Shoemaker et al., 1994; Tschakovsky et al.,
2006). This implies that enhanced vasodilatation is elicited in relation to the increased
average muscle force, exerted at higher workloads, to meet the elevated metabolic activity
(Fig. 2). However, these blood flow values are a mean of steady-state exercising blood flow
measurements, and temporary muscle contraction-induced blood flow variations may
therefore be conveyed in the mean average blood flow value (Fig. 2, 4 and 5). For human
voluntary exercise, it is of value to consider how variations in repeated muscle contractions
at target muscle strength (muscle force) directly influence exercise blood flow in conduit
arteries. Therefore, weve investigated whether sudden physiological and spontaneous
changes in exercise rhythm, and consequently workload, temporarily alter blood flow to the
working muscle.

Fig. 6. Relationship between steady-state limb femoral arterial blood flow and the
achieved workload during incremental one-legged dynamic knee-extensor exercise at 30
or 60 contractions per minute during a minute measurement in one subject (60 samplings
at each workload). Modified figure, reprinted with permission from Osada, T. &
Rdegran, G. (2009), Clinical Physiology and Functional Imaging, Vol. 29, No. 4, pp. 277-292,
by John Wiley & Sons Ltd.

Medical Imaging

306
The results showed that limb femoral arterial blood flow increased positively and linearly
(dotted line) with increasing target workload. However, limb blood flow was inversely and
linearly related (solid line) to the actually achieved workload, when measured over 60
consecutive contraction-relaxation cycle bouts for each target intensity at 30 and 60
contractions per minute, respectively (Figure 6).
Thus any sudden spontaneous increase or decrease in the achieved workload transiently
altered the relationship between limb femoral arterial blood flow and the achieved
workload. The influence upon the magnitude of limb femoral arterial blood flow, due to
fluctuations in the achieved workload from the target workload was similar at target
workload sessions of 30 and 60 contractions per minute, respectively. These findings
indicate that a transient sudden increase in the workload during rhythmic muscle
contractions more rapidly impedes limb femoral arterial blood flow, than that vasodilatation
may be elicited to restore the intensity related steady-state limb blood flow response, in
relation to the average metabolic activity. This evidence in human applied science may
contribute to the evaluation of exercise hemodynamics for rhythmic, dynamic-isotonic
exercise training, leading to exercise prescriptions (muscle contraction frequency or muscle
contraction intensity) for healthy participants as well as for patients, requiring additional
physical activity in the rehabilitation or clinical setting.
8. Conclusion
Measuring exercise blood velocity and blood flow in working skeletal muscle assessed by
Doppler ultrasound can be performed, however there are still limitations with regards to the
exercise model, target vessel, muscle contraction-intensity and frequencies, in upper or
lower limbs.
9. Acknowledgment
The authors would like to acknowledge the support given by Professor Bengt Saltin, the
staff of The Copenhagen Muscle Research Centre, as well as the volunteers who
participated in this study. This study was supported by the Danish National Research
Foundation Grant 504-14 (CMRC), in part by a Grant-in-Aid for Young Scientists (B) (No.
16700471) in 2004-2005 as well as by the Excellent Young Researchers Overseas Visit
Program (No. 21-8285) in 2010 in Scientific Research from the Ministry of Education,
Culture, Sports, Science and Technology (MEXT) of Japan and the Japan Society for the
Promotion of Science (JSPS) (T. Osada).
10. References
Andersen, P., Adams, R.P., Sjgaard, G., Thorboe, A. & Saltin, B. (1985). Dynamic knee
extension as model for study of isolated exercising muscle in humans. Journal of
Applied Physiology, Vol. 59, No. 5, pp. 1647-1653, ISSN 8750-7587
Andersen, P. & Saltin, B. (1985). Maximal perfusion of skeletal muscle in man. Journal of
Physiology (London), Vol. 366, pp. 233-249, ISSN 0022-3751

307
Gill, R.W. (1985). Measurement of blood flow by ultrasound: Accuracy and sources of error.
Ultrasound in Medicine and Biology, Vol. 11, No. 4, (July-August) pp. 625-641, ISSN
0301-5629
Hughson, R.L., Cochrane, J.E. & Butler G.C. (1993). Faster O2 uptake kinetics at onset of
supine exercise with than without lower body negative pressure. Journal of Applied
Physiology, Vol. 75, No. 5, pp. 1962-1967, ISSN 8750-7587
Hughson, R.L., MacDonald, M.J., Shoemaker, J.K. & Borkhoff, C. (1997). Alveolar oxygen
uptake and blood flow dynamics in knee extension ergometry. Methods of
Information in Medicine, Vol. 36, No. 4-5, pp. 364-367, ISSN 0026-1270
Hughson, R.L., Shoemaker, J.K., Tschakovsky, M.E. & Kowalchuk, J.M. (1996). Dependence
of muscle VO
2
on blood flow dynamics at onset of forearm exercise. Journal of
Applied Physiology, Vol. 81, No. 4, pp. 1619-1626, ISSN 8750-7587
Isnard, R., Lechat, P., Kalotka, H., Chikr, H., Fitoussi, S., Salloum, J., Golmard, J .L., Thomas,
D. & Komajda, M. (1996). Muscular blood flow response to submaximal leg exercise
in normal subjects and in patients with heart failure. Journal of Applied Physiology,
Vol. 81, No. 6, pp. 2571-2579, ISSN 8750-7587
Leyk, D., Efeld, D., Baum, K. & Stegemann, J. (1992). Influence of calf muscle contractions
on blood flow parameters measured in the arteria femoralis. International Journal of
Sports Medicine, Vol. 13, No. 8, pp. 588-593, ISSN 0172-4622
MacDonald, M.J., Shoemaker, J.K., Tschakovsky, M.E., & Hughson R.L. (1998). Alveolar
oxygen uptake and femoral artery blood flow dynamics in upright and supine leg
exercise in humans. Journal of Applied Physiology, Vol. 85 No. 5, pp. 1622-1628, ISSN
8750-7587
Osada, T., Katsumura, T., Hamaoka, T., Inoue, S., Esaki, K., Sakamoto, A., Murase, N.,
Kajiyama, J., Shimomitsu, T. & Iwane, H. (1999). Reduced blood flow in abdominal
viscera measured by Doppler ultrasound during one-legged knee extension. Journal
of Applied Physiology, Vol. 86, No. 2, pp. 709-719, ISSN 8750-7587
Osada, T., Katsumura, T., Murase, N., Sako, T., Higuchi, H., Kime, R., Hamaoka, T. &
Shimomitsu, T. (2003). Post-exercise hyperemia after ischemic and non-ischemic
isometric handgrip exercise. Journal of Physiological Anthropology and Applied Human
Science, Vol. 22, No. 6, pp. 299-309, ISSN 1345-3475
Osada, T. (2004). Muscle contraction-induced limb blood flow variability during dynamic
knee extensor. Medicine and Science in Sports and Exercise, Vol. 36, No. 7, 1149-1158,
ISSN 0195-9131
Osada, T. & Rdegran, G. (2002). Femoral artery inflow in relation to external and total work
rate at different knee extensor contraction rates. Journal of Applied Physiology, Vol.
92, No. 3, pp. 1325-1330, ISSN 8750-7587
Osada, T. & Rdegran, G. (2006). Alterations in the blood velocity profile influence the blood
flow response during muscle contractions and relaxations. Journal of Physiological
Science, Vol. 56, No. 3, pp. 195-203, ISSN 1880-6546
Osada, T. & Rdegran, G. (2006). Differences in exercising limb blood flow variability
between cardiac and muscle contraction cycle related analysis during dynamic
knee extensor. Journal of Sports Medicine and Physical Fitness, Vol. 46, No. 4, pp. 590-
597, ISSN 0022-4707

Medical Imaging

308
Osada, T. & Rdegran, G. (2009). Femoral artery blood flow and its relationship to
spontaneous fluctuations in rhythmic thigh muscle workload. Clinical Physiology
and Functional Imaging, Vol. 29, No. 4, pp. 277-292, ISSN 1475-0961
Rdegran, G. (1997). Ultrasound Doppler estimates of femoral artery blood flow during
dynamic knee extensor exercise in humans. Journal of Applied Physiology, Vol. 83,
No. 4, pp. 1383-1388, ISSN 8750-7587
Rdegran, G. & Saltin, B. (1998). Muscle blood flow at onset of dynamic exercise in humans.
American Journal of Physiology Heart and Circulatory Physiology, Vol. 274, No. 1, pp.
H314-H322, ISSN 0002-9513
Robergs, R.A., Icenogle, M.V., Hudson, T.L. & Greene, E.R. (1997). Temporal inhomogeneity
in brachial artery blood flow during forearm exercise. Medicine and Science in Sports
and Exercise, Vol. 29, No. 8, pp. 1021-1027, ISSN 0195-9131
Shoemaker, J.K., Hodge, L. & Hughson, R.L. (1994). Cardiorespiratory kinetics and femoral
artery blood velocity during dynamic knee extension exercise. Journal of Applied
Physiology, Vol. 77, No. 6, pp. 2625-2632, ISSN 8750-7587
Tschakovsky, M.E., Saunders, N.R., Webb, K.A., & ODonnell, D.E. (2006). Muscle blood-
flow dynamics at exercise onset: Do the limbs differ? Medicine and Science in Sports
and Exercise, Vol. 38, No. 10, pp. 1811-1818, ISSN 0195-9131
Walle, L. & Wesche, J. (1988). Time course and magnitude of blood flow changes in the
human quadriceps muscles during and following rhythmic exercise. Journal of
Physiology (London), Vol. 405, pp. 257-273, ISSN 0022-3751
Part 4
Advances in Medical Imaging Techniques

15
Current State of the PET/CT Hybrid
Technique and Clinical Indications
Patricia Carreo-Morn and Mara de las Nieves Gmez Len
Clinical University Hospital of Salamanca and Princesa Hospital of Madrid
Spain
1. Introduction
In the year 2000, Townsend developed a prototype that integrates FDG-PET and CT, thus
creating the possibility of performing both tests sequentially, in order to obtain an image
that is the result of the fusion of hardware of anatomical (CT) and metabolic information
(PET) of the section under study (Townsend & Beyer, 2002). Since then, PET/CT imaging
has been a revolution in the diagnosis of tumours, staging, restaging and detection of local
and distant recurrence and assessment of therapy response.
Applications have been developed in very different areas such as oncology, cardiology
(cardiac viability), neurology (Alzheimer disease and epilepsy before surgery), etc.
2. Techniques
2.1 Hybrid technique
Therefore, the CT, which was the anatomical imaging technique of choice in the staging and
monitoring of the treatment of oncological patients as well as in the planning of
radiotherapy, joined the FDG-PET, which is a more functional technique that allows for an
early detection of the disease and with which the residual lesions after treatment can be
characterized as scar tissue or neoplastic lesions. PET/CT might improve the precision of
the initial staging and the detection of residual diseases and recurrences in these patients.
This would then optimize the treatment schemes for each patient and prevent the
complications of other, more aggressive diagnostic techniques, as well as the toxicity of
unnecessary chemotherapy and/or radiotherapy (Beyer, 2011).
2.2 PET/CT imaging protocol
The patients must not eat anything for 4-6 hours prior to the test. Immediately after the
administration of a 370 MBq dose of intravenous 18FDG, the patient rests for 45-60 minutes.
During this time, each patient receives 1500 ml of oral contrast (Gastrographin 3%), except
for the studies on lung cancer, in which water is administered as an oral contrast. Nowadays
the data are obtained with a hybrid system PET/CT which combines 4-64 slice multi-
detector CT with a PET scanner of 18 bismuth germinate (BGO), lutetium oxyorthosilicate
(LSO), or gadolinium silicate(GSO) crystal detector rings. At first, a CT scan is performed at
low doses and without intravenous iodated contrast.

Medical Imaging

312
The images are taken 45-60 minutes after an injection at rest, starting at the base of the skull,
down to the upper area of the thigh. Initially, in Spain, in 2003 (Gmez-Len et al., 2007),
low dose CT images were obtained at 140 mV, 80 mA, gantry rotation time of 0.5 seconds,
2x5 cm collimation, section thickness of 5 mm and reconstruction interval of 3 mm.
Immediately after that, the PET scan acquired 4-6 contiguous volumes. Finally, the
diagnostic CT could be studied after an injection of intravenous contrast (Iobitridol 300 mg
iodine/ml), with a 50-70 seconds delay for the acquisition of the images (depending on the
pathology under study). The parameters used were the same that those for low dose CT,
except for the intensity of the current, which varied according to an automatic intensity
modulation system that depends on the weight of the patient, with a maximum of 300 mA.
A few minutes after the end of the test, PET images went through attenuation correction
with the CT data and then reconstructed, like the CT images and the combined PET/CT
images (Gmez-Len et al., 2007).

Fig. 1. PET/CT (modified from Townsend, 2002)

Current State of the PET/CT Hybrid Technique and Clinical Indications

313
Nowadays, PET/CT images can be studied without a prior low-dose CT, using the data
obtained in the diagnostic CT with intravenous contrast. There are no differences between
PET attenuation with low-dose CT and diagnostic CT with intravenous contrast (Pinilla et
al., 2010).
2.3 Assessment of the PET image or CT image
The main limitations of these techniques separately are the inability to distinguish, in
many cases, benign lesions from malignant lesions, especially in the lymph nodes, the
inability to establish the response to the treatment and the impossibility to distinguish, in
some cases, between the changes in recurrences after therapy (chemotherapy,
radiotherapy, surgery).
2.4 Advantages of combined PET/CT over PET and CT
Improvement in locating the lesions thanks to a quasi-perfect anatomical and functional
co-register.
Different physiological and pathological uptake in PET.
Functional changes precede anatomical alterations.
Planning of RT.
CT data can be used for the correction of data attenuation in PET (shorter times).
PET/CT shows a higher sensitivity and specificity than each of its components
individually, and probably higher than the combined retrospective reading of both
separated components.
The most relevant effect is the fact that CT provides more specificity to PET, which
allows for a precise anatomical location of PET uptakes and a higher degree of certainty
in the interpretation of the tests (Czernin et al., 2010).
PET, on the other hand, provides some valuable functional information.
Tumour cells present an increased glucose transport rate and an increased rate of
glucose metabolism.
2.5 Implications of CT-based attenuation correction
Artifacts due to respiratory and deglutition movements.
Artifacts due to beam hardening (upper limbs) and obesity.
Artifacts due to the use of contrasts and high-density objects.
Other artifacts due to the patients weight, the patients BMI, the patients lean weight,
the administered dose of FDG, the level of basal glycemia, crystal PET detectors (BGO
vs. LSO), etc.
Asymmetrical uptake of FDG in the muscles, previous record of traumatisms or
inflammation.
2.6 PET/CT: Physiological uptake
18FDG uptake is more intense in the brain, in which metabolism depends on glycolysis,
and in the myocardium, which uses glycol tic metabolism in basal situations.

Medical Imaging

314
18FDG is excreted through urine, and intense activity can be found in the urinary
system, the ureters and the bladder.
It is present in the liver, the spleen, the bone marrow and the renal cortex.
2.6.1 Physiological uptakes

Fig. 2, 3 & 4. Physiological uptake of the brain, thymus and myocardium.

Fig. 5a & b. Physiological uptake of lymph tissue (asterisks) and tongue (asterisks).


315

Fig. 6. Physiological uptake of the kidneys and bladder.

Fig. 7. Uptake of the endometrium (arrow) and bowel loops filled with oral contrast.

Fig. 8. Uptake of the endometrium and bowel loops filled with oral contrast.

Medical Imaging

316
2.6.2 Physiological uptakes after treatment

Fig. 9. Axial CT section with intravenous contrast, axial section of the PET component,
combination of both techniques and coronal section of the entire PET study. Brown fat
(arrows on the different sections).

Fig. 10. Sagittal sections of CT component, PET component and combined image. Coronal
PET section. Axial sections of the pelvis of the CT component, PET component and
combined image. The patient received a bone marrow transplant, with uptake in the entire
axial and peripheral skeleton due to granulocyte colony-stimulating factor.


317
2.7 FDG PET/CT: Limitations
This is a list of tumours with low uptake of FDG: Well-differentiated tumours, hypocellular
tumours, mucine-producing tumours, HCC (hepatocell carcinoma), BAC and intraductal
mucinous papillary tumours.
On the other hand, there are some difficulties in the detection of the pathological collection:
High physiological uptake: brain, tonsils
Physiological elimination: Urothelial carcinoma of the renal pelvis, bladder cancer
2.8 Applications of PET/CT in oncology
PET/CT is used in the staging and re-assessment of specific neoplasias, and it is especially
useful in non-small cell lung cancer, lymphomas, colorectal cancer, sarcomas and melanomas.
PET/CT is particularly useful in patients under suspicion of clinical recurrence with
conventional imaging techniques, as well as in the characterization of residual masses after
chemotherapy and radiotherapy.
We want to highlight an article that was published in Euroradiology in 2011, which
analyzed the increase in the use of PET/CT in Europe (Hilund-Carlsen et al., 2011). The
most notable results dealt with the fact that the diagnosis, staging and treatment may
change in up to 36% of the cases, according to some studies published in the USA: Initial
results of The National Oncologic PET registry.
Danish experience: A 3-year clinical experience in a large new Danish PET/CT centre in
relation to national and European developments. The use of PET/CT in cancer was
registered from early 2006 to 2009, in order to judge the impact on patient management and
to compare it with national and European trends.
Referrals came primarily from the departments of oncology (23.0%), hematology
(21.6%), surgery (12.6%), internal medicine (12.7%) and gynecology (5.5%).
Referral indications were diagnosis (31.3%), staging (22.3%), recurrence detection
(21.2%), response evaluation (17.0%) and other causes (8.2%). Response from nearly
60% of users showed that PET/CT caused a change in diagnosis and/or staging and/or
treatment plan in 36.0% of cases.
The working diagnosis was confirmed in 53.7% of the cases, the staging in 42.7% of the
cases and the treatment plan in 49.9% of the cases, and changes took place in diagnosis
(14.3% of the cases), in staging (22.1%), and in treatment plan (28.3%).
According to the EANM, during the study period, there was a steep increase in the
national use of FDG and in the European use of PET/CT from 166 cases (in 27
countries) to 463 cases (in 46 countries), which represents an increase of 179%, and an
increase of the number of studies of 69%.
3. Indications for the monitored use of FDG: High-level research NLCAHR
contextualized health research synthesis: PET/CT programs
1. Solitary pulmonary nodule
2. Non-small cell lung cancer
3. Recurrent colorectal cancer
4. Staging and re-staging of lymphoma

Medical Imaging

318
5. Recurrent malignant melanoma.
6. Identification of cancer of unknown primary origin
7. Malignant tumours in head and neck
8. Breast cancer.
9. PET-CT and its applications in radiation therapy.
10. The future: Neuroradiology and cardiovascular diseases (Demeter et al., 2009).
3.1 Solitary pulmonary nodule
The role of PET-CT in clinical assessment in indeterminate pulmonary nodules shows
two options: biopsy or resection and monitoring over 2 to 5 years. The use of PET-CT in
the management of pulmonary nodules helps to stratify patients better according to the
risk of malignancy.
Accuracy depends on: Nodule size 1cm and FDG-Avidity SUV 2.5: overlapping
Improvements in the increase of the uptake compared with background activity
S: 83-100% E: 63-90%, NPV: 95% and change the management in 26% of patients.
3.1.1 Limitations of PET/CT

Fig. 11. False positive: Axial section of CT, PET and combined images and coronal section of
the PET study. The pulmonary nodule of the right upper lobe shows an intense uptake of
18FDG in the PET and combined image. The anatomopathological analysis revealed a
tuberculoma in a patient with colon carcinoma.


319
Benign uptakes:
Physiological uptakes. Brown fat
Inflammation: sarcoidosis, vasculitis, etc.
Inflammation after surgery (1-2 months) or after radiation therapy (2-3 months)
Active infections
Benign tumours
False positive: Tuberculosis, granuloma, sarcoidosis, abscess and fungal infection
(aspergillosis, coccidioidomycosis), necrotizing pneumonia, benign tumours: sclerosing
hemangioma, myofibroblastic tumour and leiomyoma.
False negative: Well-differentiated adenocarcinoma, BAC (bronchioalveolar carcinoma),
carcinoid and nodules of less than 1 cm.

Fig. 12. Coronal and sagittal sections of the entire body with CT, PET and combined images.
Axial sections. False positive: Necrotizing pneumonia of middle lobe and lower right lobe
with effusion.

Medical Imaging

320

Fig. 13a & b. Axial sections of CT and PET/CT combined image. Lesion on the lower right
lobe with air bronchogram, without 18FDG uptake. False negative: Bronchioalveolar
carcinoma.
3.2 Non-small-cell lung carcinoma
PET/CT improves the identification and location of the lesions as well as the detection of
the involvement of small nodes and neoplastic lesions with low FDG avidity. Diagnostic
accuracy of PET/CT: T 70-97%; N: 78-93%; TNM: 83-96%. Global staging is significantly
better. It leads to a change of stage in approximately 26% of the patients, and it changes the
management approach in 9-19% of the cases.
Advantages of PET/CT:
Techniques: Better quality than PET study. The PET element is shorter (examination
time decreases by 40%), and it provides a more effective use of PET radiotracers.
For the patient: A single preparation session and study session, and a better clinical
management of complementary examinations
At diagnosis: More sensitivity and specificity than isolated CT and PET.
CT adds specificity: Better location, safer reading.
PET adds sensitivity: it detects infiltrations without morphological alterations.
3.2.1 Limitations of PET/CT
Tumours with a decrease in the 18F-FDG uptake: Well-differentiated tumours,
hypocellular tumours or mucin-producing tumours.
Difficult detection of pathological deposits:
High physiological uptake: brain, tonsils.
Physiological elimination: Urothelial carcinoma of the renal pelvis, bladder cancer


321
3.2.2 T staging
PET/CT is a better choice because it provides an accurate correlation between the extension
of the FDG deposit and the anatomy of the area.
Improvement of the CT component: Focal chest wall infiltration, invasion of the
mediastinum and vascular invasion.
PET Component: It distinguishes between tumours and atelectasis; malignant pleural
effusion (T4) with an accuracy of 92%; metabolic activity with prognostic value.
3.2.3 N staging
PET increases sensitivity and NPV. PET/CT adds specificity and improves accuracy:
Improvement of the exact location of the uptake in patients with atelectasis, mediastinal
deviation and anatomical variants and proper identification of the node stage.
The problems derived from false positives in granulomatous disease and inflammatory
changes caused by a coexistent lung disease leads to a lower PPV. For this reason, a
mediastinoscopy is advised if the tumour can be operated, as well as an histological
confirmation.
PET/CT is a good guide for an invasive biopsy, if the glands show positive results for the
aortopulmonary window, the anterior mediastinum or the posterior subcarinal region. It is
useful for a mediastinotomy, a transbronchial biopsy, ultrasound-guided fine-needle
aspiration or transbronchial ultrasound-guided endoscopy.

Fig. 14. PET/CT: Staging of NSCLC with mediastinal adenopathies, with an important
18FDG uptake. The invasion of the right main pulmonary artery can be seen thanks to the
intravenous contrast. The pleural effusion does not show an 18FDG uptake.

Medical Imaging

322
3.2.4 M staging
PET provides an excellent detection of thoracic metastases, detection of unsuspected distant
metastases 28% and change of treatment in 53% of the cases. PET/CT allows a more
accurate location in adrenal metastases, with diagnostic accuracy in 92 % of the cases. It
reduces the number of unnecessary thoracotomies by 25%.
Hepatic metastases: The specificity and sensitivity of PET/CT is similar to those of
other diagnostic techniques. It offers additional information for unspecific lesions.
Adrenal metastases: Up to 10% of all NSCLC include a metastatic adrenal nodule at
diagnosis. The accuracy of PET/CT is 92%, and its specificity ranges between 80 and
100% of all cases. An undetermined nodule in the CT with negative results in the PET
scan generally means that it will be benign. False negatives have been described in
adrenal metastases for very small lesions or when there is hemorrhage or necrosis.
Bone metastases: Sensitivity goes up to 90% (similar to scintigraphy with Tc99m),
although its specificity is better than scintigraphy with Tc99 (61%). The main
limitation can be seen in blastic bone metastases, in which scintigraphy with Tc99m
shows higher accuracy, or peripheral bone metastases. However, most bone
metastases of NSCLC are central and lithic, which means that PET/CT might replace
scintigraphy with Tc99m.
Limitations of PET/CT in brain metastases are explained by the difficulty of finding
small lesions due to cerebral physiological FDG activity. For this reason, brain NMR is
recommended in these uncertain cases.
3.2.5 Re-staging
Post-therapy changes are known to make CT assessment more difficult. PET/CT improves
the post-therapy accuracy, because 18F-FDG accumulates in the viable tumour cells instead
of being captured by post-therapy necrosis and fibrosis.
Recurrence diagnosis in NSCLC with PET/CT improves sensitivity (77-100%) and
specificity (62-92%), although this last factor can be reduced due to inflammation after
radiation therapy or chemotherapy. For this reason, the diagnosis should be carried out 1-2
months after chemotherapy and 2-3 months after radiation therapy.

`

Fig. 15. NSCLC staging on upper left lobe with metastases on the liver, bones (sternum and
vertebras) and peripheral pancreas (arrow). The patient is a 61-year-old male with NSCLC
(initial stage: N1, according to a recent CT scan), and with several unsuspected metastatic
lesions on the spine, liver and ribs.


323
3.3 Colorectal cancer
At diagnosis, colorectal cancer is found only in 36% of the patients. 39% of them showed
lymph node metastases, and 19% of them presented distant metastases. Up to 15-20% of
the patients presented liver metastases when their primary tumour was surgically
resected. Surgery (of the primary tumour and the liver metastases) is the only known
curative treatment.
Resection of liver metastases is only indicated for patients with 1-4 lesions located on a
single lobe (whenever there is no other evidence of more adenopathies or distant
metastases)
Up to 5% of the patients present synchronous colon carcinomas, and 30% of all patients
present adenomatous polyps. Moreover, another 5% will develop a metachronic
carcinoma of the colon in the future.
PET/CT is very useful in:
Detection of recurrences in the following cases: high CEA with conventional imaging
techniques, non specific or erroneous findings with conventional techniques or
presacral masses after treatment.
Monitoring of treatment with chemotherapy and radiotherapy.
PET/CT is not to be used in: Screening, initial diagnosis and patients with a known
disseminated disease
Recurrence of colorectal cancer takes place in 37-44% of the patients within the first two
years after the resection of the tumour with curative purposes. An early detection of
potentially resectable metastases or local recurrences leads to an increase in survival rates.
Serum CEA levels can be used to monitor the presence of recurrences (S: 59%; E: 84%), but
they cannot locate the place of the recurrence. PET showed more than 90% sensitivity and
more than 95% specificity, compared with CT levels, which were 60-85% sensitivity and 60-
90% specificity. PET modifies the therapeutic management in up to 25-35% of the patients,
and it detects up to 33% more metastases, compared with conventional methods.
Diagnostic accuracy of PET ranges between 90 and 100%, compared with 50-65% for CT.
Between 15 and 20% of patients with local recurrences are candidates for curative resection.
However, long-term survival in these cases is only 35% (due to the presence of hidden
metastases).
Correct differentiation between a viable tumour and a post-therapy fibrosis (due to surgery
and/or radiation therapy) on the pelvis is vital for a proper management of the patients
with suspicion of local recurrence. PET is very useful for a correct classification of these
patients, and it can also find additional unsuspected metastatic sources. FDG accumulation
6 months after radiation therapy or later are a sign of a residual tumour or a recurrence.
Importance in liver metastases: The use of PET in the assessment of patients before a
curative liver resection probably leads to the selection of a subgroup of ideal patients-
candidates who can benefit greatly from this technique. 3-year survival rate improves up to
77% (compared with 40-45% with CT), and 3-year disease-free survival rates also improve
up to 40% (compared with 15-28% with CT).

Medical Imaging

324

Fig. 16. I. Axial section of CT (A). 64-year-old woman with adenocarcinoma, treated with
surgery, chemotherapy and radiation therapy, with clinical suspicion of recurrence. Axial
section of PET (B) and combined image (C). PET study in a coronal section (D). A small mass
can be seen on the surgical site, showing an intense 18FDG uptake. The recurrence was
confirmed with surgery.
3.4 Lymphomas
In most HL and NHL there is an increase in avidity: HL, DLC-NHL, follicular
lymphoma
Staging and re-staging: higher sensitivity and specificity than CT and PET: sensitivity:
91-94%, specificity: 88-100%.
PET/CT detects normal-sized node diseases, characterization of residual masses; it
detects partial or slow responses and less false positives than PET
It identifies small lesions and specific locations: Base of lungs, skin and non-distant
metastasis
Residual mass: Early detection of remains/recurrence, more effective treatment.
Disease exclusion prevents invasive diagnostic procedures and unnecessary treatments
and guided biopsy towards uptake regions in the mass.
Non-Hodgkin lymphomas (NHL) can be roughly classified into 3 groups: low grade,
intermediate grade and high grade. Low-grade NHL represents 40% of all NHL, and they
are usually indolent. Most of the patients are treated with a monochemotherapy regime (or
even with a wait-and-see approach). PET can represent an important role in the assessment
of patients in which the low-grade NHL is suspected to have turned into a high-grade NHL
(this happens in 10-20% of the cases).


325
Intermediate-grade NHL represents 40% of all NHL, and high-grade NHL represent 5-10%
of all cases. Both are treated with chemotherapy and radiation therapy. PET plays several
roles in the management of these patients. It determines the extension of the disease
(staging) and the response to treatment (therefore, chemotherapy treatments can be
modified if they are not being effective).

Fig. 17. Axial section of CT with normal intravenous contrast. Axial section of PET. PET/CT
detected a pathological FDG uptake in a small adenopathy in the splenic hilium. Therefore,
the stage was increased from II to III.
Hodgkins lymphoma (HL): The extension of the disease is, when taken by itself, the most
important prognostic factor for a proper management with regard to global survival and
recurrence-free survival in HL patients. Conventional methods are often incapable of
revealing the real extension of the tumour (for example, between 20% and 30% of all
patients with a HL that is allegedly located over the diaphragm also present
infradiaphragmatic involvement when an exploratory laparotomy is performed).
PET provides an excellent contrast for the lesions and a relatively good anatomical location
thanks to the tomographic nature of the images, greatly improved with the new PET/CT.
It makes it possible to study, in a single exploration, all the affected organs.
The detection of the lesions depends on their biochemical signal instead of on anatomical
criteria. A certain correlation between the FDG uptake degree and the histological degree of
the lymphoma has been observed.

Medical Imaging

326
However, low-degree lymphomas may not accumulate enough FDG to be detected. PET
fulfills two important aspects that are being more and more taken into account in health
services: on the one hand, it reduces the number of invasive diagnostic procedures, and on
the other hand, it is a technique with a good cost-effectiveness ratio (a study which included
PET into the diagnostic algorithm of lymphoma patients showed that expenses were
reduced by 50%). Some studies have shown that PET modifies the management of
lymphoma patients in up to 62% of the cases (Pinilla et al., 2010).

Fig. 18. PET/CT can change the stage of the disease in approximately 10% of all lymphoma
patients. Study of PET/CT in initial staging of a patient with nodular sclerosis Hodgkins
lymphoma with laterocervical, supraclavicular and mediastinal involvement.
3.4.1 Lymph node involvement
PET shows high sensitivity for the detection of lymph nodes and staging the patients.
Published sensitivities vary between 62% and 100%. In spite of this variability, PET seems to
be more sensitive than CT for the staging of lymphoma patients, because it can detect more
lesions inside and outside the nodes, than conventional methods. It alters the stage of
patients, in any of both directions, in up to 44% of the cases, according to some studies.
3.4.2 Extranodal involvement
Extranodal involvement is associated with a worse prognosis of patients. PET can detect up
to 57% more extranodal lesions than CT. Diffuse infiltration in the spleen, bone marrow and


327
liver cannot be detected with CT. Up to 30% of lymphoma patients with splenomegaly do
not present tumour infiltration of the spleen and a significant number of patients presents
tumour involvement of the spleen without splenomegaly. FDG uptake by the bone marrow
represents a tumour infiltration, although an increased uptake of the bone marrow can also
be expected if the patient has undergone chemotherapy or has received colony-stimulating
factors recently (post-therapy medullary hyperplasia).
Bone marrow involvement (stage IV) means a worse prognosis in lymphoma patients, and it
is more frequent in NHL patients. In order to be detected by CT, the bone involvement must
be focused and accompanied by bone destruction. Unfortunately, a diffuse infiltration of the
bone marrow does not show bone destruction, and it is usually asymptomatic. PET can
detect both the focal involvement and the diffuse infiltration of the bone marrow. According
to several studies, it is more sensitive than bone scintigraphy with Tc99m.
3.4.3 Monitoring of response to treatment
Conventional methods (CT, NMR) only show a reduction of the size of the lesions, and they
are not very reliable as predictors of the clinical success of the lymphoma treatment. Some
studies with CT showed that less than 50% with positive findings in CT developed a
recurrence of the disease in the long term. Unlike conventional techniques, PET determines the
metabolic activity of a lesion, which reflects the mass of viable cells inside a tumour. After the
beginning of chemotherapy or radiation therapy, there is a series of metabolic changes in the
tumour before any changes on the size of the mass. Several recent studies suggest that, after
successful chemotherapy, the metabolic activity of the tumour rapidly decreases (SUV
decrease by 75-90% 7 days after the beginning of treatment). An intense and persistent FDG
uptake after the end of chemotherapy cycles shows an absence of therapeutic response.
Patients who show persistently positive PET studies after the first chemotherapy cycle present
a much higher risk of recurrence (a recurrence rate of 90%, according to one study) than
patients with a positive response (decrease in FDG uptake). Likewise, some studies observed
that 85% of all patients with a negative PET study after the first chemotherapy cycle remain in
remission (minimum monitoring time of 18 months). A PET study after the first chemotherapy
cycle also has been proven to be a better predictor for response to treatment than a PET study
at the end of the treatment (diagnostic accuracy of 87% versus 70%).
An early assessment of response to treatment would not only avoid the toxicity of an
ineffective treatment, but also reduce the costs of that therapy. In these cases, patients
could benefit from second-line therapies. A variation of SUV over 25% must be considered
as a real finding.
Differentiation of fibrotic tissue versus viable in a post-therapy residual mass: The presence of
a residual mass after the end of lymphoma therapy is a challenge for the clinical diagnosis,
because it can represent either a viable tumour or a residual diagnosis. Conventional
diagnostic techniques do not offer reliable enough signs that can tell the difference between a
viable tumour and a fibrosis, but with PET, FDG accumulates in the viable tumour, whereas
fibrosis do not reveal any accumulation. In patients with mediastinal lymphoma, up to 64% of
them show some anatomical alteration after the end of the treatment. However, only 18% of
them will show a recurrence. FDG uptake in a residual mass is associated to a higher rate of
recurrence and worse global survival rates, compared with patients without FDG uptake. The
presence of persistent metabolic activity in a residual mass may motivate a change in the

Medical Imaging

328
treatment or the addition of new chemotherapeutic agents. In spite of the fact that 25% of all
patients with negative PET results and a residual mass will have a recurrence, it will take place
in a different location from the residual mass in 80% of the cases (Cronin, 2010).

Fig. 19a & b. 45-year old patient with Diffuse large B-cell NHL with a mass that shows an
intense FDG avidity at diagnosis (a). The mass persists after the end of the treatment,
without FDG uptake. The recurrence was confirmed with monitoring, without additional
treatment (b).
3.5 Melanoma
It is one of the tumours with the highest FDG avidity, and it has the potential to metastasize
in any part of the body. Therefore, it benefits from a PET/CT assessment, which shows a
great sensitivity for the detection of metastases, with the exception of the brain.
PET is useful in the detection of adenopathies and distant metastases (particularly with
melanomas thicker than 1.5 mm Clark level 4). The sensitivity and specificity of PET in
the detection of distant metastases is 80% and 87%, respectively (higher than in
conventional methods).
PETs sensitivity for the detection of adenopathies in patients with melanoma depends on
the size of the metastases (sensitivity is almost 100% for adenopathies >1 cm, and only 23%
for adenopathies <5 mm). All data accumulated until now suggest that the diagnostic
accuracy of PET is higher for systemic staging than for local staging of regional lymph nodes
(especially in Stage I or II patients). Therefore, PET cannot replace sentinel node biopsy in
the clinical management of the patients.
3.6 Cancer of unknown primary origin
Guided biopsy identifies primary tumours in more than 33% of the cases (Wartski et al
2006) and diagnosis of distance metastasis
Controversial cost-effectiveness analysis in first test for epidermoid carcinoma
Useful in cervical metastases of adenocarcinoma.


329

Fig. 20. Patient with melanoma on the leg and inguinal and iliac node involvement.
3.7 Head and neck tumours
They are useful to distinguish, in some cases, between the changes in recurrences after
therapy (chemotherapy, radiotherapy, surgery).
3.7.1 Applications of PET/CT in the assessment of head and neck tumours
Unknown primary tumour: detection of the primary lesion in up to 33% of the cases.
Staging.
Re-staging: post-therapy changes versus recurrence.
Thyroid cancer: especially when the I-131 scintigraphy is negative.
Metachronic/synchronic tumours: respiratory and digestive tract (up to 30% of the
patients).
3.7.2 Possible interpretation errors
Asymmetrical uptake of FDG in the muscles.
Submaxillary glands
Treatment with surgery and radiation therapy.
Uptake in supraclavicular area fat.
Hashimotos thyroiditis.

Medical Imaging

330
3.7.3 Esophageal cancer
USAs Medicare currently includes two indications for PET in esophageal cancer:
Preoperative staging and post-therapy re-staging.
Monitoring of the treatment (when there is evidence of a potential recurrence).
Primary tumour: PET is not useful in the assessment of primary esophageal cancer or in
the detection of a local invasion of the tumour. A PET/CT combination improves staging.
Lymph node metastases: The presence of adenopathies on the neck, the supraclavicular
areas or the celiac trunk is a sign of M1 metastasis (which means that its detection
represents a great impact on the management of the patient).
Distant metastases: The main advantage of PET with regard to conventional imaging
techniques is its capability to detect distant metastasis (diagnostic accuracy of 84%
versus 63% for CT). The presence of distant metastases has a great impact on the
management of the patient because they are not resectable. PET detects metastatic
diseases that are not identified by conventional studies in up to 20% of the patients.
Recurrent esophageal cancer: The sensitivity of PET for the detection of distant
metastases is 95% (specificity of CT is 79%). Moreover, PET provides additional
diagnostic information in 27% of the patients with suspicion of tumour recurrence.
3.7.4 Thyroid cancer
Diagnosis: PET is not indicated in the diagnosis of thyroid cancer (not all carcinomas
attract FDG, and some benign lesions can accumulate FDG).
Staging: Although there are no data in this regard, PET probably does not play an
important role in the initial staging of patients with thyroid cancer.
Patients with suspicion of recurrent cancer and a negative I-131 scintigraphy: The
published sensitivity of PET for the detection of recurrent thyroid cancer in patients
with negative I-131 scintigraphies ranges between 60% and 94% (specificity between 42-
95%). One advantage of PET is the fact that it can detect tumour sites in nodes smaller
than 1 cm. However, PET cannot detect small lung metastases.
Prognosis: 3-year survival rates are significantly reduced in patients with postive PET
results (survival of 60%), with regard to patients with a negative result (survival of
98%), which means that PET has an important prognostic value.

Fig. 21a & b. Recurrent thyroid cancer. The patient is a 73-year-old woman with a record of
papillary thyroid cancer. She showed a recent increase of serum thyroglobulin with negative
results for I-131 scintigraphy. PET/CT: Tumour remains on the surgical site and right
laterocervical adenopathy.


331
3.7.5 Head and neck tumours
Most head and neck tumours are epidermoid carcinomas. Up to 60% of the patients
present an advanced stage of the lesion at diagnosis. The involvement of lymph nodes is
the most important prognostic factor with regard to survival. Conventional studies with
CT and/or NMR have certain limitations in the assessment of lymph node involvement,
because they are mainly based on the size of the nodes (> 1 cm) in order to classify them
as pathological nodes. However, up to 40% of all lymph node metastases take place in
nodes smaller than 1 cm. PET is particularly useful in the assessment of patients with a
clinical stage N0. In this clinical context, between 16% and 60% of all patients reveal
hidden lymph node metastases after a PET study.
Proper preoperative staging (TNM) is essential when planning the type of lymph node
dissection and when assessing the need of postoperative radiation therapy or
chemotherapy.
A relatively common presentation of head and neck tumours is the appearance of
cervical adenopathies. A careful ENT examination, together with CT/NMR can identify
the primary tumour in most cases. However, in up to 32-40% of the patients, the
primary lesion will not be identified. PET, particularly combined with CT, is especially
useful in these cases.
Assessment of response to treatment: PET (PET/CT) can be used to monitor the
response to treatment in head and neck tumours. Tumours that respond to treatment
show a reduced metabolic activity, and those that present a persistent uptake one
month after radiation therapy usually contain viable or residual tumour cells. In more
than 80% of the cases, PET can establish the difference between a residual tumour and
post-radiation therapy fibrosis.

Fig. 22a, b & c. Head and neck tumour: unknown primary tumour. The patient is a 70-year-
old male with a recent biopsy of a right cervical mass that was positive for epidermoid
carcinoma (unknown primary mucosal lesion). PET/CT shows a favorable response after
treatment with surgery and radiation therapy.

Medical Imaging

332
The quantification of FDG uptake by the tumour (through SUV) correlates with the
biological behavior of the tumour, and it has been proved that an intense FDG uptake
(SUV>5.5) identifies a group of patients who can benefit from a more aggressive
treatment.

Fig. 23a & b. Re-staging of head and neck tumour. The patient is a 72-year-old male with
adenopathy in an adenocarcinoma of the left nasal cavity with extension into the orbitary
region that was surgically operated. 6 weeks later, the tests reveal a FDG-uptake area on the
anterior region of the left zygomatic arch, compatible with persistence of viable tumour.
3.8 Breast cancer
Assessment of systemic metastases, identification of recurrence locations and possible role
in the radiotherapy planning.
CT Component: extended towards chest wall and skin.
PET provides a more accurate detection of metastases in the internal mammary gland.
PET studies can have a very significant impact on the management of patients, especially in
those with suspicion of advanced disease. PET can modify the clinical stage in up to 36% of the
cases (it increases the clinical stage in 28% of the cases and reduces it in 8% of the cases). In
20% of the cases, unsuspected metastases are discovered, and the clinical management of
patients is modified in up to 58% of the cases, once that the PET results are taken into account.
3.8.1 Local disease
Breast cancer presents a significantly variable FDG uptake. Generally speaking, a higher
proliferation rate of the tumour and a higher degree of undifferentiation mean a higher rate
of metabolic activity of the tumour and higher FDG avidity. Invasive ductal carcinoma
shows a FDG uptake significantly higher than invasive lobular carcinoma. Focal node


333
lesions also present high levels of FDG uptake, compared with infiltrative/diffuse lesions.
PET/CT has an advantage in the fact that it is not altered by the density of the mammary
tissue, and the quality of the image is not altered in case of previous surgery (radiation
therapy or mammary prosthesis). Its sensitivity ranges between 64% and 96%, and its
specificity is around 75-100%. The main limitation of PET is the fact that it cannot detect
lesions smaller than 1 cm, due to the partial volume effect and the physiological
accumulation of FDG in the mammary tissue.
3.8.2 Lymph node metastases
The sensitivity that has been published in the literature for PET detection of locoregional
adenoapthies ranges between 50% and 100%, and its specificity is around 86-92%. If nodes
of less than 1 cm and micrometastases are taken into account, the sensitivity of PET studies
significantly decreases. For these reasons, PET cannot replace axillary node dissection. PET
is particularly useful in the detection of adenopathies in the mediastinum and the internal
mammary chain (PET sensitivity: 85%; CT sensitivity: 54%). The identification of these
adenopathies has a very significant impact on the management of the patients (either
increasing the radiation field or applying a more aggressive chemotherapy regime) and in
their prognosis (Escalona et al., 2010).
3.8.3 Metastatic disease
PET is more accurate in the detection of unsuspected distant metastases, both at initial
diagnosis and during the monitoring stage.
3.8.4 Monitoring of response to treatment
The assessment of response to treatment can be established with PET before any other
diagnostic method (before there is a noticeable or measurable reduction in the size of the
tumour). A PET study after a single chemotherapy cycle can predict the response (or lack
of response) of the tumour to the treatment, with a sensitivity of 90-100% and a specificity
of 74-85%).

Fig. 24 a, b & c. Patient with breast carcinoma, recurrence on the right axilla.

Medical Imaging

334
3.8.5 Assessment of tumour recurrence
Previous surgery and/or radiation therapy can produce fibrosis/scar tissue that will
complicate the assessment of potential locoregional recurrences (particularly on the axillary
region), although potential distant metastatic areas can be detected.
3.8.6 Other applications of PET in breast cancer patients
The presence or absence of estrogen receptors has a great influence on the choice of a
systemic treatment. Between 30% and 40% of the patients with advanced breast cancer and
positive result for estrogen receptors (ER) will respond to treatment with antiestrogens
(tamoxifen). Another important percentage of patients will have a stable condition for a
clinically significant period of time. The presence of ER can be assessed with the use of an
estrogen tracer (FES). If a tumour shows SUV>1 in a PET study with FES, it is considered to
be a positive result for ER. When the tamoxifen treatment has started, ER are blocked, which
means that any tumour that responds to antiestrogen therapy will show a fast decrease of
SUV in PET studies with FES.
3.9 PET/CT and its applications in radiation therapy
Ideally speaking, the radiation therapy should administer a dose of ablative radiation to the
tumor, leaving the peritumoral healthy tissue intact. With 3D conformational radiation
therapy, the doses for the tumor and the other tissues can be calculated more accurately, the
therapeutic effect increases, and toxic complications are reduced. In order to apply this
therapy, the tumor needs to be accurately located. CT provides anatomical information, as
well as the dosimetric bases for the calculation of radiation absorption. However, it also has

Fig. 25a & b. Intensity-modulated radiation therapy in a patient with cervical cancer (pre-
and post-treatment)


335
important limitations, such as the delimitation of tumor margins, unnecessarily large
radiation areas or infiltrated areas that remain untreated. PET/TC integrates structural
and molecular information. Therefore, it provides a more accurate estimation of the tumor
size and it identifies better areas with tumor viability. The use of this new
anatomobiological outline in the planning of radiation therapy represents a clinically
significant change with regard to the volume of radiation therapy that the patients
receive. It can change the treatment approach, avoid unnecessary radical radiation
therapy and adjust the volume of treatment, with the possibility of increasing the doses in
specific locations.
PET/CT before treatment: Cervical cancer with inguinal and paraaortic adenopathies.
Definition of target volume.
Planning of radiation therapy. The axial section of the radiation field is extended 14 mm
in order to include paraaortic adenopathies, and additional radiation is applied to the
pelvic adenopathy.
Intensity modulated radiation therapy is administered in order to achieve higher
effectiveness in the treatment with less side-effects.
The response to treatment is assessed after the end of the therapy.
3.10 The future: Neuroradiology and cardiac viability
3.10.1 Drug-resistant epilepsy
MRI and PET co-registration in pediatric epilepsy FDG-PET represents a useful tool for
presurgical evaluation of epilepsy, particularly when MRI is nonlesional. Previous studies
have shown FDG-PET to have 63100% sensitivity in lateralizing temporal lobe epilepsy
(TLE) and to provide complementary information to MRI. For extratemporal lobe epilepsy
(ETLE), studies have shown FDG-PET to have slightly lower sensitivity at 3683%. In
addition to increasing detection postsurgical prognostic information independent of
information provided by MRI, FDG-PET might thus provide complementary as well as
supplementary functional information in regard to the etiology of seizure activity. Given the
parallel roles of MRI and FDG-PET in presurgical evaluation for epilepsy patients, co-
registration of MRI and FDG-PET might enhance presurgical management of intractable
epilepsy. This has not been unequivocally established, but it is already recommended that
FDG-PET images be interpreted in light of all structural imaging information. In the recent
UCLA cohort, FDG-PET and MRI co-registration demonstrated favorable postsurgical
outcomes (Engel class III) in 80% of the patients with intractable epilepsy with the
application of co-registered imaging to maximally resect the functional abnormal area. This
technique uses anatomic imaging to help define the limits of resection, despite previous MRI
findings that had been considered to be nonlesional.
3.10.2 PET and brain development in pediatrics
Functional development of the pediatric brain has been evaluated by FDG-PET. Chugani et
al. demonstrated that the metabolic pattern of a developing brain follows the order of
anatomical, evolutional and behavioral development. Increased glucose metabolism is
shown in the visual, sensorimotor cortex and the cerebellum, and this is correlated with early

Medical Imaging

336
visuospatial and sensorimotor function and primitive reflexes. Hypermetabolism in the basal
ganglia is known to be associated with developing movement and sensorimotor function.
The quantitative analysis of brain FDG-PET has demonstrated that the degree of glucose
metabolism of infants is significantly lower than that of adults. The degree of metabolic
activity of neonates is about 30% that of adults and it continues to increase with age. It is
hypothesized that increased metabolism is associated with increased metabolic demands
from neuronal plasticity development. By a childs third year, the metabolic level exceeds
that of adults, and it reaches its plateau between ages 4 and 9 with a value 1.3 times higher
than that of normal adults. After this period, the value decreases to adult level by the end of
the second decade.
Future applications include combined PET-MRI imaging and neuroreceptor imaging.
3.10.3 Dementia
Conventional anatomic imaging (e.g., MRI and CT) play a minimal role in the early
detection of dementia, especially for Alzheimers Dementia AD. There is evolving literature
that supports that FDG PET may be more sensitive and specific than conventional nuclear
medicine imaging (i.e., ECD and HMPAO imaging) in the diagnosis of dementia, especially
in early detection. A more recent multicentre study (Mosconi et al., 2008) enrolled 548
patients (110 normal, 114 with mild cognitive impairment, 199 with AD, and 125 with other
forms of dementia). Based on clinical endpoints they found that FDG PET correctly
classified 94% of normal variants, 95% of AD, and between 92% and 94% of other types of
dementia. There was significant heterogeneity in FDG patterns for patients with mild
cognitive impairment. They developed standardized automated methods to analyze FDG
brain scans and thought this automated consistent approach resulted in more homogeneous
results than previously reported in the literature.
3.10.4 Cardiac viability
A stress PET examination can reliably demonstrate myocardial blood flow using 82Rb or
13Nammonia. PET radiopharmaceuticals can be used to assess both cardiac perfusion (i.e.,
13N-ammonia, 15O-water, 82Rb) and heart muscle viability (FDG) (Machac et al., 2006). This
section will only address evidence related to FDG viability assessment. Cardiac imaging
tests can be roughly stratified into those which assess coronary blood flow/cardiac muscle
perfusion, systolic function (i.e., how well the heart is pumping), heart muscle viability (i.e.,
living, but potentially at risk, heart muscle versus dead/scarred tissue) and other less
common studies which assess very specific metabolic processes (e.g., fatty acid metabolism)
or neuromuscular function (e.g., MIBG studies). PET can be used to assess all of these areas
the CT component, can also assess coronary artery anatomy (e.g., CT angiogram) and
coronary calcium scoring. These parameters can also be assessed through conventional CT
or with SPECT/CT (Hesse et al., 2005).
4. Conclusion
PET/CT is a multidisciplinary technique which involves nuclear, radiology, radio physics,
radio pharmacology and oncology and combines the advantages of the functional
information by PET and the special and contrast resolution of CT. This improves the


337
diagnosis, staging and follow-up of some types of cancer and other pathologies, and new
indications could appear in the future.
5. References
Beyer, T.; Townsend, D.W.; Czernin, J. & Freudenberg, L.S. (2011) The future of hybrid
imaging part 2: PET/CT Insights Imaging REVIEW. Insights into imaging, Vol. 2,
No. 3, pp. 225-234.
Cronin, C.G.; Swords, R.; Truong, M.T.; Visswanathan, C.; Rohren, E.; Giles, F.J.; ODwyer,
M. & Bruzzi, J.F. (2010) Clinical utility of PET/CT in lymphoma. American Journal of
Roentgenology, Vol. 194, No. 1, pp. W91-W103, ISSN 0361-803X.
Czernin, J.; Benz, M.R. & Allen-Auerbach, M.S. (2010) PET/CT imaging. The incremental
value of assessing the glucose metabolic phenotype and the structure of cancers in
a single examination. European journal of radiology, Vol. 73, No. 3, pp. 470-480, ISSN
0720-048X.
Demeter, S.; Bornstein, S.; Butler, J.; Cramer, B.; Hollett, P. & Jones, L. (2009). The
Development of a PET/CT Program in Newfoundland and Labrador, Newfoundland
and Labrador Centre for Applied Health Research, Memorial University, St.
Johns, NL.
Escalona, S.; Blasco, J.A.; Reza, M.M.; Andradas, E. & Gmez, N. (2010). A systematic review
of FDG-PET in breast cancer. Medical Oncology, Vol. 27, No. 1, pp. 114-129, ISSN
1357-0560.
Gmez-Len, N.; Pinilla, I.; Rodrguez-Vigil Junco, B.; Coya, J.; Herndez, D.; Andradas,
E.; Reza, M. & Madero, R. (2007). Uso del sistema hbrido PET/TC: experiencia
inicial del Hospital Universitario La Paz. Radiologa, No. 49, pp. 29-36, ISSN 0033-
8338.
Hesse, B.; Tgil, K.; Cuocolo, A.; Anagnostopoulos, C.; Bardies, M.; Bax, J. et al. (2005)
EANM/ESC procedural guidelines for myocardial perfusion imaging in nuclear
cardiology. Eur J Nucl Med Mol Imaging, No. 32, pp. 855-897, ISSN 1619-7070.
Hilund-Carlsen, P.F.; Gerke, O.; Vilstrup, M.H.; Lerberg Nielsen, A.; Thomassen, A.; Hess,
S.; Hilund-Carlsen, M.; Vach, W. & Petersen, H. (2011) PET/CT without capacity
limitations: a Danish experience from a European perspective. European Radiology,
Vol. 21, No. 6, pp. 1277-1285, ISSN 0938-7994.
Machac, J.; Bacharach, S.L.; Bateman, T.M.; Bax, J.J.; Beanlands, R.; Bengel, F. et al. (2006)
Positron emission Tomography myocardial perfusion and glucose metabolic
imaging. Journal of Nuclear Cardiology, No. 13, pp. 121-151, ISSN 1071-3581.
Pawaroo, D.; Cummings, N.M.; Musonda, P.; Rintoul, R.C.; Rassl, D. & Beadsmoore, C.
(2011) Non-small cell lung carcinoma: accuracy of PET/CT in determining the size
of T1 and T2 primary tumours. American Journal of Roentgenology, Vol. 196, No. 5,
pp. 1176-1181, ISSN 0361-803X.
Pinilla, I.; Gmez-Len, N.; Del Campo-Del Val, L.; Hernandez-Maraver, D.; Rodrguez-
Givil, B.; Jover-Daz, R. & J. Coya, J. (2010) Diagnostic value of CT, PET and
combined PET/CT performed with low-dose unenhanced CT and full-dose
enhanced CT in the initial staging of lymphoma. Q J Nucl Med Mol Imaging, EPUB
ahead of print, ISSN 1827-1936.

Medical Imaging

338
Townsend D. W, Beyer T. (2002) A combined PET/CT scanners: a hardware approach to
image fusion. British J Radiol; No. 75, pp. 24-30.
16
Advances in Medical Imaging
Applied to Bone Metastases
ngel Gonzlez-Sistal
1
, Alicia Baltasar Snchez
1
,
Michel Herranz Carnero
2
and lvaro Ruibal Morell
2

1
Medical Imaging Research Laboratory, Dpt. Physiological Sciences II
Faculty of Medicine, University of Barcelona
2
Nuclear Medicine Service, Complexo Hosp. Universitario Santiago de Compostela
Molecular Imaging Group, IDIS
Spain
1. Introduction
Bone metastases are the result of a primary cancer invasion which spreads into the bone
marrow through the lymphogenous or hematogenous pathways. Bone metastases are a
common complication of cancer.The primary cancers that most frequently metastasize to
bone are breast and prostate cancer (65 - 75 %) amongst many others (thyroid 42 %, lung 36
% or kidney 35 %) (Suva et al., 2011). Although the exact incidence of bone metastases is
unknown given its dependence on the type of primary cancer, it is estimated that 350,000
people die of bone metastases annually in the United States.
1.1 Bone structure and microenvironment
Bone is the third most common site of hematogenous tumor metastases after liver and
lungs. The imbalance in bone turnover results in a favorable environment for the growth of
metastatic tumors, a process known as the vicious cycle of bone metastases (Fili et al., 2009).
Bone consists of cortical, trabecular and marrow components. Cortical bone, is compact
and has canals containing vessels. A layer of compact bone surrounds trabecular bone,
which contains the bone marrow. Most of the red marrow (hematopoietic) is located in
axial bones (spine, ribs, pelvis, proximal femora), whereas fat marrow is found in
appendicular bones (long bones). Bone tissue contain the cells types: osteoblasts,
osteoclasts, osteocytes and bone-lining cells.
Breast and prostate carcinomas have a great avidity for bone because the molecular
interactions between these cancer cells and host cells favor the establishment of osseous
lesions. Thyroid, lung, kidney, gastric, colon and skin cancers also metastasize usually to
bone but, to a lesser degree, because these cell types do not possess the properties needed
for invasion and residence in the bone microenvironment (Weilbaecher et al., 2011).
A number of factors contribute to the high incidence of bone metastases (Eriksen, 2010):
high blood flow in the red marrow; adhesive molecules on tumor cells that bind them to
marrow stromal cells and bone matrix (tumor cells increase the production of angiogenic

Medical Imaging

340
and bone-resorbing factors that further enhance tumor growth in bone); growth factors
(transforming growth factor b (TGF b), insulin-like growth factors I and II (IL-I and IL-II),
fibroblast growth factors (FBG), platelet-derived growth factors and bone morphogenetic
proteins (BMPs)) that are released and activated during bone resorption, providing fertile
ground in which tumor cells can grow (Suva et al., 2009).
The adult skeleton continually turns over and remodels itself. There is a balanced
remodeling sequence: osteoclasts resorb bone on trabecular surfaces and then osteoblasts
form bone at the same site (Hadjidakis et al., 2006).
1.2 Types and localization of bone metastases
Most tumor implants occur through the hematogenous pathway. The preference for axial
involvement is also due to the greater vascularity of the red marrow found in the axial
skeleton as opposed to the yellow marrow (high-fat) found in the appendicular bone. Bone
metastases are characterized as lytic(bone destructive), blastic / sclerotic (bone forming)
or mixed according to the radiographic and/or pathologic appearance of the lesion. This
classification represents the dysregulation of the normal bone remodeling process mediated
by osteoblasts, osteoclasts and tumor cells.
Patients with breast cancer have predominantly osteolytic lesions (Trinkaus et al., 2009),
although 15 to 20 percent of them have osteoblastic lesions. Bone metastases from kidney,
lung or thyroid cancers more often are osteolytic. In addition, secondary formation of bone
occurs in response to bone destruction. Only in multiple myeloma do purely lytic bone
lesions develop. In contrast, the lesions in prostate cancer are predominantly osteoblastic.
(Logothetis & Lin, 2005; Ye et al., 2007).
1.3 Clinical features
Bone metastases represent a major cause of morbidity and mortality, once tumors metastasize
to bone they are usually incurable (Coleman et al., 2011). Bone metastases are rarely silent.
They can give rise to a number of life-threatening complications (Coleman et al., 2006).
Osteolytic metastases can cause severe pain, pathologic fractures, decreased mobility,
hypercalcemia, anemia, spinal cord compression and other nerve-compression syndromes.
Patients with osteoblastic metastases have bone pain and pathologic fractures because of the
poor quality of the bone produced by the osteoblasts. Patients with fewer metastases or
solitary lesions appear to have a better prognosis than those with multiple metastatic deposits.
1.4 Biomarkers of bone turnover
Biochemical markers of bone turnover have been used to assess the response to therapy or
for the detection of bone metastases. Levels of bone-specific alkaline phosphatase,
osteocalcin and type I procollagen C-propeptide are serum markers of osteoblast activity.
Whereas, serum levels of C-terminal telopeptide of type I collagen or tartrate-resistant acid
phosphatase and urinay levels of type I collagen cross-linked N-telopeptides are indicators
of osteoclast activity (Roodman, 2004).
There are limitations to the clinical utility of many of these bone markers. Recently, some
new markers have been described as potentially important role: CXXL16/CXCR6 (Ha et al.,
2011), OPG/RANKL (Mercatali et al., 2011) or CCN3 (Ouellet et al., 2011).

Advances in Medical Imaging Applied to Bone Metastases

341
2. Pre-clinical cancer research optical imaging modalities
Optical imaging is based on the detection of photons emitted from living cells, tissues or
animals. It can be divided into bioluminescence imaging (BLI) and fluorescence imaging
(FLI). The role of molecular imaging in pre-clinical research is evolving. In small animal
models optical imaging technologies are used to visualize normal as well as aberrant
cellular processes at a molecular-genetic or cellular level of function (Chaudhari et al., 2005;
Kwon et al., 2010; Snoeks et al., 2011).
The mechanism for creating luminescent light is that luciferase, which acts as a catalyst in
the presence of oxygen and ATP, converts chemical luciferin into oxyluciferin, releasing
light in the process. The substrate D-luciferin is injected, distributes rapidly throughout the
body of the animal and is taken up by the cells. The emitted light is detected by a cooled
charged coupled device camera (CCD) and has a wavelength from 500 to 620 nm which is
sufficient to penetrate small animal tissue (Ntziachristos et al., 2005).
As regards FLI, fluorescence occurs when the excited state is caused by external stimulation
by light (Leblond et al., 2010). As for BLI, luminescence is caused by a chemical reaction
(either a natural, biological one- bioluminescence, or a purely chemistry based once
chemiluminescence) (Kim et al., 2010). Despite its distinctive features each modality
revealed differences in sensitivity, signal-to-noise-ratio (SNR) and background emission
from tissues. Autofluorescence of tissue reduces the signal-to-noise ratio, so in fluorescence
imaging the SNR is expected to be greater than in bioluminescence imaging.
In vivo expression of reporter genes encoding bioluminescent or fluorescent proteins can be
detected externally by sensitive detection systems. BLI could easily image in vivo a bone
metastatic lesion that in the end-point histological analysis results in a tumor of 0.5 mm
3

volume, corresponding to 1.7x10
4
cells.
Optical imaging only provides semi-quantitative data because of tissue-dependent signal
attenuation and poor positional information due to photon scattering. However, new
advances have made it possible to extend BLI and FLI to 3D imaging by optical
tomography, ensuring better quantification of photon emission. As a result of the
development of fluorescence molecular tomography (FMT) and other 3D fluorescence and
bioluminescence data capturing methods, it is now possible to acquire 3D optical data and
use different modalities (radiography, CT, PET, SPECT or MRI) to ensure spatial resolution
and anatomical detail (Nahrendorf et al., 2010).
The main advantage of optical imaging is by a noninvasive study its ability to visualize
biological processes, such as angiogenesis, inflammation or matrix degradation in the
development and growth of bone metastases (Snoeks et al., 2011).
3. Diagnosis imaging modalities: Algorithm choice
Imaging plays a major role given that early identification of skeletal metastases could lead to
changes in patient management and quality of life. Imaging modalities are based on either
direct anatomic visualization of the bone or tumor or indirect measurements of bone tumor
metabolism (figure 1). Clinical evaluation demands multimodal diagnostic imaging owing
to the limitations of the diagnostic techniques. Four main modalities are currently used in
clinical practice: radiography, computed tomography (CT), scintigraphy and magnetic

Medical Imaging

342
resonance imaging (MRI). At present, positron emission tomography (PET) and single
photon emission computed tomography (SPECT) have a potential for evaluation (Rybak et
al., 2001; Costelloe et al., 2009).
These techniques differ in performance in terms of sensitivity and specificity, but none of
the modalities alone seem to be able to yield a reliable diagnostic outcome. There is
currently no consensus about the best modality for diagnosing the bone metastases and for
assessing their response to treatment. In clinical practice, most oncologists do not even use
the same criteria, which results in disparate assessments of bone metastases (Gonzlez-
Sistal, 2007; Welch & Black, 2010).
Although bone metastases can be treated, their response to treatment is considered
unmeasurable, which excludes patients with cancer and bone metastatic disease from
participating in clinical trials of new treatments (Hamaoka et al, 2010).
Positive
99m
Tc-MDP Scintigraphy
Radiography
Normal
Bone metastases Suspicious lesion Benign process
CT or MRI
Bone metastases Benign process Normal
STOP
STOP
2,3, a, b
1, c
2,a,b,c
3, a,b,c
SPECT
FDG-PET
18F-FLUORIDE-PET
4, c
1, c
1,a,b

Fig. 1. Protocol for detection of bone metastases. Each imaging modalities visualize different
aspects of bone tissue or tumor. (Visualization: 1 bone metabolism, 2 cortical/trabecular bone,
3 bone marrow / tumor, 4 tumor metabolism; Extension: a local, b regional, c whole body)
The algorithm of figure 1 shows a classical protocol for detection of bone metastases. The
proposed algorithm is based in:
Scintigraphy remains the technique of choice in asymptomatic patients in whom
skeletal metastases are suspected (whole-body screening). This technique, albeit very
sensitive, is poorly specific, and thus a bone scan finding is double-checked with an
additional examination (radiography, CT or MRI).
Radiography is used to evaluate symptomatic sites (bone pain) or confirm findings of
other imaging modalities


343
CT and MRI can depict anatomic changes in more detail. CT is preferable for assessing
axial bone metastases, regardless of whether the main tumor involves the bone marrow
or cortex. MRI, on the other hand, is better for detecting bone marrow disease or spinal
cord compression.
If MRI and CT cannot detect the disease and clinical suspicion of bone metastases
remains, a PET/CT is made
SPECT is not typically used for the initial detection of metastatic disease. SPECT-CT is
useful for the assessment of lesions that are not determined in scintigraphy
4. Gamma-ray techniques
4.1
99m
Tc-MDP bone scintigraphy
Bone scintigraphy is a nuclear medicine tomographic imaging technique using gamma rays.
Technetium-99m bound to methylene diphosphonate (
99m
Tc-MDP) is the most frequent
radionuclide used in scintigraphy. The patient is injected this radiopharmaceutical and then
he is scanned with a gamma camera (a device sensitive to the radiation emitted by the
patient) which it produces a 2D picture.
99m
Tc-MDP bone scintigraphy is the method used to screen the whole body for bone
metastases. It detects increases in osteoblastic activity and the flow of the blood level. It is a
marker of bone turn-over or bone metabolism (Moore et al., 2007).
Scintigraphy is indicated for the following procedures: screening or staging in
asymptomatic patients, evaluation of persistent bone pain with a negative radiography,
determination of the extent of bone metastases in patients with positive radiography,
differentiation of metastatic from traumatic fractures and determination of the therapeutic
response to bone metastases (Hamaoka et al., 2004).
A classic pattern is the presence of randomly distributed focal lesions throughout the
skeleton (figure 2) which appear as areas of increased tracer uptake (hot spots). Other
typical patterns in scintigraphy are superscan (diffuse metastatic disease), cold lesions (due
to complete absence of reactive bone, they are associated with aggressive carcinomas),
normal scintiscan, flare phenomenon (osteoblastic activity that reflects bone healing after
chemotherapy but not advancing disease) and hypercaptation in soft-tissue lesions.
As regards the degree of confidence, scintiscans are non specific for determining the cause
of increased uptake (lower specificity) since the findings of scintigraphy reflect the
metabolic reaction of bone. Bone scintiscans have a poor spatial and contrast resolution.
Many benign processes can produce an isotope uptake that mimics metastases (false-
positive finding) and approximately one third of patients show a solitary area associated
with a benign process. Thus, biopsy confirmation may also be needed for the final
diagnosis of suspicious lesions. The differential diagnosis includes metabolic illness,
osteomalacia, trauma, arthritis, osteomyelitis, infarctions and Pagets disease. Therefore,
suspicious lesions or a single hot spot should be verified by other imaging modalities
such as radiography, CT, or MRI.
Scintigraphy does not detect pure lytic metastases when bone turnover is slow or when the
site is avascular. The main advantages of scintigraphy are widely available and can screen
rapid whole-body images at a reasonable cost.

Medical Imaging

344
When assessing the therapeutic response, scintigraphy should be supplemented by images
obtained with other modalities to provide a valid baseline for the assessment of bone tumor.
In addition, it can take six months or longer to detect a response in the scintigraphy because
of the confounding effect of the flare phenomenon.

Fig. 2.
99m
Tc bone scintigraphy images of a patient with prostatic carcinoma. This is a classical
pattern with the presence of randomly distributed focal lesions throughout the skeleton. Bone
metastases appear as areas of increased tracer uptake (hot spots) (red arrows).
4.2 Single-photon-emission computed tomography (SPECT)
SPECT is a nuclear medicine tomographic imaging technique using gamma rays.
99m
Tc-
MDP, the same radionuclide used in conventional skeletal scintigraphy is the most
frequent radionuclide used. It is very similar to conventional nuclear medicine planar
imaging using a gamma camera. However, images are acquired in a cross-sectional rather


345
than a planar fashion. It is able to provide 3D information. Compared with a conventional
scintigraphy, the precise anatomic localisation provided by SPECT ensures a better
differentiation between benign and malignant diseases. SPECT has greater sensitivity and
specificity than scintigraphy for detecting vertebral and pelvis metastases (sensitivity 87-
92% and specificity 73-100%). Like scintigraphy, it is possible that SPECT does not detect
lytic lesions (Beheshti et al., 2009).
Because of limited availability and poor diagnostic imaging quality, SPECT is not typically
used for the initial detection of metastatic disease (Chua, 2009). SPECT-CT is useful for the
assessment of lesions that are not determined in scintigraphy. When scintigraphy are
reviewed immediately, SPECT or SPECT-CT imaging can be performed in the same session
without administering a second dose of radionuclide.
4.3 Positron emission tomography (PET)
PET is a nuclear medicine imaging technique that produces tomographic images through
the detection of high-energy photon pairs emitted during positron decay of a radioisotope.
PET visualizes the uptake of positron-emitting radioisotope by tissues, but, for skeletal
metastases, two radiopharmaceuticals are typically used:
18
F-Fluoride, a bone turnover
tracer, and
18
FDG a tumor tracer. PET can be used for whole-body scanning to detect
metastases in either soft tissue or bone. PET provides a 3D image of tracer concentration
within the body that is then constructed by computer analysis. In modern scanners, 3D
imaging is often accomplished with the CT or MRI on the patient in the same machine. PET
has a limited spatial resolution and complementary CT scanning or MRI is required to
localize the specific area (Costelloe et al., 2009; Cook ,2010).
Its main advantages are: high sensitivity (84-100%), biochemical and molecular information
(does not always coincide with morphological explorations). PET is an effective technique
that is used in the evaluation of skeletal metastatic disease, particularly when combined
with CT, because of its high resolution and coverage of the whole body (Ben-Haim, 2009 &
Israel, 2009). PET has the advantage of permitting quantification of therapeutic response
using the maximum standard uptake (mSUV) value. PET can be a sensitive modality for
diagnosing and monitoring osteolytic response (Beheshti et. al, 2009; Wahl et al., 2009).
However, the disadvantages may include PET cyclotron or the generator method, its high
radiation, its high cost, its limited availability, and the additional time required for scanning
with respect to other imaging modalities.
4.4
18
FDG-PET
18
FDG (2-deoxy-2-[18F]-fluoro-D-glucose or fluorodeoxyglucose) has been used to measure
glucose metabolism in many types of primary cancer and can be useful for distinguishing
benign from malignant bone lesions. It is an analog of glucose and represents the most
widely used PET radiotracer in daily practice. The mechanism of uptake in tumor cells
consist in the diffusion facilitated by glucose transporters (GLUTs), phosphorylation by
hexokinase and subsequent metabolic trapping within the cell. It is a reflection of glucose
metabolism consumption or tissue level.
Although it is a very sensitive marker, it is not very specific, since increases in tissue uptake
of
18
FDG is not always synonymous with cancer. The main clinical application of this

Medical Imaging

346
technique consists in diagnosing, staging and restaging of various cancer types (figure 3). It
is very effective in detecting metastases of breast, lung, esophagus, colon, thyroid, head,
neck, melanoma and lymphoma.
From the dosimetric point of view, it should be noted that an administration of 350 MBq of
18
FDG results in a radiation exposure of approximately 6,3-7 mSV and PET/CT examination
may result in a radiation exposure of more than 22-23 mSv, near the upper range of the dose
given for regular diagnostic abdomen CT (3,5-25 mSv).

Fig. 3. PET images in a patient with lung carcinoma. (A)
18
FDG PET shows the primary
tumor (red arrow, 4x4.3cm, mSUV=27) with contralateral clavicle nodal metastases (green
arrow, 2.7x2.5 cm, mSUV=19.2). (B) Incidental right humerus focal uptake is noted (red
cross) which is suspicious for bone metastases modifying the initial disease staging from the
patient to stage IV.


347
4.5
18
F-fluoride-PET
The mechanism of
18
F-Fluoride uptake is similar to that of
99m
Tc-MDP, the tracer used in
scintigraphy: its accumulation depends on osteoblastic activity and local blood flow. It is an
indicator of bone turnover, but not of a specific tracer of bone metastases. At present, it is
the second most commonly used technique in oncology, playing an important role in the
detection of bone involvement. A number of studies have shown that the diagnostic
accuracy of
18
F-fluoridePET is better than scintigraphy (95,7% vs 75,4%) for detecting bone
metastases in lung, prostate, thyroid and breast cancer, especially when using PET/CT
(Withofs et al., 2011). Moreover, it appears to be more sensitive than SPECT.
The specificity of
18
F-Fluoride when used with PET alone is 62% (Groves et al., 2007), which
demands a CT, mainly pelvic and lumbar injury, and may also detect extra-osseous
metastases. An advantage is that it can detect early osteoblastic changes in slow growing
tumors, where
18
FDG has a limited value.
18
F-fluoride appears to be equally sensitive to
osteoblastic and osteolytic metastases and can identify extremely early osteoblastic changes
in response to metastatic deposits. It also seems to be superior in the detection the low
avidity of bone tumor metastases for
18
FDG as in some thyroid cancers.
Moreover, quantitative
18
F-Fluoride PET can determine kinetics of fluoride incorporation
into bone as a measure of fluoride transport, bone formation and turnover. Kinetic analysis
(fluoride transport and flux) may be useful in assessing changes in bone turnover in
response to therapy, and in bone metastases with values exceeding those of normal bone
(Doot et al., 2010).
5. X-ray techniques
5.1 Radiography
Radiography is an x-ray imaging technique that produces a 2D picture of human body
structures superimposed on each other, whose insides absord different amounts of radiation
depending on the densities of its components. Bone metastases can appear on radiography
as areas of absent density (osteolytic), as disrupted bone structure or as high density
(osteoblastic). Radiography is a good method for characterizing bone metastases: osteolytic,
osteoblastic or mixed. The ocurrence of bone metastases suggests a primary tumor, eg.
osteolytic metastases are typical in breast and lung carcinomas whereas osteoblastic
metastases are more common in prostatic carcinoma (Hamaoka et al., 2004). Radiography is
used to evaluate symptomatic sites (bone pain) or confirm findings of other imaging
modalities (especially for evaluating suspicious lesions on scintigraphy). Radiography may
be used to assess the risk of a pathological fracture. False positive diagnosis may result
because osteolytic metastases can mimic osteoarthritis. Osteoblastic metastases may be
difficult to distinguish from other sclerotic bone lesions such as tuberous sclerosis.
The disadvantage is that between 30 and 50% of normal bone mineral content must be lost
before lytic lesions become apparent on radiography (Wahl et al., 2009). Moreover, lesions in
trabecular bone are more difficult to detect by radiography than cortical. The main
advantages of radiography are that it is cost effective and widely available. However, it is
not recommended for screening because of its limited sensitivity, which depends partly on
location (Rybal & Rosenthal, 2001).

Medical Imaging

348
Radiography can detect the response of osteolytic lesions by depicting active bone
formation (blastic change) or the reappearance and normalization of bone structure. But,
it may be difficult to differentiate bone metastases from healing or previously
unidentified sclerotic lesions. In several studies, indicators of response to treatment on
radiography are correlated with clinical symptoms or other changes better than
scintigraphy. Nevertheless, changes in radiography are not apparent until three to six
months after the initiation of treatment (Even-Sapir, 2005).
5.2 Computed tomography (CT)
CT consists of a large series of 2D x-ray images taken around a single axis of rotation. The
computer assisted reconstruction is used to generate 3D pictures. The radiologic appearance
in the CT bone window setting offers skeletal detail because it can distinguish between
materials of different densities. It is more sensitive than radiography in the detection of bone
metastases. CT is useful for evaluating radiography negative areas in symptomatic patients
or where metastases are suggested clinically. CT is important in the evaluation of focal
abnormalities observed in bone scintigraphy that cannot be confirmed using radiographs.
Osteolytic, osteoblastic and mixed bone metastases are well despicted on CT. It can detect
metastases in the marrow. Bone metastases appear more attenuated than the marrow. But
the usefulness of CT in detecting early deposits in bone marrow is limited. Although CT is
superior to radiography some advanced destructive lesions on trabecular bone may not be
visible in the absence of cortical bone involvement, so CT is less apparent than the marrow
changes visualized on MRI (Hamaoka et al., 2004). CT is also better than radiography and
scintigraphy for depicting lesions in the spine and calvarium. CT is useful in guiding needle
biopsy in bones such as vertebrae or ilia.
The disadvantage of CT is that its high radiation dose makes CT unsuitable as a screening
tool. Thus, limited anatomic areas can be scanned simultaneously (Rybal & Rosenthal, 2001).
When assessing the therapeutic response, sclerosis of a lytic component on CT suggests a
response to treatment. Notwithstanding, lysis or the appearance of new lysis or an
increase in the size of blastic lesion represents a disease progression (Hamaoka et al., 2009;
Wahl et al., 2009).
5.3 Dual-energy X-ray absorptiometry (DXA)
DXA is a method for the evaluation of bone mineral density (BMD) and monitoring
patients with osteopenia or osteoporosis, entailing minimal exposure to radiation (x-ray).
Two x-ray beams with differing energy levels are aimed at the patient's bones. When soft
tissue absorption is subtracted out, the BMD can be determined from the absorption of
each beam by bone.
BMD reflects the balance between bone formation and resorption. This method has also
been applied in several clinical trials to assess the therapeutic outcome of patients with bone
metastases receiving systemic treatments (Vassiliou et al., 2011). The BMD of skeletal
metastases increases in patients responding to treatment and was significantly correlated
with the changes imaged on skeletal x-rays and CT.


349
6. Magnetic resonance imaging (MRI)
MRI produces high quality images of the inside of the human body. It uses a powerful
magnetic field to align the magnetization of some atoms in the body (hydrogen nuclei or
protons of the water molecules) and a radiofrequency fields to systematically alter the
alignment of this magnetization. When a patient is subjected to the powerful magnetic
field of the scanner, the magnetic moments of some of these protons change and align
with the direction of the field. An image can be constructed because the protons in
different tissues return to their state of equilibrium at different rates (tissue variables: spin
density, T1 and T2 relaxation times, flow and spectral shifts) (Wu, 2011). Unlike the other
techniques, MRI does not use ionizing radiation.
MRI diagnosis of bone metastases is characterized by long T1 relaxation times, whereas
T2 relaxation times are variable, depending on tumor morphology. Metastases are seen as
focal or diffuse areas of hypointensity on T1-weighted images which contrasts with the
surrounding high signal marrow fat. Metastases can often be distinguished from focal
deposits of red marrow because the latter are more focal and may have centrally located
fat (bulls eye sign). In the case of T2-weighted images, bone metastases are seen as areas
of intermediate or high signal intensity which contrast with normal marrow because of
their high water content, and they are commonly, surrounded by a rim of bright signal
(halo sign) (Schmidt, 2007).
MRI can identify bone metastases at an early stage. CT can also visualize bone marrow
lesions, but the resolution is poor. The advantages of MRI include distinguishing benign
(osteoporotic) from malignant causes of vertebral compression fracture, detecting spinal
cord compression and the capability to obtain sagittal views of the entire spine to be
assessed in one imaging session. Nevertheless, MRI is less advisable than radiography or CT
for detecting the destruction of cortical bone structure because, cortical appears black on T1
and T2-weighted sequences.
Difussion-weighted imaging (DWI) is a MRI technique based on the imaging of molecular
mobility of water (i.e. diffusion). Recently, DWI has been proposed as a time-and cost-
effective detection to image tumor deposits throughout the whole body. DWI is
considered a highly sensitive method for the detection of bone metastases (Luboldt, 2008;
Takenaka, 2009).
With whole-body MRI, technical problems arise from the relatively long examination times
(4560 min) and the limited depiction of metastases in small bones. However, the
development of ultrafast pulse sequences (Turbo STIR (short tau inversion recovery)) for
whole-body MRI may substantially decrease imaging time. MRI scanning has
contraindications, these include: pacemakers, metallic implants or ferromagnetic foreign
bodies (e. g. shell fragments). Many benign process can produce the appearance of bone
metastases on MRI ( e.g in vertebral column: degenerative disk disease, osteomyelitis,
benign compression fracture, infarcts or Schmorls nodes).
When assessing the therapeutic response, MRI is optimal for showing spinal cord status and
changes in the bone marrow but not suited to showing lytic or blastic change in bone
structure.

Medical Imaging

350
7. Imaging modalities in major cancer entities
7.1 Breast cancer
The skeleton is the most common site of distant metastases in breast cancer (30%85%). The
vertebral column is the most common site of spread followed by the ribs. Bone metastases of
breast cancer are lytic, blastic or mixed. Scintigraphy is indicated in patients with advanced
disease or when bone involvement is clinically suspected. The use of
18
F-FDG PET it is not a
routine staging procedure.
18
F-FDG PET allows for the detection of both soft-tissue and
skeletal sites of disease. The superiority of
18
F-FDG PET was reflected mainly in the
detection of bone metastases, which were predominately lytic. By using PET/CT, sclerotic
lesions overlooked by the PET part of the study can be identified on the CT part. In this
setting, the high sensitivity of
18
F-FDG PET for detecting marrow and lytic lesions and the
high sensitivity of CT for detecting sclerotic lesions are complementary. Notwithstanding,
some types of breast cancer, primarily well-differentiated histologic subtypes including
some of the tubular and lobular ones, are less
18
F-FDG avid and so are their metastases. A
recognized scintigraphy effect of antiestrogen therapy commonly applied in patients with
breast cancer is the flare reaction. Clinically, it may be difficult to differentiate the flare
reaction from disease progression. The initial agonist effect to therapy is also associated with
increased tumor
18
F-FDG uptake. A change in
18
F-FDG uptake can be detected as early as 10
days after initiation of treatment compared with several weeks that are often required to
make this assessment on the basis of clinical symptoms (Niikuraet al., 2011).
7.2 Prostate cancer
Prostate cancer frequently metastatizes to bone (35-85%). Bone metastases of prostate cancer
are predominantly blastic. The vertebral column (lumbar), pelvis, sternum, ribs and femur
are the most common sites of spread. Staging of newly diagnosed prostate cancer is essential
for guiding treatment. Patients with low-risk prostate cancer are unlikely to have metastatic
disease on scintigraphy. Patients are referred for scintigraphy mainly if they are considered
to be at high risk for bone metastases, with high PSA levels, a locally advanced disease, or a
high Gleason score. Scintigraphy is the most widely used method for evaluating skeletal
metastases of prostate carcinoma. The role of
18
F-FDG PET seemed to be limited in this type
of cancer as both the soft-tissue sites of disease and bone metastases were reported to be
18
F-
FDG negative or to show only a low-intensity uptake in many patients. PET/CT was
suggested to overcome the problem of pelvic tumor sites being obscured by the radioactive
urine. Using
18
F-FDG PET for monitoring the response to treatment, a decline in tumor
glucose uptake measured as early as 48 h after androgen withdrawal, preceding any change
in tumor volume or in PSA levels. Other PET tracers suggested for assessment of prostate
cancer include
11
C- or
18
F-labeled choline and acetate,
11
C-methionine,
18
F-
fluorodihydrotestosterone, and
18
F-fluoride. The latter may be highly sensitive for detecting
bone metastases in patients with prostate cancer (Even-Sapir, 2007).
7.3 Lung cancer
Bone metastases are diagnosed at initial presentation in 4%60% of patients with non-small-
cell lung cancer (NSC). Bone pain is usually considered an indicator of skeletal metastases,
but up to 40% of lung cancer patients with proven bone metastases are asymptomatic.


351
Surgical resection offers the highest probability of a favorable outcome in patients with NSC
lung carcinoma. However, the survival of patients who undergo surgery remains low,
probably because of presurgical understaging. Clinical staging at presentation has been
performed by means of CT of the thorax through the liver and adrenals, CT or MRI of the
brain, and scintgraphy for assessment of bone involvement. This staging algorithm remains
the most commonly used in places where
18
F-FDG PET is not a routine staging modality of
lung cancer. Including SPECT in the acquisition protocol of scintigraphy could improve the
diagnostic accuracy scintigraphy in detecting bone metastases in patients with lung cancer.
If necessary, scintigraphy can be complemented by CT or regional MRI for further
assessment of unclear lesions.
18
F-FDG PET and PET/CT were recently reported to be of
value in assessing the presence of soft-tissue and bone spread in patients with NSC lung
cancer. They reported that
18
F-FDG PET had a high positive predictive value and a lower
false-positive rate compared with scintigraphy (Even-Sapir, 2007).
The utility of PET in the detection of bone metastases is demonstrated in the following case
(figures 4 and 5). A patient with lung carcinoma where the scintigraphy did not detect pure
lytic metastases caused by slow bone turnover or an avascular site. The PET scan revealed a
focus of increase FDG uptake suspicious for osteolytic metastases in the spine. MRI was
requested for confirmation of diagnosis.

Fig. 4.
99m
Tc-MDP bone scintigraphy images of a patient with lung carcinoma. It is a normal
scintiscan. The bone scan does not revealed suspicious foci of uptake

Medical Imaging

352

(A) PET show the primary tumor (red arrow). (B) A subsequent PET scan revealed a focus of increased
FDG uptake in the spine (L2 vertebral body, red cross) which is suspicious for a osteolytic metastases.
Fig. 5. PET images in a patient with lung carcinoma. The utility of PET in the detection of
bone metastases is demonstrated in this case. The bone scan, of the same patient, is showed
in figure 4: a normal scintiscan (false negative case).
8. References
Beheshti M, Langsteger W, Fogelman I. (2009). Prostate cancer: role of SPECT and PET in
imaging bone metastases. Semin Nucl Med 39: 396-407
Ben-Haim S, Israel O. (2009). Breast cancer: role of SPECT and PET in imaging bone
metastases. Semin Nucl Med 39(6):408-15
Chaudhari AJ, Darvas F, Bading JR, Moats RA, Conti PS, Smith DJ et al 2005. Hyperspectral
and multispectral bioluminescence optical tomography for small animal imaging.
Phys Med Biol 50:54215441
Chua S, Gnanasegaran G, Cook GJR. (2009). Miscellaneous cancers (lung, yhyroid, renal
cancer, myeloma, and neuroendocrine tumors): Role of SPECT and PET in imaging
bone metastases. Semin Nucl Med 39: 416-430
Coleman RE. (2006). Clinical features of metastatic bone disease and risk of skeletal
morbidity. Clin Cancer Res 15;12(20 Pt 2):6243s-6249s
Coleman RE & McCloskey EV. (2011). Bisphosphonates in oncology. Bone 49(1):71-6
Costelloe CM, Rohren EM, Madewell JE, Hamaoka T, Theriault RL, Yu TK, Lewis VO, Ma J,
Stafford RJ, Tari AM, Hortobagyi GN, Ueno NT. (2009). Imaging bone metastases
in breast cancer: techniques and recommendations for diagnosis. Lancet Oncol
10(6):606-14
Cook GJ. (2010) PET and PET/CT imaging of skeletal metastases. Cancer Imaging 10: 1-8


353
Doot RK, Muzi M, Peterson LM, Shubert EK, Gralow JR, Specht JM et al.(2010). Kinetic
analysis of 18F-fluoride PET images of breast cancer bone metastases. J Nucl Med 51:
521-7
Eriksen EF. (2010). Cellular mechanisms of bone remodeling. Rev Endocr Metab Disord
11(4):219-27
Even-Sapir E. (2005). Imaging of malignant bone involvement by morphologic,
scintigraphic, and hybrid modalities. J Nucl Med 46(8):1356-67
Even-Sapir E. (2007). PET/CT in malignant bone disease. Semin Musculoskelet Radiol
11(4):312_321
Fili S, Karalaki M, Schaller B. (2009). Mechanism of bone metastasis: the role of
osteoprotegerin and of the host-tissue microenvironment-related survival factors.
Cancer Lett 28;283(1):10-9.
Gonzlez-Sistal, A. Baltasar Snchez, A (2006). A complementary method for the detection
of osteoblastic metastases on digitized radiographs. J Digit Imaging 19(3):270-5
Gonzlez-Sistal A. (2007) Diagnostic Imaging Modalities of Osteoblastic Metastases Hospital
Imaging & Radiology Europe 2 (1): 16-17
Groves AM, Win T, Haim SB, Ell PJ. (2007). Non-[18F]FDG PET in clinical oncology. Lancet
Oncol 8(9):822-30
Ha HK, Lee W, Park HJ, Lee SD, Lee JZ, Chung MK. (2011). Clinical significance of
CXCL16/CXCR6 expression in patients with prostate cncer. Mol Med Report
4(3):419-24
Hadjidakis DJ, Androulakis II. (2006). Bone remodeling. Ann N Y Acad Sci 1092:385-96
Hamaoka T, Madewell JE, Podoloff DA, Hortobagyi GN, Ueno NT. (2004) Bone imaging in
metastatic breast cncer. J Clin Oncol 15;22(14):2942-53
Hamaoka T, Costelloe CM, Madewell JE, Liu P, Berry DA, Islam R, Theriault RL,
Hortobagyi GN, Ueno NT. (2010). Tumour response interpretation with new
tumour response criteria vs the World Health Organisation criteria in patients with
bone-only metastatic breast cancer. Br J Cancer 16;102(4):651-7
Kim JB, Urban K, Cochran E, Lee S, Ang A, Rice B, Bata A, Campbell K, Coffee R, Gorodinsky
A, Lu Z, Zhou H, Kishimoto TK, Lassota P. (2010). Non-invasive detection of a small
number of bioluminescent cancer cells in vivo. PLoS One 23;5(2):e9364
Kwon H, Enomoto T, Shimogawara M, Yasuda K, Nakajima Y, Ohmiya Y. (2010).
Bioluminescence imaging of dual gene expression at the single-cell level.
Biotechniques 48(6):460-2
Leblond F, Davis SC, Valds PA, Pogue BW. (2010). Pre-clinical whole-body fluorescence
imaging: Review of instruments, methods and applications. J Photochem Photobiol B
21;98(1):77-94
Logothetis CJ, Lin SH. (2005). Osteoblasts in prostate cancer metastasis to bone. Nat Rev
Cancer 5(1):21-8
Luboldt W, Kufer R, Blumstein N, Toussaint TL, Kluge A, Seemann MD, Luboldt HJ. (2008).
Prostate carcinoma: diffusion weighted imaging as potential alternative to
conventional MR and 11C-choline PET/CT for detection of bone metastases.
Radiology 249: 1017 1025
Mercatali L, Ibrahim T, Sacanna E, Flamini E, Scarpi E, Calistri D, Ricci M, Serra P, Ricci R,
Zoli W, Kang Y, Amadori D. (2011). Bone metastases detection by circulating
biomarkers: OPG and RANK-L. Int J Oncol 39(1):255-61
Moore AE, Blake GM, Fogelman I. (2007). A study to determine the dependence of 99mTc-
MDP protein binding on plasma clearance and serum albumin concentration. Nucl
Med Commun 28(3):187-92

Medical Imaging

354
Nahrendorf M, Keliher E, Marinelli B, Waterman P, Feruglio PF, Fexon L, Pivovarov M,
Swirski FK, Pittet MJ, Vinegoni C, Weissleder R.(2010). Hybrid PET-optical imaging
using targeted probes. Proc Natl Acad Sci USA 27;107(17):7910-5
Niikura N, Costelloe CM, Madewell JE, Hayashi N, Yu TK, Liu J, et al (2011). FDG-PET/CT
Compared with conventional imaging in the detection of distant metastases of
primary breast cncer. Oncologist; 16 (8):1111-9
Ntziachristos V, Ripoll J, Wang LV, Weissleder R. 2005 Looking and listening to light: the
evolution of whole-body photonic imaging. Nat Biotechnol 23:313320
Ouellet V, Tiedemann K, Mourskaia A, Fong JE, Tran-Thanh D, Amir E, Clemons M, Perbal B,
Komarova SV, Siegel PM. (2011). CCN3 impairs osteoblast and stimulates osteoclast
differentiation to favor breast cancer metastasis to bone. Am J Pathol 178(5):2377-88
Roodman GD. (2004). Mechanisms of bone metastasis. N Engl J Med 15;350 (16):1655-64.
Rybak LD, Rosenthal DI. (2001). Radiological imaging for the diagnosis of bone metastases.
Q J Nucle Med 45(1):53-64
Schmidt GP, Schoenberg SO, Schmid R, Stahl R, Tiling R, Becker CR, Reiser MF, Baur-
Melnyk A. (2007). Screening for bone metastases: whole-body MRI using a 32-
channel system versus dual-modality PET-CT. Eur Radiol 17(4):939-49
Snoeks TJ, Khmelinskii A, Lelieveldt BP, Kaijzel EL, Lwik CW (2011). Optical advances in
skeletal imaging applied to bone metastases. Bone 48(1):106-14
Suva LJ, Washam C, Nicholas RW, Griffin RJ.(2011). Bone metastasis: mechanisms and
therapeutic opportunities. Nat Rev Endocrinol 7(4):208-18
Takenaka D, Ohno Y, Matsumoto K, Aoyama N, Onishi Y, Koyama H, Nogami M,
Yoshikawa T, Matsumoto S, Sugimura K. (2009). Detection of bone metastases in
non-small cell lung cancer patients: comparison of whole-body diffusion-weighted
imaging (DWI), whole-body MR imaging without and with DWI, whole-body
FDG-PET/CT, and bone scintigraphy. J Magn Reson Imaging 30(2):298-308
Trinkaus M, Ooi WS, Amir E, Popovic S, Kalina M, Kahn H, Singh G, Gainford MC,
Clemons M. (2009). Examination of the mechanisms of osteolysis in patients with
metastatic breast cancer. Oncol Rep 21(5):1153-9
Vassiliou V, Andreopoulos D, Frangos S, Tselis N, Giannopoulou E, Lutz S. (2011). Bone
Metastases: Assessment of Therapeutic Response through Radiological and Nuclear
Medicine Imaging Modalities. Clin Oncol (R Coll Radiol) [Epub ahead of print]
Wahl RL, Jacene H, Kasamon Y, Lodge MA. (2009). From RECIST to PERCIST: Evolving
Considerations for PET response criteria in solid tumors. J Nucl Med 50 Suppl
1:122S-50S
Welch HG, Black WC. (2010). Overdiagnosis in cncer. J Natl Cancer Inst. 5;102(9):605-13
Weilbaecher KN, Guise TA, McCauley LK. (2011). Cancer to bone: a fatal attraction. Nat Rev
Cancer 2011 Jun;11(6):411-25
Withofs N, Grayet B, Tancredi T, Rorive A, Mella C, Giacomelli F, et al. (2011). 18F-
fluridePET/CT for assessing bone involvement in prostate and breast cancers. Nucl
Med Commun 32: 168-76
Wu LM, Gu HY, Zheng J, Xu X, Lin LH, Deng X, Zhang W, Xu JR. (2011). Diagnostic value
of whole-body magnetic resonance imaging for bone metastases: a systematic
review and meta-analysis. J Magn Reson Imaging 34(1):128-35
Ye XC, Choueiri M, Tu SM, Lin SH.(2007). Biology and clinical management of prostate
cancer bone metstasis. Front Biosci 1;12:3273-86
Anan Liu
1
*
, Kang Li
2
and Tong Hao
3
1
School of Electronic Information Engineering, Tianjin University
2
Microsoft Corporation
3
School of Chemical Engineering and Technology, Tianjin University
1,3
China
2
USA
1. Introduction
Measurement of the proliferative behaviors of cells in vitro is important to many biomedical
applications ranging from basic biological research to advanced applications, such as drug
discovery, stem cell manufacturing, and tissue engineering. Critical to such measurement is
accurate identication and localization of mitosis, which is the process whereby a eukaryotic
cell separates the chromosomes in its cell nucleus into two identical sets in two daughter
nuclei. In the early years, it is possible to manually identify incidents of mitosis because
mitotic cells in culture tend to retract, round up, and exhibit intensied surrounding halos
under phase contrast illumination (Fig.1) for short-period, small-scale studies. Recently,
many cell proliferation assays have been developed for high throughput cell imaging and
analysis. Especially, phase contrast microscopy is a superior imaging modality because it
enables continuous monitoring of live cells without requiring destructive methods of cell
manipulation, such as cell lysis and staining. Consequently, the need for extended-time
observation and the proliferation of high-throughput imaging have made automated image
analysis mandatory.
The state-of-the-art mitosis detection methods can be categorized into three classes,
tracking-based, tracking-free, and hybrid approaches. Tracking-based approaches rely on
cell tracking to determine individual cell trajectories, and then identify mitosis based on
the temporal progression of cell features along their trajectories (Al-Kofahi et al., 2006; Bise
et al., 2009; Debeir et al., 2005). The dependency on cell tracking is a severe burden because
tracking per se is a challenging task. Tracking-free approaches alleviate this burden and can
detect mitosis directly in an image sequence. One representative technique was proposed
by Li et al (Li, Kanade, Chen, Miller & Campbell, 2008), which applies a cascade classier
to classify volumetric sliding windows of an image sequence with 3D Haar-like features.
Major drawbacks of this approach include the requirement of a large amount of training
*
Corresponding Author

A Hierarchical Framework for Mitosis Detection
in Time-Lapse Phase Contrast Microscopy
Image Sequences of Stem Cell Populations
17
Fig. 1. Mitosis Sample. All the frames in one mitosis sequence are concatenated into one
gure to show the visual pattern transition. One mitosis sequence can be visually divided
into three stages, before mitosis, during mitosis, and after mitosis. Especially, all mitotic
regions within the stage of during mitosis appear like 8 shape.
data and the lack of location specicity of detection. Hybrid approaches aim to construct a
self-contained solution by leveraging the advantages of the previous two methods. These
approaches typically consist of candidate sequence detection, sequence feature extraction,
and classication as consecutive steps. To detect mitosis candidates, earlier methods (Eccles
& Klevecz, 1986) apply thresholding and morphological ltering to extract bright halos
surrounding potentially mitotic cells in each image, and then group the extracted regions in
successive images based on their spatial relationship. Subsequently, to identify mitosis, Eccles
et al (Eccles & Klevecz, 1986) employed a ring shape detector to locate the mother and two
daughter cells; Gallardo et al (Gallardo et al., 2004) adopted a hidden Markov model to classify
candidates based on temporal patterns of cell shape and appearance features; Liang et al
(Liang et al., 2007) utilized conditional random eld model to identify mitosis with shape and
texture features in the mitotic regions. However, these methods can only recognize mitosis
event without precise spatial and temporal localization during cell proliferation because of
the lack of the ability to analyze the latent structure of mitosis progression.
Different from previous work simultaneously identifying and localizing mitosis event, we
divided the problem of mitosis event detection into two sub-tasks: 1) mitosis identication:
identifying the visual pattern transition from 0 shape mother cell to 8 shape mother cell and
nally to two separate daughter cells in Fig.1; 2) mitosis localization: localizing the starting
point, the rst 8 shape pattern, and ending point, the last 8 shape pattern, during mitosis
progression in Fig. 1. The proposed framework follows the spirit of the hybrid approach. It
takes a phase contrast microscopy image sequence as input, and automatically pinpoints the
time point at which mitosis occurs. First, model-based microscopy image preconditioning
(Li & Kanade, 2009) and volumetric segmentation are utilized to extract spatiotemporal
sub-regions in the input image sequence where mitosis potentially occurred. Then, a hidden
conditional random eld classier (Quattoni et al., 2007) is applied for mitosis identication.
By making signicant extension on our previous work on mitosis sequence identication
(Liu et al., 2010a;b), a conditional random eld model (Lafferty et al., 2001) is implemented
for mitosis localization. The rst two stages jointly optimize the recall and precision for
mitosis identication while the third stage pinpoints the time point at which mitosis occurs,
minimizing the mean error and standard deviation of mitosis location. The superiority
lies in three aspects: 1) nonnegative mixed-norm preconditioning method can avoid over
segmentation and ad hoc image processing; 2) volumetric region grow method can avoid the
explicit cell tracking; 3) latent contextual information can be discoveredby hidden conditional
random eld and conditional random eld classiers for both sub-tasks. Consequently the
356 Medical Imaging
A Hierarchical Framework for Mitosis Detection in Time-lapse Phase Contrast Microscopy Image Sequences of Stem Cell Populations 3
proposed framework has high generalization ability and can be straightforwardly adapted to
different cell types.
The rest of the paper is structured as follows. In Section 2, we introduce the hierarchical
framework for mitosis event detection. Then, the experimental methods and results are
respectively detailed in Section 3 and 4. At last, the conclusion and future work are presented.
2. Mitosis detection framework
As shown in Fig.2, the proposed hierarchical framework for mitosis event detection consists
of three steps: Mitosis Candidate Extraction, Mitosis Identication, and Mitosis Localization. We
will present the technical details of each step in the subsequent sections.
Fig. 2. Mitosis detection framework
357
A Hierarchical Framework for Mitosis Detection in Time-Lapse
Phase Contrast Microscopy Image Sequences of Stem Cell Populations
Fig. 3. Outputs by the key steps of the proposed framework. (a) Original image sequence; (b)
Outputs of image preconditioning; (c) Outputs of mitosis candidate extraction; (d) Outputs
of mitosis identication; (e) Outputs of mitosis localization.
2.1 Mitosis candidate extraction
This step serves the purposes of eliminating "easily" negative regions in the image sequence
where mitosis is unlikely to occur and extracting temporally continuous sub-sequences with
potential mitosis to facilitate subsequent sequence classication. The algorithm consists of
two sub-steps, image preconditioning and sequence segmentation.
2.1.1 Image preconditioning
Image segmentation in microscopy, especially in non-uorescence optical microscopy
modalities, is notoriously challenging due to inherent optical artifacts. Rather than treating
phase contrast microscopy images as general natural images and rushing into the image
processing warehouse for solutions, we study the properties of phase contrast optics to model
its image formation process. Li et als research revealed that the phase contrast imaging
system can be relatively well explained by a linear imaging model (Li, 2009; Li & Kanade,
2009; Yin et al., 2010). With this model, a quadratic optimization function with sparseness
and smoothness regularizations can be formulated to enhance cell regions and transforms the
original image into an ideal image with zero background and nonzero foreground regions
that correspond to potential mitotic cells as shown in Fig.3.(b). The detailed method contains
following two steps.
(a) Algebraic image model
The generic model for microscopy images consists of three components: 1) an imaging model
h() that represents the image formation process of the microscope; 2) an additive bias b(x, y)
358 Medical Imaging
that compensates for a nonzero background level, nonuniform illumination, and spatial
sensitivity variations of the detector; and 3) a noise model n() that accounts for imaging
and detection noise. The model can be written as:
g(x, y) = n(h( f (x, y)) + b(x, y)), (1)
with g(x, y) being the observed image, and f (x, y) being the ideal object image that we want to
retrieve, which could represent the optical path length distribution in the object, uorescence
intensities, or an phenomenological image that simply facilitates object segmentation. Under
the assumption of additive noise, the model above can be expressed succinctly in a linear
algebraic framework, given by
g = Hf + b + n Hf + constant, (2)
where g denotes a vectoried representation of the observed image, which is formed by
concatenating the image pixels in raster order. The vectors f , b, n are dened likewise for
the object, bias, and noise, respectively. The imaging model is expressed as a matrix-vector
multiplication between the NN transfer matrix H and f , which is adequate for representing
a wide range of microscopy image formation processes. Depending on the physics of phase
contrast microscopy, we choose the image formation model, H, dened by an effective point
spread function (or EPSF) in (Li, 2009):
EPSF(x, y) (x, y) (e
(x
2
+y
2
)/
2
1
e
(x
2
+y
2
)/
2
2
) (3)
where () is a Dirac delta function, and , are scaling factors. The EPSF approximates the
imaging function of phase contrast optics, which accounts for the formation of halo effects
around imaged cells.
(b) Nonnegative mixed-norm preconditioning
With the image model specied in Eq.2 and the assumption of additive noise, we need to
compute the ideal object image f given an observed image g. This inverse problem can be
tackled through a two-step process. Then an observedimage which is unfriendly for computer
analysis can be transformed into the one that facilitates automated object segmentation and
measurement.
a.Bias Elimination:As the rst step of preconditioning, a bias-corrected image can be obtained
by estimating the bias eld froman image. Under the assumption that the bias eld is smooth
and spatially slowly varying, it can be modeled as a K-th order polynomial surface:
b = Xp (4)
where p = (p
0
, p
1
, p
2
, . . .)
is the coefcient vector, and X is a matrix of N rows and (K +

1)(K + 2)/2 columns with the i-th row being (1, x
i
, y
i
, x
2
i
, x
i
y
i
, y
2
i
, . . .). To eliminate the bias,
the optimal coefcients can be computed rstly by solving the over-determined linear system
Xp = g. This amounts to solving the least-squares problem p
= argmin
p
||Xp g||
2
2
, which
has a closed-form solution p
= (X
X)
1
X
g. The bias-corrected image is then computed

as g
= g Xp
.
359
b.Object Reconstruction via Convex Optimization: The second step of preconditioning
reconstructs the object from the bias-corrected image. It is achieved by minimizing:
O( f ) = ||g
Hf ||
2
2
+Smoothness( f ) + Sparsity( f ), s.t. f 0 (5)
where || ||
2
denotes a L2 norm; the positive coefcients and control the importance; the
rst termpromotes data delity; the second and third terms encourage the spatial smoothness
and sparseness of the reconstructed image, respectively.
By employing an L2 (Tikhonov) smoothness term (Tikhonov, 1977) and a weighted L1
sparseness term (Donoho, 2004), Eq.5 can be transformed into:
O
1
( f ) = ||g
Hf ||
2
2
+||Rf ||
2
2
+ ||W f ||
1
, s.t. f 0 (6)
The smoothness term penalizes the L2 norm of the Laplacian of f , where the N N matrix
R represents an algebraic Laplacian operator with symmetric boundary condition. The
sparseness term penalizes the weighted L1 norm of f , where W is a diagonal matrix with
positive weights w
1
. . . w
N
on the diagonal and zeros elsewhere.
By rewriting O
1
( f ) in terms of the symmetric positive denite matrix Q = H
H +R
R and
the vector l = H
, and letting w denote the weight vector (w

1
. . . w
N
), the minimization
problem can be expressed as the following nonnegative-constrained quadratic program:
f
= argmin
f
1
2
f
Qf + (l + w)
f , s.t. f 0 (7)
We utilized the simple and efcient iterative algorithm, the sparseness-enhanced
multiplicative update (SEMU) algorithm (Li & Kanade, 2009), to solve the object function
above.
2.1.2 Sequence segmentation
After preconditioning, 3D seeded region growing (Silvela & Portillo, 2001) is applied to
the transformed image sequences to extract spatiotemporal sub-regions that correspond to
candidate mitosis sequences (Fig.3(c)). The algorithm relies on two automatically-determined
thresholds: a seeding threshold computed by Otsus optimal thresholding algorithm (Otsu,
1979)is used to detect seeds; and a lower threshold determined by Rosins unimodal
thresholding algorithm (Rosin, 2001) is used as the stopping criterion of region growing.
2.2 Mitosis identication
With the extracted candidate mitosis sequences, the goal of mitosis identication is to classify
the candidate mitosis sequences detected in the previous step as mitosis or not as shown
in Fig.3(d). To take advantages of the spatiotemporal context and the exible constellation
of latent structure during mitosis progression, the hidden conditional random eld (HCRF)
classier is utilized to model the visual pattern transition shown in Fig.1.
HCRF is a powerful discriminative model for temporal inference (Quattoni et al., 2007).
Comparing to generative models, like Hidden Markov Models (HMMs) (Rabiner, 1989),
HCRF does not assume observations are independent of each other and therefore can
360 Medical Imaging
incorporate long range dependencies for inference. On the other hand, different from other
discriminative models, like Conditional Random Fields(CRFs) (Lafferty et al., 2001), HCRF
use intermediate hidden variables to model the latent structure of the input domain and
can infer a single label for the entire input sequence with the training data not explicitly
labeled. Therefore, several literatures show that HCRF always outperforms HMMs and CRFs
(Gunawardana et al., 2005; Quattoni et al., 2007; Wang et al., 2006).
Mitosis identication can be formulated as the problem of predicting a label y given an
observation sequence X {x
t
}
T
t=1
, where x
t
denotes the feature descriptor of the t-th
frame in the observation sequence and y is a member of a sequence-wise label set Y
(Mitosis, Non mitosis). HCRF model also consists of a vector of hidden variables h
{h
t
}
T
t=1
where h
t
H captures the latent structure of the input domain and H is the set
of potential hidden states. A graphical representation of the HCRF model G = (V, E) where
V denotes the vertex set and E denotes the edge set, is shown in Fig.4 (a).
Given the denitions of the sequence-wise label y, the observation sequence X, the hidden
variables h and the model parameters , the HCRF model can be dened by:
p(y|X, ) =
h
p(y, h|X, ) =
1
Z
h
exp((y, h, X; )) (8)
where Z =
y
Y,hH
exp((y
, h, X; )) is a partition function; (y, h, X; ) R is a

potential function parameterized by as:
(y, h, X; ) =

jV
pP
1
f
1, p
(j, y, h
j
, X)
1, p
+

(j,k)E
pP
2
f
2, p
(j, k, y, h
j
, h
k
, X)
2, p
(9)
where f
1, p
and f
2, p
denote the P
1
node feature functions and P
2
edge feature functions
respectively;
1, p
,
2, p
are the components of , corresponding to node and edge parameters.
For mitosis identication, we dene the two kinds of node feature functions, f
1,1
and f
1,2
, as
follows:
f
1,1
(j, h
j
, X) = (x
j
), f
1,2
(j, h
j
, y) =
1, if h
j
H and y is Mitosis,
0, otherwise.
(10)
where (x
j
) is the feature descriptor of frame x
j
. Correspondingly,
1, p
consists of two parts,
1,1
for f
1,1
measuring the compatibility between observation x
j
and hidden state h
j
, and
1,2
for f
1,2
measuring the compatibility between latent variable h
j
and segment-wise label y. We
also dene the edge feature function as follows:
f
2,1
(j, k, y, h
j
, h
k
, X) =
1, if (h
j
, h
k
) T and y is Mitosis,
0, otherwise.
(11)
where T denotes all possible hidden state transitions from h
j
to h
k
within a mitosis sequence.
Correspondingly,
2,1
for f
2,1
measures the compatibility between an edge with hidden state
h
i
and h
j
and the segment-wise label y.
361
Given Tr = {(X
1
, y
1
), (X
2
, y
2
), . . . , (X
N
, y
N
)} be a set of training examples, the model
parameters can be learned by optimizing the objective function (Lafferty et al., 2001):
L() =
N
i=1
log p(y
i
|X
i
, )
1
2
2
||||
2
(12)
where N is the total number of training sequences. The rst termin Eq.12 is the log-likelihood
of the data; the second term is the log of a Gaussion prior with variance
2
, i.e., p()
exp(
1
2
2
||||
2
). The optimal model parameter
= argmax
L() can be obtained with

gradient ascent algorithm, such as BFGS (Avriel, 2003). Given an unseen test sequence X,
the best corresponding label y
can be computed by
y
= argmax
y
p(y|X,
) (13)
Fig. 4. A comparison of graphical structure. (a)HCRF; (b)CRF
2.3 Mitosis localization
Comparing to previous work only focusing on coarse level mitosis localization (Gallardo et al.,
2004; Liang et al., 2007), it is more important to localize the start and end of mitosis event in
spatio-temporal domain which will benet real-time quantitative analysis of mitotic rate and
the correlation between the mother cell and the daughter cells during cell division (Li, Miller,
Chen, Kanade, Weiss & Campbell, 2008). Different from our previous work only considering
the start and end of mitosis as a spatial visual pattern like 8 shape (Liu et al., 2010b), we
regard mitosis progressionas the spatio-temporal visual pattern changes from0 shape mother
cell to 8 shape cell and nally to two separate daughter cells. Consequently we propose
to model the visual pattern transition between adjacent frames with Conditional Random
Model (CRF), the most notably discriminative model proposed by Lafferty et al(Lafferty et al.,
2001), to localize the start and end of one mitosis event as shown in Fig.3(e). Comparing
to Support Vector Machine (Liu et al., 2010b) using only spatial information for mitotic cell
modeling, CRF can take good advantages of spatio-temporal context to rene the frame-wise
362 Medical Imaging
annotation for accurate mitosis localization. Furthermore, CRF can incorporate long range
dependencies between observations and their labels without the assumption of generative
dynamic Bayesian network models that the observations are independent given the values of
hidden variables.
The task of mitosis localization can be considered as the problem of predicting a sequence of
labels L = {l
t
}
T
t=1
given an observation sequence X = {x
t
}
T
t=1
, where l
t
is a member of a
frame-wise label set L of all possible labels. For one mitosis event, we designate L contains
three members, before mitosis (BM), during mitosis (DM), after mitosis (AM). Therefore, the
rst and the last frames of a segment with label DM within one mitosis sequence respectively
correspond to the start and end of mitosis event. CRF is an undirected probabilistic graphical
model (G
= (V
, E
)) as shown in Fig.4 (b).

Given the denitions of the frame-wise label sequence L, the observation sequence X, and
the model parameters , the CRF model can be mathematically formulated by:
p(L|X, ) =
1
Z
exp((L, X; )) (14)
where Z
=
L
exp((L
, X; )) is a partition function; (L, X; ) R is a potential

function parameterized by as:
(L, X; ) =

jV
qQ
1
f
1,q
(j, l
j
, X)
1,q
+

(j,k)E
qQ
2
f
2,q
(j, k, l
j
, l
k
, X)
2,q
(15)
where f
1,q
and f
2,q
denote the Q
1
node feature functions and Q
2
edge feature functions
respectively;
1,q
and
2,q
are the components of , corresponding to node and edge
parameters. We dene the node feature function as follows:
f
1,1
(j, l
j
, X) = (x
j
) (16)
where (x
j
) is the feature descriptor of frame x
j
and
1,1
for f
1,1
measures the compatibility
between observation x
j
and frame-wise label l
j
. We also dene the edge feature function as
follows:
f
2,1
(j, k, l
j
, l
k
, X) =
1, if (l
j
, l
k
) T
,
0, otherwise.
(17)
where T
denotes all possible state transitions from l

j
to l
k
within a mitosis sequence and
2,1
measures the compatibility between an edge linking label l
j
and l
k
and the frame x
j
and x
k
.
Given Tr
= {(X
1
, L
1
), (X
2
, L
2
), . . . , (X
M
, L
M
)} be a set of training examples, the model
parameters can be learned by optimizing the objective function:
L() =
M
i=1
log p(L
i
|X
i
, )
1
2
2
||||
2
(18)
where M is the total number of training sequences. The rst term in the objective function is
the log-likelihood. The second term denotes the Gaussian prior with variance
2
. Learning
the parameters of a CRF model is a convex problem so that the global optimality
=
argmax
L() can be guaranteed and obtained with gradient ascent algorithm, such as BFGS
363
(Avriel, 2003). Given a testing sequence X, with the optimal model parameters
, the label
L
of the input sequence is

L
= argmax
L
p(L|X,
) (19)
3. Experimental method
3.1 Data
Five phase contrast image sequences of C3H10T1/2 mouse mesenchymal stem cell
populations (American Type Culture Collection, Manassas, VA) were acquired, each
containing 1000 images. The multipotent C3H10T1/2 stemcells serve as a model for the adult
human mesenchymal stem cells that can differentiate into a variety of cell types. The growing
environment consists of Dulbeccos Modied Eagles Media (DMEM; Invitrogen, Carlsbad,
CA), 10% fetal bovine serum (Invitrogen, Carlsbad, CA) and 1% penicillin streptomycin
(PS; Invitrogen, Carlsbad, CA). The cells were observed during growth under a Zeiss
Axiovert 135TV inverted microscope with a 5, 0.15 N.A. objective and phase contrast optics.
Time-lapse image acquisition was performed every 5 minutes using a 12-bit Qimaging Retiga
EXi Fast 1394 CCD camera at 500ms exposure with a gain of 1.01. Each image consists of
13921040 pixels with a resolution of 19 m/pixel.
3.2 Feature representation
From Fig.1, it is obvious that mitotic regions typically do not have rich features, and their
shapes and appearances are highly irregular and exible. To represent the feature of each
frame within one mitosis sequence, we computed the GIST descriptor for each frame to
formulate (x
i
). GIST is a well-known visual feature for image classication and can
represent all levels of processing, from low-level features (e.g., color, spatial frequencies)
to intermediate image properties (e.g., surface, volume) and high-level information (e.g.,
objects, activation of semantic knowledge) (Oliva & Torralba, 2001). The superiorities of
GIST for mitotic region description lie in three aspects: 1) GIST representation can bypass
the segmentation of individual objects. Precise cell region segmentation in phase contrast
microscopy is a notoriously difcult problem due to the phenomena of halo or shade-off.
Consequently, shape features, like Histogram of Oriented Gradients (Dalal & Triggs, 2005)
and Shape Context (Belongie et al., 2002), are usually not satisfactory descriptors for this task.
Comparatively, the computation of GIST only uses spectral and coarsely localized information
and therefore formulates a holistic representation of entire image without requirement to
precisely extract edge or boundary in the image; 2) GIST is invariant to scale and rotation.
Mitotic regions usually experience drastic changes in both size and orientation during
mitosis progression. GIST physically represents the dominant spatial structure of an image
including a set of perceptual dimensions, naturalness, openness, roughness, expansion, and
ruggedness. Therefore GIST is a robust descriptor for mitotic regions; 3) GIST is a compact
descriptor. GIST can be formulated from localized energy spectrum (spectrogram) with both
intensity and local structure information. Due to the high dimension and redundancy of
spectrogram, dimensionality reduction of Karhunen-Loeve Transform (KLT) is implemented
on it to formulate a compact representation.
364 Medical Imaging
3.3 Experiments
(a)Mitosis Identication: The proposed model-based microscopy image preconditioning and
sequence segmentation methods are implemented on the original image sequences to extract
mitotic candidate sequences. Then, an HCRF classier is learned with manually labeled
mitotic and non-mitotic candidate sequences by the previous step for mitosis identication. In
HCRF model, we can incorporate long range dependencies controlled by a sequence window
length wwhich denes the amount of past and future observations to be used when predicting
the state at time t (w = 0 indicates only the current observation is used). To select the best
window size w, we tuned this parameter and trained the corresponding HCRF classiers to
compare their performances with Area Under Curve (AUC) of ROCcurve. After obtaining the
best window size, we randomly selected one image sequence as training set and tested on the
other four image sequences. Four different outcomes of the mitosis identication algorithm
are examined: 1) true positive (TP): the identied sequence contains one mitosis event; 2)
false positive (FP): the identied sequence does not contain any mitosis event; 3) true negative
(TN): the discarded sequence does not contain any mitosis event; 4) false negative (FN): the
discarded sequence contains mitosis events. Then, Precision (the ratio of TP to TP+FP) and
recall (the ratio of TP to TP+FN), and the F1 score (a harmonic mean of precision and recall) are
used as quantitative metrics for measuring the accuracy of mitosis identication. To evaluate
the performance of the proposed method, it is quantitatively compared with two different
methods: 1) to show the superiority of HCRF for sequence classication, it is compare to CRF
model (Liang et al., 2007) which can not discover the latent structure of the entire sequence
for classication; 2) to validate the advantage of integrating temporal information, it is also
compared to a frame-by-frame classication approach using a support vector machine (SVM)
classier as we did previously (Liu et al., 2010b).
(b)Mitosis Localization: With the identied mitosis sequences, we manually annotate
the three states, before mitosis, during mitosis, and after mitosis, with label 1, 2, and 3
correspondingly. Then, a CRF classier is trained for mitosis localization. In CRF model,
we can also incorporate long range dependencies with a sequence window length w
which
denes the amount of past and future observations to be used when predicting the state
at time t. To select the best window size w
, we tuned this parameter and trained the

corresponding CRF classiers to compare their performances with Area Under Curve of ROC
curve. After getting the best windowsize, we selected the same training set for model learning
and tested on the other four image sequences. To evaluate the accuracy of mitosis localization,
we calculated the mean and standard deviation (SD) of the timing error of the start and end
of the state of during mitosis. The timing error is dened as the absolute difference of the
rst/last frame with label 2 by the proposed method against the manually-labeled ground
truth.
4. Experimental results
4.1 Mitosis identication
With preconditioning and volumetric region growing, we extracted candidate mitosis
sequences in each input sequence. This step achieved 100%recall with relatively lowprecision
around 42%. To improve precision, we trained HCRF models with GIST features to rene the
identication results. To optimize the performance of HCRF model, 4 HCRF classiers were
365
learned by varying window size from 0 to 3. From Fig.5, the optimal window size can be set
with 2 (AUC=0.929) by plotting the ROC curve of each model and comparing the AUC values.
0 0.2 0.4 0.6 0.8 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
ROC of HCRF(GIST)
False positive rate
T
r
u
e

p
o
s
i
t
i
v
e

r
a
t
e

WindowSize = 0 & AUC = 0.907
Fig. 5. ROC curves of HCRF models with different sequence window sizes from 0 to 3.
Thereafter, the HCRF classier was trained with GIST feature and window size 2 on the
training set. The ROC curves of the test results on the four test sequences are shown in Fig.6.
By choosing the threshold corresponding to the maximum F score, we identify the mitosis
events in each input sequence. Fig.7 shows the four kinds of samples decided by our method.
From Table 1, we can achieve an average precision of 0.83 and an average recall of o.92 with a
corresponding maximum F score of 0.88.
Sequence No. Precision (%) Recall (%) MaximumF (%)
1 85 95 90
2 83 90 87
3 79 94 86
4 84 90 87
Average 83 92 88
Table 1. Precision and recall of mitosis identication by HCRF
To demonstrate the superiority of HCRF for sequence classication, we compared its
performance to the CRF model trained with fully-labeled sequences. To utilize CRF for
366 Medical Imaging
0 0.2 0.4 0.6 0.8 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
HCRF:WindowSize=2;GIST
False positive rate
T
r
u
e

p
o
s
i
t
i
v
e

r
a
t
e

Seq1: AUC = 0.92
Seq2: AUC = 0.93
Seq3: AUC = 0.91
Seq4: AUC = 0.86
Fig. 6. ROC of HCRF with GIST feature and window size = 2
mitosis identication, it is rst applied to label the full sequence. Then, a candidate sequence
is classied as mitosis if the number of frames assigned with labels 2 (during mitosis) is
greater than a certain threshold. By varying the threshold, we plot the ROC curves to
evaluate the performance of CRF models and obtained the optimal AUC of 0.78. Moreover,
to show the advantage of integrating temporal information, we compare its performance to a
frame-by-frame classication approach using a support vector machine (SVM) classier. We
implemented a mitotic cell detector using SVM with a radial basis function (RBF) kernel. The
best parameters for the SVM models were selected by cross validation. The detector was
applied independently to each frame of a candidate sequence. Different from the training
strategy of CRF, we labeled the frames during mitosis as positive samples, and the other
frames as negative samples. A candidate sequence is classied as mitosis if the number of
frames assigned to be mitotic exceeds a certain threshold. By varying the threshold, we plot
the ROC curves to evaluate the performance of SVM models and obtained the optimal AUC
of 0.77.
To compare the overall classication performances of HCRF, CRF and SVMtrained with GIST
features, we utilize the balanced F score as a complementary metric to AUC. We separately
computed the AUC and the best achievable F score of each sequence with each classier. From
Table.2, the results indicate that the HCRF classier consistently outperformed both CRF and
367
Fig. 7. Mitosis identication results by HCRF.
SVM, with a best-case performance of 92% AUC and 90% maximum F score when precision
and recall are 85% and 95% respectively.
Sequence No.
AUC (%) Maximum F (%)
HCRF CRF SVM HCRF CRF SVM
1 92 78 77 90 84 80
2 93 73 71 87 83 77
3 91 73 70 86 84 81
4 86 62 55 87 75 80
Table 2. Comparison between HCRF, CRF and SVM
4.2 Mitosis localization
To localize the starting and ending points of mitosis events, we usedCRF to label each frame of
the identied mitosis candidates by HCRF. To optimize the performance of CRF model, 4 CRF
classiers were learned by varying window size from 0 to 3. From Fig.8, the results showed
that the model trained with GIST features and window size 1 outperformed the others with
the best AUC value of 0.933.
The CRF classier was trained with GIST feature and window size 1 on the training set. The
ROC curves of the test results on the four test sequences are shown in Fig.9. By choosing the
368 Medical Imaging
0 0.2 0.4 0.6 0.8 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
ROC of CRF(GIST)
False positive rate
T
r
u
e

p
o
s
i
t
i
v
e

r
a
t
e

WinodwSize = 0 & AUC = 0.916
Fig. 8. ROC curve of CRF models with different sequence window lengths from 0 to 3.
0 0.2 0.4 0.6 0.8 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
CRF:WindowSize=1;GIST
False positive rate
T
r
u
e

p
o
s
i
t
i
v
e

r
a
t
e

Seq1: AUC = 0.93
Seq2: AUC = 0.96
Seq3: AUC = 0.92
Seq4: AUC = 0.94
Fig. 9. ROC of CRF with GIST feature and window size = 1.
threshold corresponding to the maximum F score, we located the starting and ending points
in each input sequence as shown in Fig.10. The experimental result shows that the proposed
method can achieve an average localization error of 1.35 1.46 frames for the start of mitosis
369
Fig. 10. Mitosis localization results by CRF. (a) Both starting and ending points are labeled
correctly; (b-c) The starting point is labeled correctly while the ending point is labeled
wrongly; (d-e) The starting point is labeled wrongly while the ending point is labeled
correctly; (f-i) Four cases of locating both starting and ending points wrongly.
event and 1.68 1.73 frames for the end of mitosis event. From Fig.10, it is understandable
that the start point can usually be localized more precisely than the end point because the
appearances and shapes of mitotic cells always change drastically when two daughter cells
separate with each other physically.
5. Conclusion
We propose a hierarchical framework of mitosis event detection for cell populations imaged
with time-lapse phase contrast microscopy. The method consists of three stages: mitosis
candidate extraction, mitosis identication, and mitosis localization, which jointly optimize
recall and precision for mitosis identication and minimize the mean error and standard
deviation of mitosis location. The proposed detection method achieved two kinds of excellent
results in very challenging image sequences of multipolar-shaped C3H10T1/2 mesenchymal
stem cells. For mitosis identication, we can achieve an average precision of 0.83 and a recall
of o.92 with a corresponding maximum F score of 0.88. For mitosis location, the proposed
method can achieve an average localization error of 1.35 1.46 frames for the start of mitosis
370 Medical Imaging
event and 1.68 1.73 frames for the end of mitosis event. Moreover, the proposed method
does not depend on empirical parameters, ad hoc image processing, or cell tracking and
consequently can be straightforwardly adapted to different cell types.
6. Acknowledgement
We would like to thank Prof. Takeo Kanade, Dr. Phil G. Campbell, Dr. Lee E. Weiss,
Dr. Mei Chen, Dr. Eric D. Miller, Dai Fei Elmer Ker, Dr. Sung Eun Eom, Dr. Zhaozheng
Yin, Jiyan Pan, and Ryoma Bise for their helpful discussions. This work was supported
in part by the National Natural Science Foundation of China (61100124, 21106095), Tianjin
Research Program of Application Foundation and Advanced Technology (10JCYBJC25500),
China Postdoctoral Science Foundation (2011M500512), 2010/2011 Innovation Foundation of
Tianjin University.
7. References
Al-Kofahi, O., Radke, R., Goderie, S., Shen, Q., Temple, S. & Roysam, B. (2006). Automated
cell lineage construction: A rapid method to analyze clonal development established
with murine neural progenitor cells, Cell Cycle 5(3): 327335.
Avriel, M. (2003). Nonlinear Programming: Analysis and Methods, Dover Publishing.
Belongie, S., Malik, J. & Puzicha, J. (2002). Shape matching and object recognition using shape
contexts, IEEE Trans. Pattern Anal. Mach. Intell. pp. 509522.
Bise, R., Li, K., Eom, S. & Kanade, T. (2009). Reliably tracking partially overlapping neural
stem cells in dic microscopy image sequences, MICCAI Workshop on Optical Tissue
Image Analysis in Microscopy Histopathology and Endoscopy, Vol. 12, pp. 6777.
Dalal, N. &Triggs, B. (2005). Histograms of oriented gradients for human detection, Proc. 2005
IEEE Int. Conf. Computer Vision and Pattern Recognition (CVPR05), pp. 886893.
Debeir, O., Ham, P. V., Kiss, R. & Decaestecker, C. (2005). Tracking of migrating cells under
phase-contrast video microscopy with combined mean-shift processes, IEEE Trans.
Med. Imag. 24: 697711.
Donoho, D. L. (2004). For most large underdetermined systems of linear equations the
minimal l1-norm solution is also the sparsest solution, Comm. Pure Appl. Math
59: 18481853.
Eccles, B. A. & Klevecz, R. R. (1986). Automatic digital image analysis for identication
of mitotic cells in synchronous mammalian cell cultures, Anal. Quant. Cytol. Histol.
8: 138147.
Gallardo, G., Yang, F., Ianzini, F., Mackey, M. & Sonka, M. (2004). Mitotic cell recognition with
hidden Markov models, Proc. SPIE: Medical Imaging, Vol. 5367, pp. 661668.
Gunawardana, A., Mahajan, M., Acero, A. & Platt, J. C. (2005). Hidden conditional random
elds for phone classication, Proc. Int. Conf. Speech Communication and Technology,
pp. 11171120.
Lafferty, J. D., McCallum, A. &Pereira, F. C. N. (2001). Conditional randomelds: Probabilistic
models for segmenting and labeling sequence data., ICML01, pp. 282289.
Li, K. (2009). Large-Scale Stem Cell Population Tracking in Phase Contrast and DIC Microscopy
Image Sequences, PhD thesis, Carnegie Mellon University, 5000 Forbes Avenue,
Pittsburgh, PA.
371
Li, K. & Kanade, T. (2009). Nonnegative mixed-norm preconditioning for microscopy image
segmentation, Proc. Int. Conf. Information Processing in Med. Imaging, pp. 362373.
Li, K., Kanade, T., Chen, M., Miller, E. D.and Weiss, L. E. & Campbell, P. G. (2008). Computer
vision tracking of stemness, Proc. IEEE Int. Symp. Biomed. Imaging, pp. 847 850.
Li, K., Miller, E., Chen, M., Kanade, T., Weiss, L. & Campbell, P. (2008). Cell population
tracking and lineage construction with spatiotemporal context, Med. Image Anal.
12(5): 546566.
Liang, L., Zhou, X., Li, F., Wong, S., Huckins, J. & King, R. (2007). Mitosis cell identication
with conditional random elds, Proc. Life Sci. Sys. App. Workshop, pp. 912.
Liu, A., Li, K. & Kanade, T. (2010a). Mitosis sequence detection using hidden conditional
random elds, Proc. 7th IEEE International Symposium on Biomedical Imaging: From
Nano to Macro, pp. 580583.
Liu, A., Li, K. & Kanade, T. (2010b). Spatiotemporal mitosis event detection in time-lapse
phase contrast microscopy image sequences, 2010 IEEE International Conference on
Multimedia and Expo (ICME10), pp. 161166.
Oliva, A. & Torralba, A. (2001). Modeling the shape of the scene: A holistic representation of
the spatial envelope, Int. J. Comput. Vision. pp. 145175.
Otsu, N. (1979). A threshold selection method from gray level histograms, IEEE Trans. Syst.,
Man, Cybern. 9: 6266.
Quattoni, A., Wang, S., Morency, L.-P., Collins, M. & Darrell, T. (2007). Hidden conditional
random elds, IEEE Trans. Pat. Anal. Mach. Intel. 29(10): 18481852.
Rabiner, L. R. (1989). A tutorial on hidden markov models and selected applications in speech
recognition, Proceedings of the IEEE, Vol. 77, pp. 257286.
Rosin, P. L. (2001). Unimodal thresholding, Patt. Recog. 34(11): 20832096.
Silvela, J. & Portillo, J. (2001). Breadth-rst search and its application to image processing
problems, IEEE. T. Image. Process. 10(8): 11941199.
Tikhonov, A.-I. N. (1977). Solutions of Ill Posed Problems (Scripta Series in Mathematics), Vh
Winston.
Wang, S. B., Quattoni, A., Demirdjian, M. D. & Darrell, T. (2006). Hidden conditional random
elds for gesture recognition, Proc. 2006 IEEE Int. Conf. Computer Vision and Pattern
Recognition, pp. 15211527.
Yin, Z., Li, K., Kanade, T. & Chen, M. (2010). Understanding the optics to aid microscopy
image segmentation., Proc. International Conference on Medical Image Computing and
Computer Assisted Intervention (MICCAI10), pp. 209217.
372 Medical Imaging
0
Safety of Interactive Image-Guided Surgery
Alain Beaulieu
Royal Military College of Canada
Canada
1. Introduction
In this chapter, we discuss Interactive Image-Guided Surgery (IIGS) and present the current
state of the art in the analysis and evaluation of this kind of computer-supported medical
procedure. We also present the results of our research to evaluate the safety of IIGS. During
our discussion of the current state of the art, we review the various approaches that have been
proposed to analyze and evaluate IIGS systems including their strengths and weaknesses.
The original hypothesis that initiated our research was that: it is possible to improve the
safety analysis of computer-guided surgery if we model this kind of system as a process
control model." The hypothesis was rst elaborated in an initial research planning meeting
while we were investigating a better way of developing a Verication and Validation method
for IIGS. During the evolution of this research, this hypothesis has evolved signicantly from
a basic idea into a modelling language and a protocol that have been evaluated using a
particular set of IIGS systems used for orthopedic surgery.
2. Using medical imaging for surgery
For many decades, medical imaging has been rapidly evolving as one of the most useful
applications of scientic and engineering knowledge to improve the quality of health care.
Medical imaging provides technology that enables physicians to see subcutaneous structures
and tissues without the use of more invasive procedures
1
.
The use of medical imaging increases the observability of areas of interest inside the patient.
Increased observability allows for more complete diagnosis and better planning of surgical
procedures leading to potential improvements in surgical outcome. The rst use of medical
imaging with clinical intent can be traced back to January 1896 in Birmingham, United
Kingdom where an x-ray was used to plan the removal of a needle from a womans hand
Viergever (1998). Medical imaging for clinical procedures has therefore been around for over
100 years. The rst use of medical images was static, but it demonstrated the usefulness
of medical imaging in planning surgeries. Medical imaging progressed over the next 100
years with improvements in the quality of existing modalities and image processing, as
1
The debate on the degree of invasiveness of ionizing and non-ionizing radiation is ongoing. Although
it is clear that their side effects cannot be overlooked Assessment and Management of Cancer Risks
from Radiological and Chemical Hazards (1998); on Radiological Protection (1960-Present); Proo (1979);
Tenforde (1979), medical imaging is far less invasive and risky than open exploratory surgery Hindus
(2001).
18
well as the addition of new imaging modalities. In todays medical imaging, there are
numerous modalities that can be used in complementary fashion to improve diagnosis.
Some of these modalities include: X-Ray, CAT
2
, Magnetic Resonance Imagery (MRI), Single
Photon Emission Computed Tomography (SPECT)
3
, Positron Emission Tomography (PET),
ultrasound and several other specialized forms of imaging Msges & Lavalle (1996).
In recent years, it has been possible to acquire intraoperative volumetric data with the use
of mobile medical imaging (CT, MRI and ultrasound). Mobility of 3-Dimensional medical
imaging in the operating room has allowed surgeons to track surgical instruments through
tissues in real-time during the surgery, leading to less invasive procedures or Minimally
Invasive Surgery (MIS) Patel et al. (1998); van der Weide et al. (1998); Viergever (1998).
The benets of MIS are signicant. Although no single reference could be found that provided
a comprehensive list of the potential benets of MIS, several authors refer to such benets
to highlight the relevance of their work Choi et al. (2000); Gary et al. (2006); Hindus (2001);
Msges & Lavalle (1996); Patel et al. (1998); Peters (2001); Pieper et al. (1994); Robb (1996);
van der Weide et al. (1998); Viergever (1998); Zamorano et al. (1994).
During the last decade, increased ability to process computer graphics and images in real-time
has made possible the intraoperative use of 3-Dcomputer models obtained fromsegmentation
and volume rendering of preoperative imaging from CAT, MRI and ultrasound Viergever
(1998). Without volumetric data, the surgeon would have to form his/her own mental model
from imaging slices.
3. Interactive image-guided surgery
3.1 General
This section covers the general principles of orthopedic IIGS. We present the processes and
methods involved in IIGS; the intent is to provide an overviewand to illustrate the complexity
of such systems. IIGS can be represented as a pipeline of operations which can be separated
into six distinct phases as displayed in Figure 1. The parts of the process are dened as follows:
1. Image Capture - Preoperative images of the region of interest in the patient are acquired.
The images of interest are represented by a set of CAT slices or a scene
4
.
2. Visualization - Various complex operations are applied to images to extract information.
Visualization has also been called rendering or more loosely refered to as image processing.
3. Surgery Planning - Preoperative planning of the surgery is conducted using the scenes from
visualization that have been generated by the IIGS system. The planning phase can also
include rehearsal of operations on some guidance systems. Training can also be included
as a requirement.
4. Registration - In the operating theatre, the visualized images are registered to the patient to
align the reference systems.
2
Also known as Computed Axial Tomography or CT
3
AKA tomoscintigraph
4
A scene is a constructed 3-D image from a region of interest in the patient.
374 Medical Imaging
Safety of Interactive Image-Guided Surgery 3
5. Intraoperative Guidance - During surgery, the position of surgical instruments is tracked and
displayed superimposed on the preoperative visualized images, allowing the surgeon to
accurately navigate through the area of interest in the patient.
6. Postoperative Followup - After surgery, the surgeon may want to conrm surgical outcome
using the surgery planning information.
Image
Capture
Visualization
Surgery
Planning
Registration
Intraoperative
Guidance
Postoperative
Followup
- - -
- - - -
Fig. 1. The IIGS Pipeline
The above phases have also been presented in the literature as models containing three or
four phases Breeuwer et al. (1998); Gauldie (2002). A three phase model could be composed
of preoperative (phases 1-3 above), intraoperative (phases 4-5) and postoperative (phase 6).
Regardless of the separation of phases or actual requirement of the system, the system follows
this linear progression of operations. Below, we address the six phases of IIGS. Planning and
post operative followup phases can be represented as single processes and depend in large
measure on information generated during the other four phases. These two phases are highly
dependent on competent human intervention derived from medical knowledge.
3.2 Image capture
The imaging modality depends on the procedure to be performed and the target area to
be imaged. Apart from the physics of the modality, it is important to note that for a
given modality, various health facilities will have different machines each with its own pixel
resolution, slice thickness and storage format. The increased need to communicate medical
images over networks and between machines has pushed the adoption of medical imaging
standards such as Digital Imaging and Communication in Medicine-3 (DICOM-3). This
standard only supports sets of 2-D slices and not volume models Starreveld et al. (2001).
Further processing of the images as described in the following section depend on the ability of
the algorithms to adapt to the parameters of various CAT scan machines and format of images
that come from other systems.
Image capture can be preceded by the surgical implantation of ducial markers if ducial
registration is to be performed. Physical markers such as these are made visible on medical
images by selecting appropriate material for the markers.
3.3 Visualization
Prior to intraoperative mobile 3-D medical imaging and IIGS, the surgeon had a series of 2-D
slices that he/she used to form a mental 3-D model. Visualization (sometimes referred to as
375 Safety of Interactive Image-Guided Surgery
rendering
5
) deals with the display of 3-D information on a 2-D matrix of pixels (computer
screen). Using computer graphics methods and todays technology, this 3-D visualization
can be built from the 2-D slices obtained from medical imaging modalities. Constructing a
3-D model is a complex process that is best viewed as a pipe-and-lter architecture for image
transformations. This view is useful since the selection and ordering of various operations
will determine the characteristics and quality of the visualization.
Several image processing operations can be applied to medical images including: ltering,
segmentation, classication, interpolation, quantication and image rendering. There is a
great number of algorithms and techniques described in the literature for each of these
operations. Most techniques require interactive or manual assistance from an expert in 3-D
image processing, but recently, automated algorithms are starting to appear in the literature
Radau et al. (2000); Udupi et al. (2001); Zoroo et al. (2001). One of the most helpful references
in the domain of visualization is a survey paper on 3-D rendering in medicine by Jayaram
K. Udupa Udupa (2001). In trying to create a concise survey of the operations used in
visualization, we found that there is no consensus on terminology or clean interfaces between
various operations. For example ltering and classication are often included as part of
segmentation. We consider each operation as a step in visualization. The operations are
presented in the following subsections.
Images on computer screens are represented by a set of pixels. Likewise, medical images
produced from modalities like CAT are formed from matrices of pixels of various intensities
identifying different tissues. Pixels and voxels are sometimes used synonymously in the
literature. In 3-D visualization, it makes more sense to talk about voxels, the 3-D equivalent of
a 2-D pixel Hindus (2001); Udupa (2001). This mixup in terminology can be explained when
we consider that a 2-D slice has a thickness that is determined by the beam width of the CAT
machine.
3.3.1 Filtering
Filtering is an operation that takes a scene and changes the intensity
6
of its voxels to reduce
noise, distortion and other artifacts that have not been eliminated by the imaging machine.
Filtering smoothes the edges of structures to provide clearer and crisper images. Filtering can
either be linear or optimal Ayache et al. (1996).
There are two approaches to linear ltering: gradient or Laplacian. Both approaches
assume a white Gaussian noise. If the noise does not have zero mean or there are a lot of
irregularities in structures, linear ltering may require large convolution masks as well as
complex algorithms. Because medical images are very large, the computational requirements
of complex algorithms are very high.
Another less computation intensive approach for noisy images is optimal ltering. Canny,
Deriche and Shen have proposed such lters Ayache et al. (1996).
5
Visualization and rendering are often used interchangeably in the literature, but they are not
synonymous. Visualization represents the entire process of image transforms where rendering is its
last step of displaying the image.
6
Depending on the modality of imaging, and the author, intensity is sometimes replaced by density. This
is especially true for scenes that have vector-valued voxel intensity such as obtained with MRI.
376 Medical Imaging
3.3.2 Segmentation
Segmentation is crucial for the identication and extraction of information and structures from
a given scene. For orthopaedic surgery the structures consist of bones, ligaments, cartilage and
tendons. The ability to accurately locate structures in the body for planning and intraoperative
tracking is directly linked to surgical outcome Ayache et al. (1996); Dean et al. (1998); Udupa
(2001).
There are two main classes of approaches for segmentation: region-based segmentation which
includes a set of voxels describing a sub-scene and boundary-based segmentation that outputs
a set of voxels or geometric information from voxels (polygons) about the boundary of the
object Udupa (2001).
Another way to characterize segmentation algorithms is in the way set membership is decided
for voxels. In a hard segmentation, also called binary segmentation, set membership is either
0 or 1. Gray or fuzzy segmentation uses voxel intensity to decide the degree of belonging to
the set with real values ranging between 0 and 1. Some methods, such as thresholding and
clustering, use either binary or gray segmentation Udupa & Gonalves (1996).
Region-based methods can be divided in two different groups: clustering and region-growing.
Clustering uses voxel features to dene each set in a scene. The simplest clustering method
is thresholding. Region-growing, as the name indicates, grows sets of voxels based on
homogeneity properties. These methods grow regions of interest from seed points by
considering neighboring voxels for inclusion.
There is a great number of boundary-based (edge detection) methods described in the
literature. Only a few are mentioned here. Filtering is often used as a rst step in the process
of edge detection. The following are examples of boundary based methods:
1. Simple thresholding and hysteresis-thresholding can be used to nd contours. The
problem with these simple methods is that there are many discontinuities between edges
in the contour Ayache et al. (1996). These contours can be closed by domain experts who
have knowledge of anatomy.
2. Active contours also called snakes, have been successfully used to solve the 2-D contour
problem. In order to use these algorithms in a 3-D environment, one must nd a 3-D
generalization of the energy dening the border of an object. The user denes an initial
contour and the algorithm optimizes this contour based on a balance between inside and
outside energies.
3. Isosurface methods track the surfaces of structures through the assumption that the
boundary of an object in a scene has a constant density. One such algorithm is Lorensens
marching cubes Lorensen & Cline (1987), which has been adapted for one of Queens
University IIGS systems. The marching cubes method constructs the surface of each
structure by nding, for each voxel, the local direction of an isosurface through it.
From a validation perspective, it is important to note that not only are the results from
boundary-based methods highly dependent on the algorithm but also on the skills of the user.
3.3.3 Classication
Classication is basically a problem of creating an opacity scene from a density scene Udupa
& Gonalves (1996). Classication deals with connecting like-pixels with the use of a
luminescence histogram Ayache et al. (1996). Objects are dened by their belonging to a
range of luminescence. Classication and segmentation are two different operations but gray
segmentation is synonymous with classication Udupa & Gonalves (1996).
3.3.4 Interpolation
In order to facilitate the processing of information and to complete the 3-Drepresentation from
slices, it may be necessary to combine or split voxels through interpolation. The preferred
shape for a voxel in most algorithms is a cube. Because slice spacing (z-dimension) is 2-8 times
greater than the voxel size in the x or y dimension, we want to ll the gap with ctitious voxels
based on interpolated density information fromneighboring voxels. If the slices overlap, some
voxels may need to be split, again requiring interpolation. The result will provide an isotropic
3-D representation.
3.3.5 Rendering
Rendering is the last process in the visualization pipeline. As such the image that is rendered
depends on several operations, as well as the sequencing of these operations. There are two
kinds of rendering: surface and volume.
Surface rendering for which density has either value [0,1] portrays the boundary surface
of a structure. Surface rendering can be displayed as a set of polygons obtained from
segmentation. Each polygon has its own intensity level that is calculated based on the
orientation of the surface with respect to the screen. This gives the impression of depth to
the structures. Surface rendering can also be portrayed through volume rendering techniques
but the opacity of the surface voxels will hide the elements behind them.
Volume rendering for which density has a real value in the interval [0..1] portrays a translucent
structure through two distinct operations: projection and pixel intensity calculations. The
projection can be done with either direct projection of object polygons or with ray casting.
Pixel intensity calculations can be done in several ways, depending on the determination of
normal vectors to the object surface. The quality of the rendition is highly dependant on the
estimation of the normal vectors Udupa & Gonalves (1996).
3.3.6 Safety implications of visualization
As we can observe by the number of transformations described above, visualization, can
be the source of some signicant accuracy errors in IIGS. Each of the transformations is a
single point of failure and has cumulative effects on the next set of transformations. The
results from visualization are fed to the subsequent phases. Hazards created by failures in the
transformations will propagate to these phases.
As we have found during our research, tests with phantoms and computer code inspections
of algorithms represent the two main mitigation strategies.
378 Medical Imaging
3.4 Surgery planning
As indicated above, surgery planning can be reduced to a single process that is highly
dependent on information provided by the previous phases of IIGS. During this phase, the
surgeon uses a part of the IIGS system to perform a simulated surgery where s/he can make
changes to the virtual or visualized skeletal structure of the patient. The planning tool can
provide the surgeon with a variety of options to visualize corrections. Depending on the tool,
the surgeon can change the amount of correction applied to the bone structure, choose and
change prosthesis size, view post simulated surgery results. More complex planning tools
may allow the surgeon to obtain technical measurements and even view dynamic models
after a correction.
3.5 Registration
Registration is the process of aligning coordinate systems so that we can match geometric
data. Registration is also called co-registration, matching and fusion Taylor et al. (1996).
Registration is done between coordinate systems of various modalities such as the registration
of CAT and MRI 3-D models, and/or between imaging data and patient. Registration is
critical to IIGS because it is directly linked to the execution of a preoperative plan through
intraoperative navigation.
There are several approaches to registration. We describe three of them here, namely: ducial
markers, anatomical landmarks and surface based methods. Other methods are described in
Gauldie (2002); Lavall (1996). In the selection of a registration technique, there is a tradeoff
between accuracy and invasiveness.
Fiducial markers are small posts that are surgically secured to bone tissue prior to imaging.
Markers are chosen to be visible for the particular medical imaging modality. Immediately
prior to surgery, registration is performed by touching each marker with a tracked probe. This
procedure is the gold standard for registration because it minimizes errors in transformation.
This procedure is, however more invasive than other methods Birkfellner et al. (1998);
Starreveld et al. (2001).
Anatomical landmarks are an option for registration when it is desired to eliminate the extra
surgical procedure needed by ducial methods. The problem is to nd a set of distinctive and
unambiguous points that can be registered reliably. Depending on the procedure, the surgeon
may only expose surfaces that are devoid of such features rendering this method difcult to
use.
Surface-based registration is accomplished by acquiring points from the surface to be
registered. These points form a description of the surface that must be matched to the 3-D
model. The most prevalent method for acquiring these points is by touching the exposed
surface of the bone at several locations with a tracked probe. The problem with this type
of point collection is that the surgeon will expose as little bone as possible. Recently,
A-Mode ultrasound has been used to perform transcutaneous surface registration at Queens
University Gauldie (2002). A-Mode ultrasound has also been used to perform anatomy-based
registration for the spine Lavall et al. (1996), but the point gathering method and the
matching technique used by the authors resemble that of surface-based registration.
After the points are matched, the registration algorithm tries to nd the best transformation
that will best align the coordinate systems. For bones, a rigid-body transformation (
F (x))
consisting of a 3 by 1 translation vector
T and a 3 by 3 rotation matrix
R is sufcient because
the deformations of images are negligible Lavall (1996). The transformation therefore has the
form:
F (x) =
R x +
T (1)
Registration must be accurate in order to maximize surgical outcome. From the many papers
on the topic, it is generally accepted that registration accuracy of less than 1 mm and 0.5
degrees is surgically acceptable. No paper could be found in the literature that provided
empirical justication for the acceptability of these limits on surgical accuracy. Some papers
also talk about sub-pixel accuracy for localization. This provides a vague requirement, since
pixels vary in size and geometry depending on the imaging machine and the settings.
3.6 Intraoperative guidance
After the scenes are registered to the patient, it is possible to display digitized models of the
tools as part of the visualization, to provide these scenes in real-time and provide guidance
information to the surgeon. Other technical data such as numerical values of location, angles
and forces can also be displayed. As an example, during a High Tibial Osteotomy (HTO)
which involves the removal of a wedge of bone tissue, the position and angle of a drill used
to insert guide wires into the leg can be tracked and errors displayed on the preoperative
plan in real-time. Force monitoring of the drill can be ascertained by measuring the current to
determine torque Msges & Lavalle (1996).
The accurate display of surgical instrument models depends on the tracking system used.
There are two basic categories of instrument tracking: articulated arms and triangulated
emitters Galloway et al. (1994).
3.6.1 Articulated arms
Articulated arms can resolve positions though the use of instrumented joints (potentiometers,
optical encoders, or resolvers) and simple kinematics. The surgeon manipulates a surgical
instrument that is attached to the end of the arm. The arm can also have other useful
characteristics such as limiters that are set to respect the preoperative plan or brakes that keep
the position of the instrument xed for long periods.
Articulated arms have good accuracy but this accuracy is limited by uncertainties related to
deformations, backlashes, alignment errors and the like. Articulated arm signals cannot be
obscured. However, arms have other limitations: accuracy is inversely proportional to the size
of the operating envelope, only one instrument can be attached to a single arm, and sterility
is a problem Galloway et al. (1994).
3.6.2 Triangulated emitters
Three types of triangulated emitters are discussed in the literature, each having multiple
implementations.
380 Medical Imaging
Ultrasound time-of-ight was one of the rst methods used to triangulate instruments. This
technology is simple and cheap to implement. A piezoelectric emitter on the instrument is
synchronized with several receivers whose locations are known precisely. Time-of-ight of an
ultrasound pulse provides distances between each receiver and the transmitter. Intersection of
the spheres in space at the known radii provides the location of the instrument. The two main
problems with ultrasound are that the speed of sound changes depending on the temperature
and the signals can be obscured by the presence of surgical staff.
Electromagnetic systems can also be used. Three emitters generate magnetic elds in
orthogonal planes and the signals are captured by receivers. The main advantages of this
technology are that the signals do not suffer from line-of-sight obscuration and they are not
sensitive to temperature changes. It is also possible to attach such a device to an endoscope
and locate the end position of these instruments inside the body. The main problems with
this technology are the susceptibility to metallic objects in the operating theater as well as
electromagnetic interference from other electrical equipment Galloway et al. (1994).
Passive and active optical tracking can also be used for triangulation. Queens University
uses an active system composed of Charge-Coupled Detector (CCD) cameras in a linear
arrangement and a series of Infrared Light Emitting Diode (IRED) mounted on surgical
instruments. The IREDs on each instruments are mounted at specic locations and are
sequentially pulsed at time-specic intervals. The position of an IRED can be located in less
than 300 microseconds. This technology does however suffer from line-of-sight obscuration
which can be limited by careful placement of the cameras.
In terms of accuracy, the active optical tracking systemis the most accurate, capable of locating
instruments within +/- 0.2 mm in a volume of (500mm)
3
Galloway et al. (1994); Msges &
Lavalle (1996).
3.7 Postoperative followup
After the surgery the patient will be evaluated to determine surgical outcome and general
well-being. Some of the postoperative assessments will be done without the use of the IIGS
systems, using physician evaluations and/or patient questionnaires. Some evaluations will
make use of a subset of the IIGS system such as the Medical Imaging and Visualization to
compare preoperative and postoperative results. It can also be useful to compare the plan to
the actual result. Such postoperative followups should be undertaken with care considering
the additional radiation that the patient has to be subjected to.
3.8 Failures in IIGS systems
It is important to identify what is meant by a failure within the scope of Integrated
Image-Guided Surgery. Apart fromthe obvious hardware and software failures that can cause
a system to crash or malfunction, we must consider errors that may affect surgical outcome.
A concept that is advanced by Schneider and Hines, is that of patient vulnerability from data
7
provided by medical software Schneider & Hines (1990). In their paper Schneider and Hines
dene patient safety as freedom from harm by a medical device", while patient vulnerability
7
Data as discussed here includes medical imaging and models derived from volume rendering,
registration and tracking information.
is dened as potential harm due to erroneous system output". During orthopaedic surgery
involving IIGS, the decisions made by the surgeon depend in large measure on the IIGS
system
8
. Vulnerabilities in IIGS therefore come from the data that is provided in the form
of rendered images, real-time display of surgical instruments within the model and numerical
information. As indicated by Fitzpatrick, the knowledge of accuracy is as important as
the accuracy itself Fitzpatrick et al. (1998) both of which are crucial to managing patient
vulnerability. Since geometric accuracy is linked with surgical outcome and the skeletal
structure of the body is modied from this information, any errors in accuracy that can have
an impact on patient vulnerability are potential hazards. Failures must be identied based
on the severity and likelihood of hazards. Another problem of introducing complex medical
devices is that the surgeon does not know the level of condence he can have Fitzpatrick et al.
(1998).
3.9 IIGS conclusion
Interactive Image-Guided Surgery includes a complex set of process and interfaces which
have traditionally been represented as a pipeline of operations. The image processing portion
of the system can be represented with a pipe-and-lter software architecture. We argue that
although this representation is excellent to illustrate an introductory discussion to IIGS, it is
not sufcient to perform a safety analysis on IIGS. Throughout the discussion of the various
phases of IIGS, there are many interfaces, equipment and interactions that are not visible in the
pipeline architecture of the system. For example, the quality of the visualization depends on
the selection of the operations applied to the medical images, the algorithmused to implement
each operation, the ordering of the operations in the pipe-and-lter architecture and the skills
of the operator performing the image transformations.
The potential sources of errors from devices, processes and interfaces vary widely in range,
which can affect the quality of visualization, planning, registration, guidance and ultimately
the surgical outcome. As well, we must identify unwanted disturbances and noise that
interfere with the ideal operation of the IIGS system. In order to increase the completeness
of the safety analysis, we must rst identify and expose all system interfaces.
4. State of the art in the analysis and evaluation of IIGS systems
Before we describe our approach to safety analysis of IIGS systems, we present work that has
been done in the area analysis and evaluation of these systems. The intent is to show through
various approaches taken, the complexity of studying these systems froma safety perspective.
Little in the published research is directly relevant to the topic of analysis, evaluation or
validation and verication of IIGS systems. The literature, however, contains many papers
on spatial delity and accuracy of visualization, segmentation and registration algorithms.
The words verication and validation are often used by these papers, but they mainly refer to
mathematical and procedural accuracy verication. Accuracy and spatial delity are crucial
to IIGS systems and it is generally accepted that they can signicantly affect surgical outcome.
Although necessary to the conduct of safe IIGS procedures, accuracy and spatial delity
are, however, not synonymous with system safety. Saying that software is correct and the
8
The surgeon also has his/her experience and coarse visual and tactile feedback to rely on.
382 Medical Imaging
calculations are accurate says nothing about possible omissions of safeguard requirements
that could have made the system safer Gowen & Yap (1993).
Several papers have also been published on validation of user interfaces and clinical trials but
none of the proposed methods or protocols are applicable or extendable to general clinical
practice. We dene general clinical practice as a hospital with no integral research and
development department that would have knowledge of the internal design of the systems
and could be called upon to operate IIGS technology.
In the following subsections we present several methods that have been used to perform
some form of analysis, evaluation or validation activities for IIGS systems. At the end of each
section, an evaluation of the scope, focus and aim of each method is given. The scope refers to
the phase(s) that the method concentrates on: preoperative, intraoperative or postoperative.
The focus localizes the area of interest or where the method is applied and can be one or more
of: software, hardware, system, interfaces, patient, or surgeon. The aim identies the goal of
the method; the quality factor that is sought after: safety, reliability, usability, or effectiveness.
4.1 Clinical relevance
Two papers that closely address the validation of IIGS systems are from Verbeeck et al. (1995)
and from Breeuwer et al. (1998). Both papers approach the validation of IIGS systems from
the clinical evaluation perspective. Also, both papers recognize the necessity for verication of
accuracy using phantoms. Breeuwer et al. also look at CAT and MRI image distortion and the
impact of registration errors as part of their validation efforts.
In their paper, Verbeeck et al. establish some strong arguments for setting up a Clinical
Relevance validation that in effect justies or rejects the requirement for the system. The basic
idea of Clinical Relevance validation is to use a set of questionnaires that ask the surgeon to
evaluate the system based on a number of criteria which are rated on a three point scale. In
their paper, they compare the results of an existing manual
9
method of performing stereotactic
brain surgery and a new system for planning and guidance. The evaluation of the new IIGS
system is done in two distinct phases, each phase including its own questionnaires.
The Functional Specication Phase uses prototyping and a functional specication
questionnaire containing twelve questions to rene the system. Each question in this
questionnaire has a weight factor, assigned by the surgeon, that depends on the patients need
for the new system, but does not depend on the technology. A second questionnaire that does
not involve the patient and deals specically with the surgeons interaction with the new IIGS
system contains ten questions without weight factors. This questionnaire deals mainly with
the usability, usefulness and robustness of the new system from the surgeons perspective.
The Clinical Acceptability Phase includes two questionnaires. The rst questionnaire contains
seven questions and compares the old and new IIGS systems as well as their dened medical
protocols
10
. The questions deal with the clinical information that can be obtained from
each system. The questions also evaluate the degree by which any additional information
requirement from the new system outweighs the necessary activities to get this information.
9
The method is said to be manual even though a computer is used to assist in the static planning phase.
10
Because the two systems, manual and IIGS, differ in scope and information requirements, each must
have its own medical protocol for use by the surgical team.
Because the protocols differ and the information requirements are not the same for both
systems, a not available" score can be assigned to a question. Because the information needs
within a protocol may vary in importance, weight factors are assigned to each question by
the surgeon. The second questionnaire for this phase contains six questions and addresses the
benecial aspects of the new technology for the patient. Here the surgeon answers questions
that deal with overall benets, risk, comfort, time, and perceived accuracy of the systems.
Breeuwer et al. also use a questionnaire for assessment by the surgeons. The eight questions
posed to the surgeon are on a visual analog scale scored from -5 to +5. This is similar to a
Likert scale Likert (1932).
Both validations ask a question about the perceived safety from the surgeons point of view
following a procedure. Breeuwer also asks a question about the effect upon the condence of
the surgeon during the procedure.
The questionnaire method used by both research groups is an efcient instrument to evaluate
a quality factor such as Clinical Relevance or to measure the condence of the surgeon in the
new technology. This, however, is not the appropriate kind of instrument to use to validate or
verify system safety. Another weakness of questionnaires is that they explore known issues
at the time the questionnaire was developed; safety critical systems require instruments that
bring to light issues we did not think about a priori. Also because of the small samples used
in both studies, the issue of statistical signicance is questionable. Finally, as seen from both
papers, the questionnaires are quite short and contain no validating questions to ensure that
answers are consistent.
Clinical relevance and effectiveness remain important aspects of introducing new technology
in the operating theater and should be evaluated as part of a validation protocol. On the
other hand, the use of system designers to operate the new IIGS system in the operating
rooms is not the norm in clinical practice. System designers have a complete understanding
of the underlying technology of the IIGS system compared to general clinical technical staff
so the results obtained from clinical evaluation in a research setting may not be transferable
to a general clinical setting. Clinical relevance covers the intraoperative phase and is surgeon
centric. The study aims at relevance froma usability analysis perspective. Safety is mentioned
but as a single indivisible global quality factor.
4.2 Human error analysis
A paper that describes another aspect of the state of the art in the analysis of IIGS systems is
by Jiang et al. (1998). This paper is most closely related to our research. In their research, the
authors perform a tailored Failure Mode Effect Analysis (FMEA) technique to study human
errors in IIGS. Their research takes an approach similar to that of conventional safety analysis
of safety critical systems Chudleigh et al. (1995); Fries et al. (1996); Gowen (1994); Herrmann
(1995); Jones et al. (2002); Leveson (1995); Loesh et al. (1999); Mojdehbakhsh et al. (1994);
Murthy (1995) but explicitly examines only the software modules.
While the technique used by Jiang et al. has a lot of merit, their approach also has signicant
limitations. The aim of their research is to identify surgical errors by focussing on human
factors and human errors that can occur in using software. Their analysis is limited to software
component failure modes and human errors that could cause each software component to fail.
The authors concentrate on errors caused by the operation of the software not on systematic
384 Medical Imaging
faults. Because the computer human interactions are part of the study, errors in interface
design may be uncovered by this approach. In order to study this human interaction errors
from design the study should include the design of the human machine interfaces.
There are several reasons why FMEA alone is not sufcient for this kind of analysis:
1. The underlying assumption made by the authors in only using FMEA, is that there are
only single points of failure in the system. FMEA is a stovepipe analysis that does not look
at module interfaces or error propagation between modules. For example, if two human
errors are made, one during segmentation and one during registration, the effect of one of
the errors may not be signicant enough to activate a protection measure, but the accrued
errors could be surgically signicant and not be discovered.
2. Because the analysis only studies the failures of the software modules that are in the current
system architecture, and does not look at global system safety, there is a high probability
of missing a safety requirement that could be solved by adding a new module or function
such as a watchdog or built-in test.
3. Errors caused by hardware malfunctions may be overlooked if not anticipated as part of
the failure modes of the software modules.
4. Because the technique used by the authors focuses on mistakes and failures, they do not
analyze the effects of the new medical protocol as part of the safety analysis. There is
no mention of modifying surgical gestures, system interfaces, human-machine interfaces,
noise, or disturbances. This is important because some procedures change with the
introduction of IIGS
11
.
5. System failures that are not caused by software failures or human errors may be
overlooked by this approach. For example the effects of motion artifacts, deformation
of tissue due to changes in patient position between the imaging and the operating room,
and system latency during tracking, may be overlooked.
The analysis would be more complete if complemented with other safety analysis techniques.
The authors however, look at methods of prevention and protection against software failures
and human errors by applying the concepts of FMEA to software. It is the only work
of its kind that could be found in the open literature for the analysis of IIGS systems.
Human-error analysis is software centric and includes preoperative and intraoperative
phases, an improvement on other methods presented. The method aims at increasing safety
by reducing failures induced by interaction with the software.
4.3 Component-based trusted architecture pattern
A recent paper suggests a drastically different approach to ensure safety of IIGS. The
researchers claim that by using encapsulated Moore state machines and a trusted architectural
pattern of components" they can guarantee safety and reliability in the system Gary et al. (2006).
They base this claim on the hypothesis that state machines ensure that component behavior is
deterministic and that all components are in a known and error-free state at any given moment."
11
For example for wrist alignment surgery the old protocol was to cut, distract, drill holes; the IIGS
procedure changes this order to drill holes, cut and distract.
The paper also claims that the approach can predictably integrate third-party software by
bounding these untrusted components behavior using state machine wrappers.
The approach gives little consideration to the fact that the IIGS software interacts with a
physical environment. The techniques they suggest do not consider safety hazards that
are not related to the software and for which software design could provide mitigation
and/or removal. This does not follow sound safety design principles elaborated in countless
references Chudleigh et al. (1995); Elliott et al. (1994); Fei et al. (2001); Gowen (1995); Gowen
& Collofello (1994); Halang et al. (1998); Jiang et al. (1998); Jones et al. (2002); Knight (1990);
Lefngwell & Norman (1993); Leveson (1995). Leveson states that ...the quality of a safety
program is measured by its ability to inuence design" Leveson (1995); this ability comes in the
form of safeguard requirements that comes from a safety analysis. These safeguards would
be missed if we only use the trusted architecture paradigm as proposed by the authors.
By their own admission, communication with subject matter experts such as surgeons, to
conrm requirements is difcult. In their research, the authors used Unied Modeling
Language (UML) activity diagrams to show the process of state transitions and the safe
behavior of the software.
The approach is entirely software centric and specically addresses only the intraoperative
phase of IIGS. The Component-Based Trusted Architecture Pattern is a sound design approach
that may produce reliable
12
software, not safe systems.
4.4 Conclusion on state-of-the-art
The analysis and evaluation of IIGS systems discussed above focuses on narrow concerns and
isolated parts of larger systems. This aspect of IIGS research is indeed in its infancy. There are
several reasons for this lack of maturity. The complexity of the systems makes verication and
validation activities difcult to identify and perform. Due to the nature of these interactive
systems, with person in the loop, validation efforts are complex and difcult to express. Our
research aims to increase the expressiveness of models in the conduct of safety analysis for
these systems. By increasing expressiveness of the models we will better communicate with
domain experts. Our aimis also to develop a systemcentric focus including pre, intra and post
operative phases in our modeling and safety analysis. This allows to have a more complete
analysis at the system level.
5. Abstract process safety analysis model
In this section, we present our protocol and methods used to perform safety analysis on
IIGS systems. At the core of the protocol is our diagramming technique which is part of the
Abstract Process Safety Analysis Model (APSAM).
In traditional digital control theory, the controllers (computer control systems) are generally
composed of ve classes of components Dunn (2003). These components, illustrated in
Figure 2 are:
1. The application (process or plant), the physical entity that the control system monitors and
controls.
12
Software is said to be reliable if it behaves according to its specication requirement with few failures.
386 Medical Imaging
2. The sensors, convert the applications measured/observed physical properties into
corresponding electrical signals that will be inputs for the computer system.
3. The effectors, convert electrical signals from the computers outputs to corresponding
physical actions that control an applications functions.
4. The operator the human(s) who monitors or activates the computer system in real-time (or
near real-time).
5. The computer/controller, composition of software and hardware, uses the sensors and the
effectors to monitor and control the application.
Application - Physical
(Process under control)
Effectors Sensors
Computer/Controller
(Virtual control policy)
Process Input
Disturbances
Process Output
Operator
Fig. 2. Traditional Digital Control System Components
A simple control system view of the guidance phase of an IIGS system is presented in
Figure 3
13
. There are several elements that make this representation different from traditional
digital control theory as displayed in Figures 2. Some of the ve components listed above
change between the physical and virtual realms. The controller is external to the computer. The
process under control (application) is in a virtual environment. The controlled variable is the
accurate placement of surgical instruments in a human body. The surgeon, as the controller,
13
This control system model was elaborated early in the research to represent a class of problems; it does
not represent any specic system. It is presented here because it was the initial idea behind the research.
has limited observability
14
of the controlled variable in the physical domain. The observable
behavior is the calculated position of surgical instrument models superimposed on the virtual
environment. What is more interesting is that some of the views provided by the system are
not there in the physical realm. The surgeon controls the calculated position of the instruments
in a computer generated model. The effectors are an interactive system involving surgeon,
display and surgical instruments. This is the inverse of what is normally considered a classical
digital control problem, where the controller implements a control strategy through a series of
specication functions and the process observable behavior is in the physical environment.
The model of Figure 3 only represents the guidance phase of IIGS and must be augmented
to include other phases, input variables, noise and disturbances. This will give us a system
centric view of the safety problem.
One important aspect of this process control model is that it is independent of the software
architecture of a particular system. Another added benet of using process control diagrams
in representing a virtual process is that the physical and conceptual (virtual) components,
such as software supported processes and protocols, can both be represented in the same
abstraction and be analyzed as part of a complete system. This visual representation of
all components is critical to the complete description and analysis of the system. It allows
for a more complete hazard analysis that is not compartmentalized. Contributing factors to
accuracy errors, as well as the propagation of these errors through the system, will be more
visible.
Move Targets
(Surgeon)
Controller
Locate Targets in 6 DOF
Estimate Placement of
Surgical Instruments
Display in Real-Time
6
m - -
Placement of Visible
Portion of Instruments
(Manipulated Variables)
Surgical
Plan
?
?
?
Computer Process Under Control (Virtual)
Estimated Instruments Placement
(Controlled Variables)
Fig. 3. Process Control View of Interactive Image-Guided Surgery
5.1 Process control model structure diagram
In this subsection we discuss the modeling diagramused as part of the Abstract Process Safety
Analysis Model. The structure diagram is used to model the process of IIGS and will be used
to guide the safety analysis performed at the lower layers of the architecture.
14
For some procedures, the surgeon will not be able to observe the end of instruments and must rely on
the computer model. In other cases, the surgeon can observe some aspects of the physical instruments
and provide coarser placement in the absence of the computer model, but with reduced accuracy.
Observability and controllability in the later case affect the ability to maximize surgical outcome.
388 Medical Imaging
The structure diagram is also used to display the current status of the system analysis based
on the collation of safety analysis information. Each element of the Process Control Model
Structure Diagram is to be coloured with the status information and the safety risk will be
recorded numerically with each symbol on the structure diagram.
We rst highlight the diagrammatic considerations that are sufcient to describe and
substantiate our approach and to model the structure of IIGS systems as Process Control
Models. We then introduce the syntax and notation used to model the structure of Process
Control Models such as IIGS systems.
5.2 Diagrammatic modeling
In simple cases, a diagrammatic representation can model an idea to fully meet a given
purpose; studying spatial relations, modeling the relationships in an organization or a
systems structure (i.e. the command relationship in an organization or the oor plan of an
ofce). For such simple cases, one item on the diagram fully represents a sufcient abstraction
of this item from the physical world. For the case where the entire abstraction of the system
can be represented on a diagram, we intend to capture, represent and analyze structure
only. For this kind of analysis, we simply need to draw a diagram with components having
recognizable and meaningful symbols (notation), rules for connecting the symbols (syntax)
and appropriate naming labels or markings. There is no need for further meaning (semantics).
We say that these diagrams have shallow semantics"
15
because the entire design or idea is
conveyed in one level of abstraction and no further semantics are required.
We use shallow semantics to convey the idea of representing the physical and virtual
characteristics of a system on a single layer diagram of components and connectors. By
comparison, deep semantics conveys the idea of adding layers to the model and adding
meaning to describe the safety properties of the system being analyzed.
Process control modeling is the kind of problem domain that requires precise syntax and
deep semantics. If we want to model the process control of a hydraulic turret in a missile
defence system, and analyze it for safety we need to augment the turrets structure diagram
with safety analysis techniques such as Failure Mode Effect Analysis (FMEA) or Fault Tree
Analysis (FTA) IEC (1985; 1990); Leveson (1995). In a simplistic way, the structure diagram
then describes the structure of physical components of the model such as hoses, valves,
transducers, ow direction and pressure. The structure diagram is only part of the entire
model. When we augment the structure diagram with mathematical models, analytical
methods and information we further increase the semantics of the analysis.
The approach that we have taken to model IIGS systems to conduct safety analysis reects
the approach taken in process control modeling such as the one described in the paragraph
above.
5.3 Design of the APSAM syntactic types
In this section, we describe the notation and syntax that will be used to create the APSAM
structure diagram.
15
Here we dene shallow semantics. The expression has been used in various context but no reference
could be found as a source to provide a denition.
5.3.1 Types
Syntactic types represent the different classes of components and connectors on a diagram
and help to distinguish between elements in a system Dean & Cordy (1995). Syntactic types
are similar to abstract data types in programming languages. They have shallow semantics
and must be further detailed during instantiation as typed nodes Dean & Cordy (1995) onto
the Process Control Model structure diagram (subsection 5.3.4).
5.3.2 Component types
Fig. 4. Notation Types for Components.
The set of Types for components is displayed in Figure 4.
Real-Time Physical Processes are those processes that change the physical environment
and which have observe-and-react timing requirements that suffer from latency which
can affect overall system performance and/or surgical outcome. Such processes include
the movement of effectors, changes to model data and direct observation of the physical
environment.
Real-Time Virtual Processes are processes that change the virtual environment, may be
subject to latency effects and may also contribute to these effects by means of computation
delays.
Near Real-Time Physical Processes must provide good quality of service in response time,
but these processes can be interrupted without jeopardizing system outcome. It should be
understood that the interruption should not be excessive to the point where the person in
the loop forgets where he was in the process of transformation. Near Real-Time processes
could have been called quality of service time as well because they closely characterize the
timing requirements.
390 Medical Imaging
Near Real-Time Virtual Processes must provide a good quality of service in response time
to the person in the loop. These processes produce the input necessary to the conduct of
the Real-Time processes (both virtual and physical).
Sensors provide estimates of physical data for use by the virtual environment. They
increase the visibility onto the physical world provided by the virtual environment.
Effectors are physical components that are moved or used in the physical environment.
Digital models of effectors are created to show their location in the virtual environment.
We use the same syntactic type to represent sensors and effectors since labels can provide
enough characterization without adding to the notation. Prostheses are also modeled using
this syntactic type.
Disturbances are unwanted input to the system that may change physical and virtual
processes. They can cause deviation from the control policy as well as reduce visibility
and controllability in the system. Disturbances can come from various sources including
but not limited to: electronic noise, errors in images or digitization, motion artifacts, light
noise (infrared camera interference), obscuration, etc...
Person is any human participant in the process. Only those people who are directly
involved in the process and can affect safety issues should be modeled.
Labels are used to identify all components and provide a role for the component as part of
the system.
5.3.3 Connector types
The set of types for connectors is displayed in Figure 5.
The interface connector represents the unidirectional interface between two components.
The interface may be physical or a ow of information. The ow of information is in the
form of an analog or digital stream and the information is not retained or saved between
the two components, it is not persistent. If one of the components is a process and it stops,
the stream of information is interrupted.
The information owwith memory is a unidirectional connector that has the responsibility
to transfer data between components and provide persistence between successive
executions of the processes. The round end of the connector is at the source of the
information transfer. The arrow is at the sink end. There is no need to identify on the
structure diagram where the information is persistent. This can happen at the source, sink,
at both ends or in the middle. The importance of this connector is to highlight limitations
and requirements for freshness of information, security, mishandling, misplacement, stable
storage, recovery, duplication, bandwidth, coherence and transport. The issues listed here
are unique to this connector because of possible time lag affecting information ow with
memory.
The interactive connector identies the interfaces in the system that require continuous
ow of information. It has signicant symbiotic relationships between components.
The summation point is a junction for connectors. The summation point is the only element
on the APSAM structure diagram that is not part of the safety analysis. Its only purpose is
to help clarify the diagram.
Labels are used to identify the interface or relationship between components.
Fig. 5. Notation Types for Connectors.
An important distinction to be made is for the Disturbance component type. In classical
process control modeling (Figure 2) input variables, set points and disturbances are normally
represented as arrows that come from nowhere and point to where they induced an effect.
When we developed the notation we wanted to differentiate between a connector that
represents a ow of information and one that represents a disturbance, which is an unwanted
manifestation representing uncontrollable effects that come fromthe environment. Our choice
was to show such phenomena as a black dot representing the environment with an arrow to
attach to the model at the point of induction. Disturbances can connect to components to show
that the effect is specic or internal to the component, or to connectors to represent that the
disturbance inuences the ow of information or signal between components. Disturbances
that can occur at the interface between the connector and the component are attached to the
connector.
5.3.4 Process control model structure diagram
The Process Control Model Structure Diagram, or simply structure diagram, is a concept
for the safety analysis expert to instantiate syntactic types to model system structure and
behavior Dean & Cordy (1995). The modeler is responsible for providing each instantiated
type with an identity in the form of a label.
Connectors are instantiated as rst class objects in the model Shaw & Garlan (1996).
Connectors as rst class objects convey the idea that they not only show relationships on a
diagram between two components, but that they have deeper semantics similar to that of
components. The analyst uses connectors to represent relationships between components
covering all meaningful system interfaces. Some interfaces need not be represented as they
have no impact on the safety analysis. These interfaces could be represented on a separate
view for completeness.
6. Results
We present some of the most interesting or signicant ndings of the application of the safety
analysis protocol. The ndings that were generated during the iterative elaboration of the
392 Medical Imaging
analysis methods, design and the related research activities were key to the development
of the protocol and conceptual architecture as well as in the identication of information
requirements and analysis forms.
The APSAM structure diagram we elaborated for our orthopedic systems contains 28
Components and 29 Connectors, each was validated by computer system specialists that
designed parts of the IIGS systems that were used during the analysis.
After the APSAM structure diagram was completed a System Hazard Analysis (SHA) Sha
(1999) was performed for each component and connector. SHA analysis identied 57
Composite Hazards. Each of the hazard was identied with the system expert that validated
the hazard. 33 of the composite hazards were identied to have propagation effects. The
propagation effects were also described as part of the analysis and are recorded in an integral
database. The Fault Tree Analysis (FTA) IEC (1990) that followed the SHA identied 139
Hazardous conditions (leaves). 49 of the FTA leaves were mitigated by competent human
intervention. The analysis revealed only 7 uncontrolled or unresolved conditions. These
uncontrolled or unresolved conditions did not imply that the system is unsafe but the experts
believe that there is a need to further investigate these hazardous conditions. As part of the
analysis we identied 31 Safeguard requirements; most were already are current practice but
were not documented.
The ndings below were elicited from various sources. They stem from direct observation
during surgeries and observed surgical gestures of the physicians while using the IIGS
systems for the surgeries; some of which are discussed in this section. Other ndings came
fromthe professional judgment and observations of the attending physicians during surgeries
which were recorded.
The physicians would often address the clinical research scientist who operated the computers
during the surgery on the operation of the system, limitations and improvements. Interviews
with computer system specialists and a clinical research scientist were used to get the more
technical information. The safety analysis protocol and the prototype of the architecture as
well as the Process Control Model Structure Diagram were used to guide and record the
iterative elicitation of hazards and their underlying hazardous conditions.
The most important nding is that there is a requirement for further research to evaluate
the person machine interfaces in the system to better present information and reduce the
loss of situational awareness. There are four different kinds of information that can be
presented to the surgeon: Anatomic and morphologic using visualization of medical images,
visual tracking, medical plans that show desired corrections and implants, and technical
data which provides accuracy, measurements, depth, applied force and the like. Surgeons
sometimes temporarily lose situational awareness during computer-guided surgery. This loss
of situational awareness can occur if the presentation of the view is different from the normal
physical views. Loss of situational awareness can also occur if the virtual surgical site is not
the one the surgeon recognizes from the plan. In some cases the temporary loss in situational
awareness can be caused if the image was not rotated in the right direction or the right plane
by the computer operator. Zooming errors can also cause delays. Through all the surgeries
that were witnessed for this research, the surgeons regained situational awareness in less
than a minute. However, the temporary confusion could lead to longer procedures or loss
of condence in the tool in general clinical practice.
Complete identication of single points of failure is also very important. In some safety critical
systems, if you cannot eliminate or mitigate single points of failure, your system is deemed
unsafe and you must redesign or add redundant systems Dunn (2003). For IIGS we can allow
the presence of single points of failure or degradation
16
in the system as long as competent
human intervention can be used to mitigate the risk. That requires that there is visibility (or
observability) of the failure. For example if a system fails to register, it represents a hazard,
but the likelihood of occurrence is low and criticality is low because the surgeon can detect
this condition in the large majority of cases and bail out of using the system.
Another important nding that quickly came to the forefront is that our systemlevel approach
encouraged thinking in terms of component and interface modications to introduce
safeguards or risk reduction mechanisms for introduced and attendant hazards. This system
level approach makes visible potential hazards and ways to remove or mitigate those hazards.
By analyzing the leaves of the FTA trees we found that there are three main mitigation
methods that can be applied to IIGS systems:
Competent Human Intervention: Competent human intervention was very useful in
providing mitigation information for hazardous conditions where the surgeon can use
empirical geometric rules to decide if the procedure is still within operating ranges. The
idea to create the state-based representation of competent human intervention came from
observations in the operating room during several surgeries.
Phantom Tests: The use of tests with phantoms can be used as a mitigation factor where
lack of accuracy or spatial delity constitute a hazard. However, motion artifacts and
the resulting propagation effects cannot be mitigated with phantom tests. Motion artifacts
need to be further investigated to quantify their effects on IIGS. Articulated cadavers could
be used to simulate tensing and patient movements during imaging. The movements
could be measured mechanically and the effects on motion artifacts quantied on the
resulting images.
Code Inspections: Fagan style code inspection or some more agile inspection methods
can be used to review parameters and transformation in the code of the IIGS systems.
Pair programming or peer review could be used if the resources are not available. The
calibration criteria for tracking cameras are set by programmers as are parameter les,
marker types, lens models, interface packing routines, cable types and many other settings.
The processing of tracking raw data from the 3-D cameras can also lead to inaccuracies.
These settings, calibration data and calculations using the data to provide the registration
and guidance information are not visible to the operator or the surgeon. Code and data
inspections can be useful to ensure correct values are used.
7. Conclusion and future work
The research aimed at improving several attributes while performing safety analysis for IIGS
systems. The safety analysis using the prototype implementation of the protocol provided
us with the ability to validate traceability and visibility attributes. The notation used with
the prototype helped us demonstrate the expressiveness of the APSAM. The maintainability
16
We use failure to indicate that the system cannot be used and degradation if the system can be used in
a degraded mode with concurrence from the surgeon.
394 Medical Imaging
attribute is validated by the integration of the safety analysis methods under a single
high-level abstraction, the APSAM structure diagram and a repository that enforces relational
integrity. Maintainability was identied as one of the main benets of the protocol and its
supporting conceptual architecture.
Finally the validation of the IIGS APSAM for orthopedic surgery with domain experts using
the prototype helps us support the conclusion that the completeness of the safety analysis
has been increased from what it would have been using the baseline methods for comparison
described in Section 4.
We can also support the conclusion that the scale of rigor provided by the APSAM and its
safety analysis protocol provides the necessary due care for safety analysis for such systems
where competent human intervention is present. The traceability and visibility of the hazards
and their resolution should be sufcient to obtain Food and Drugs Administration market
clearance for like systems. The level of the analysis also provides for efcient use of domain
experts.
Future research is required to apply the process control model including APSAM to other
interactive systems where the operator or expert makes control decisions based on a virtual
representation of the environment. Much work needs to be done to quantify system latency
and its effects on safety. This quantication of latency is key when systems that warn surgeons
of impending surgical errors like the one proposed by Luz et al. (2010) are considered.
Recently a specication for an extension to the Unied Modeling Language called the Systems
Modeling Language or SysML was published under control of the Object Management
Group Team(2006). The intent of the document is to retain the main structure and components
of UML 2.0 and to perform system engineering domain modeling. Much of SysML makes use
of newstereotypes but some interesting extensions have been proposed because UML was too
restrictive to represent some system engineering concepts. Safety has been considered as part
of this proposed standard to specify safety under stereotyped classes for requirements. There
are no other syntactic elements or semantics to further model these system safety requirement
classes or to drive and track a system safety analysis. The use of SysML is likely to become a
standard for system engineering modeling similar to the use of UML 2.0 to model software.
An interesting future research direction would be to propose an extension to SysML based on
APSAM and implemented on Eclipse.
The generic notation and syntax, as well as the protocol, that have been produced as part of
this research are not limited to IIGS systems. Trying the model in other safety critical system
domains that have abstract interactive processes and person in the loop is certainly possible
and should be investigated. Any system that provides estimated information for decision
making in a critical environment could benet from an analysis conducted with the APSAM
and its protocol.
8. References
Assessment and Management of Cancer Risks from Radiological and Chemical Hazards (1998). Report
H39-428/1998E, Health Canada, Atomic Energy Control Board and Ontario Ministery
of Environment and Energy.
Ayache, N., Cinquin, P., Cohen, I., Cohen, L., Leitner, F. & Monga, O. (1996). Segmentation
of complex three-dimensional medical objects: A challenge and a requirement for
computer-assisted surgery planning and performance, in R. H. Taylor, S. Lavalle,
B. G. C. & R. Msges (eds), Computer-Integrated Surgery - Technology and Clinical
Applications, MIT Press, Cambridge, Massachusetts, collection of new and reedited
papers. 4, pp. 5974.
Birkfellner, W., Watzinger, F., Wanschitz, F., Ewers, R. & Bergmann, H. (1998). Calibration
of tracking systems in a surgical environment, IEEE Transactions on Medical Imaging
17(5): 737742.
Breeuwer, M. et al. (1998). EASI Project - Improving the effectiveness and quality of
image-guided surgery, IEEE Transactions on Information Technology in Biomedicine
2(3): 156168.
Choi, J. J., Cleary, K., Zeng, J., Gary, K., Freedman, M., Watson, V., Lindisch, D. & Mun,
S. K. (2000). I-SPINE: A software package for advances in image-guided and
minimally invasive spine procedures, Proceedings of SPIE - The International Society
for Optical Engineering, 28th AIPR Workshop: 3D Visualization for Data Exploration
and Decision Making, Oct 13-Oct 15 1999, Washington, DC, USA, Vol. 3905, SPIE;
AIPR Executive Committee Publisher: Society of Photo-Optical Instrumentation
Engineers, Bellingham, WA, USA, Society of Photo-Optical Instrumentation
Engineers, Bellingham, WA, USA, pp. 242247.
Chudleigh, M., Berridge, C., May, R., Butler, J. & Poole, I. (1995). SADLI project: Safety critical
software research in the medical diagnostic domain, Computing & Control Engineering
Journal 6(5): 211215.
Dean, D., Kamath, J., Duerk, J. L. & Ganz, E. (1998). Validation of object-induced MR
distortion correction for frameless stereotactic neurosurgery, IEEE Transactions on
Dean, T. R. & Cordy, J. R. (1995). A syntactic theory of software architecture, IEEE Transactions
on Software Engineering 21(4): 302313.
Dunn, W. R. (2003). Designing safety-critical computer systems, IEEE Computer 36(11): 4046.
Elliott, L., Mojdehbakhsh, R. & Tsai, W.-T. (1994). Process for developing safe software,
Proceedings of the 1994 IEEE 7th Symposium on Computer-Based Medical Systems, Jun
11-12 1994, Winston-Salem, NC, USA, IEEE Symposium on Computer-Based Medical
Systems, IEEE Computer Society, IEEE, Los Alamitos, CA, USA, pp. 241246.
Fei, B., Ng, W. S., Chauhan, S. & Kwoh, C. K. (2001). The safety issues of medical robotics,
Reliability Engineering and System Safety 73(2): 183192.
Fitzpatrick, M. J., West, J. B. & Maurer, Jr, C. R. (1998). Predicting error in rigid-body
point-based registration, IEEE Transactions on Medical Imaging 17(5): 694702.
Fries, R. C., Pienkowski, P. J. & Jorgens, III, J. (1996). Safe, effective, and reliable software
design and development, Biomedical Instrumentation & Technology 30(2): 136149.
Publisher: Assoc for the Advancement of Medical Instrumentation, Arlington, VA,
USA.
Galloway, Jr, R. L., Maciunas, R. J., Bass, W. & Carpini, W. (1994). Optical localization for
interactive, image-guided neurosurgery, Proceedings of SPIE - The International Society
for Optical Engineering, Medical Imaging 1994: Image Capture, Formatting, and Display,
Feb 13-14 1994, Newport Beach, CA, USA, Vol. 2164, pp. 137145.
Gary, K., Ibez, L., Aylward, S., Gobbi, D., Blake, B. M. & Cleary, K. (2006). Coping with
defective software in medical devices, IEEE Computer 39(4): 4045.
Gauldie, D. R. (2002). Calibration and registration with 3d a-mode ultrasound, Masters thesis,
Queens University, Kingston, Ont.
396 Medical Imaging
Gowen, L. D. (1994). Specifying and verifying safety-critical software systems, Proceedings
of the 1994 IEEE 7th Symposium on Computer-Based Medical Systems, Jun 11-12 1994,
Winston-Salem, NC, USA, IEEE Symposium on Computer-Based Medical Systems,
IEEE Computer Society, IEEE, Los Alamitos, CA, USA, pp. 235240.
Gowen, L. D. (1995). Summary of the 94 CBMS symposiums software safety workshop,
Proceedings of the 8th IEEE Symposium on Computer-Based Medical Systems, Jun 9-10
1995, Lubbock, TX, USA, IEEE Symposium on Computer-Based Medical Systems,
IEEE; SPIE, IEEE, Los Alamitos, CA, USA, pp. 330337.
Gowen, L. D. & Collofello, J. S. (1994). Assessing traditional verications effectiveness on
safety-critical software systems, Journal of Systems and Software 26(2): 103115.
Gowen, L. D. & Yap, M. Y. (1993). Traditional software developments effects on safety,
Proceedings of the 6th Annual IEEE Symposium on Computer-Based Medical Systems,
Jun 13-16 1993, Ann Arbor, MI, USA, IEEE Symposium on Computer-Based Medical
Systems, pp. 5863.
Halang, W. A., Sniezek, M. & Colnaric, M. (1998). Computerized controllers for safety critical
medical applications, International Journal of Medical Informatics 49(2): 139155.
Herrmann, D. S. (1995). Preview of IEC safety requirements for programmable electronic
medical systems, Medical Device and Diagnostic Industry 17(6): 4pp.
Hindus, L. (2001). Medical visualization realities now: Less fantastic voyages through the
body, Advanced Imaging 16(2): 42, 44, 50.
IEC (1985). IEC 60812:1985 International Electrotechnical Commission - Analysis techniques for
system reliability - Procedures for Failure Mode and Effects Analysis (FMEA), rst edn.
IEC (1990). IEC 61025:1990 International Electrotechnical Commission - Fault Tree Analysis - FTA,
rst edn.
Jiang, Z., Miao, S., Zamorano, L., Li, Q., Gong, J. & Diaz, F. (1998). Technology and human
errors in image-guided surgeries, Proceedings of Surgical-Assist Systems, Jan 25-28
1998, San Jose, CA, Vol. 3262 of Proceedings of SPIE - The International Society for
Optical Engineering, IBOS; SPIE, The International Society for Optical Engineering,
Bellingham, WA, USA, pp. 7884.
Jones, P. L., Jorgens, III, J., Taylor, Jr, A. R. & Weber, M. (2002). Risk management in the
design of medical device software systems, Biomedical Instrumentation and Technology
36(4): 237266.
Knight, J. C. (1990). Issues of software reliability in medical systems, Proceedings of the Third
Annual IEEE Symposium on Computer-Based Medical Systems, Jun 3-6 1990, Chapel Hill,
NC, USA, IEEE Computer Soc; IEEE Engineering in Medicine & Biology Soc; IEEE
Eastern North Carolina Section; Univ of North Carolina; North Carolina State Univ;
et al; Research Triangle Institue, pp. 153160.
Lavall, S. (1996). Registration for computer-integrated surgery: Methodology, state of the art,
in R. H. Taylor, S. Lavalle, B. G. C. & R. Msges (eds), Computer-Integrated Surgery -
Technology and Clinical Applications, MIT Press, Cambridge, Massachusetts, collection
of new and reedited papers. 32, pp. 7797.
Lavall, S., Troccaz, J., Sautot, P., Mazier, B., Cinquin, P., Merloz, P. & Chirossel, J.-P. (1996).
Computer-assisted spinal surgery using anatomy-based registration, in R. H. Taylor,
S. Lavalle, B. G. C. & R. Msges (eds), Computer-Integrated Surgery - Technology and
Clinical Applications, MIT Press, Cambridge, Massachusetts, collection of new and
reedited papers. 32, pp. 425449.
Lefngwell, D. A. & Norman, B. (1993). Software quality in medical devices: a top-down
approach, Proceedings of the 6th Annual IEEE Symposium on Computer-Based Medical
Systems, Jun 13-16 1993, Ann Arbor, MI, USA, IEEE Symposium on Computer-Based
Medical Systems, pp. 30731l.
Leveson, N. (1995). Safeware: System Safety and Computers, Addison Wesley, Boston.
Likert, F. (1932). A technique for measurement of attitudes, Archives of Psychology (140).
Loesh, R. E., Gosnell, A. B., Wyskida, R. M., Johannes, J. D. & Benoist, T. F. (1999). Engineering
approach to critical software certication, Proceedings of the 1999 32nd Annual Hawaii
International Conference on System Sciences, HICSS-32, Jan 5-8 1999, Maui, HI, USA,
IEEE, IEEE Comp Soc, Los Alamitos, CA, USA, p. 282.
Lorensen, W. & Cline, H. (1987). Marching cubes: A high resolution 3d surface construction
algorithm, Computer Graphics 21(4): 163169.
Luz, M., Mueller, S., Strauss, G., Dietz, A., Meixensberger, J. & Manzey, D. (2010). Automation
in surgery: The impact of navigated-control assistance on the performance, workload
and situation awareness of surgeons, Proceedings of the Human Factors and Ergonomics
Society, 54th Human Factors and Ergonomics Society Annual Meeting 2010, HFES 2010,
September 27, 2010 - October 1, 2010, Vol. 2 of Human Factors and Ergonomics Society
Annual Meeting, pp. 889893.
Mojdehbakhsh, R., Tsai, W.-T. & Kirani, S. (1994). Retrotting software safety in an
implantable medical device, IEEE Software 11(1): 4150.
Msges, R. & Lavalle, S. (1996). Multimodal information for computer-integrated surgery,
in R. H. Taylor, S. Lavalle, B. G. C. & R. Msges (eds), Computer-Integrated Surgery -
Murthy, S. (1995). Retrospective validation of medical device software, Medical Device and
Diagnostic Industry 17(5): 7pp. Publisher: Canon Communications Inc, Santa Monica,
CA, USA.
on Radiological Protection, I. C. (ed.) (1960-Present). Annals of the ICRP, Elsevier.
Patel, K. C., Duerk, J. L., Zhang, Q., Chung, Y.-C., Williams, M., Kaczynski, K., Wendt,
M. & Lewin, J. S. (1998). Methods for providing probe position and temperature
information on MR images during interventional procedures, IEEE Transactions on
Peters, T. (2001). Image-guided surgery and therapy: Current status and future directions,
Proceedings of SPIE - The International Society for Optical Engineering, Medical Imaging
2001: Visualization, Display, and Image-Guided Procedures, Feb 18-20 2001, San Diego,
CA, Vol. 4319, pp. 112.
Pieper, S., McKenna, M., Chen, D. & McDowall, I. E. (1994). Computer animation
for minimally invasive surgery: Computer system requirements and preferred
implementations, Proceedings of SPIE - The International Society for Optical Engineering,
Stereoscopic Displays and Virtual Reality Systems, Feb 8-10 1994, San Jose, CA, USA,
Vol. 2177, IS&T - Soc for Imaging Science and Technology, Springeld, VA USA;
SPIE - Int Soc for Opt Engineering, Bellingham, WA USA, Society of Photo-Optical
Instrumentation Engineers, Bellingham, WA, USA, pp. 401408.
Proo, A. E. (1979). Radiation Shielding and Dosimetry, Wiley, New York.
Radau, P. E., Slomka, P. J., Julin, P., Svensson, L. & Wahlund, L.-O. (2000). Automated
segmentation and registration technique for HMPAO-SPECT imaging of Alzheimers
patients, Proceedings of SPIE - The International Society for Optical Engineering, Medical
398 Medical Imaging
Imaging 2000: Image Processing, Feb 14-Feb 17 2000, San Diego, CA, USA, Vol.
3979, Society of Photo-Optical Instrumentation Engineers, Bellingham, WA, USA,
pp. 372384.
Robb, R. (1996). VR assisted surgery planning, IEEE Engineering in Medicine and Biology
15(2): 6069.
Schneider, P. & Hines, M. L. (1990). Classication of medical software, Proceedings of the 1990
Symposium on Applied Computing, Apr 5-6 1990, Fayetteville, AR, USA, IEEE Computer
Soc; Arkansas Soc for Computer & Information Technology (ASCIT); ACM, IEEE,
Piscataway, NJ, USA, pp. 2027.
Sha, L. (1999). High assurance bio-medical device control, Proceedings, v2, 1999, Proceedings of
the 1999 IEEE Engineering in Medicine and Biology 21st Annual Conference and the 1999
Fall Meeting of the Biomedical Engineering Society (1st Joint BMES / EMBS), Oct 13-Oct
16 1999, Atlanta, GA, USA, Annual International Conference of the IEEE Engineering
in Medicine and Biology, p. 879.
Shaw, M. & Garlan, D. (1996). Software Architecture: Perspectives on an Emerging Discipline,
Prentice Hall, Upper Saddle River, New Jersey.
Starreveld, Y., Gobbi, D., Finnis, K. &Peters, T. (2001). Software components for medical image
visualization and surgical planning, Proceedings of SPIE - The International Society
for Optical Engineering, Medical Imaging 2001: Visualization, Display, and Image-Guided
Procedures, Feb 18-20 2001, San Diego, CA, Vol. 4319, pp. 546556.
Taylor, R. H., Lavalle, S., C., B. G. & Msges, R. (eds) (1996). Computer-Integrated Surgery -
Technology and Clinical Applications, MIT Press, Cambridge, Massachusetts.
Team, S. M. (2006). Systems modeling language sysml specication draft, Technical Report
OMG Document ad/2006-03-01 Version 1.0, Object Management Group.
Tenforde, T. S. (ed.) (1979). Magnetic Field Effect on Biological Systems: Based on the Proceedings
of Biomagnetic Effects Wrokshop held at Lawrence Berkeley Laboratory, University of
California, on April 6-7, 1978, Plenum Press, New York.
Udupa, J. (2001). Three-dimensional rendering in medicine: Some common misconceptions,
Proceedings of SPIE - The International Society for Optical Engineering, Medical Imaging
2001: Visualization, Display, and Image-Guided Procedures, Feb 18-20 2001, San Diego,
CA, Vol. 4319, pp. 660670.
Udupa, J. K. & Gonalves, R. J. (1996). Imaging transforms for volume visualization, in
R. H. Taylor, S. Lavalle, B. G. C. & R. Msges (eds), Computer-Integrated Surgery -
Udupi, V., Raghavendra, A. & Inamdar, H. (2001). Computer vision method for biomedical
image analysis, IETE Technical Review (Institution of Electronics and Telecommunication
Engineers, India) 18(5): 365373.
van der Weide, R., Zuiderveld, K. J., Bakker, C. J., Hoogenboom, T., van Vaals, J. J. &Viergever,
M. A. (1998). Image guidance of endovascular interventions on a clinical MR scanner,
IEEE Transactions on Medical Imaging 17(5): 779785.
Verbeeck, R., Michiels, J., Nuttin, B., Knauth, M., Vandermeulen, D., Suetens, P., Marchal, G.
& Gybels, J. (1995). Protocol for the clinical functionality assessment of a workstation
for stereotactic neurosurgery, IEEE Transactions on Medical Imaging 14(3): 577586.
Viergever, M. A. (1998). Image guidance of therapy, IEEE Transactions on Medical Imaging
17(5): 669671.
Zamorano, L., Jiang, Z. &Kadi, A. M. (1994). Computer-assisted neurosurgery system: Wayne
Sate University hardware and software conguration, Computerized Medical Imaging
and Graphics 18(4): 257271.
Zoroo, R., Nishii, T., Sato, Y., Sugano, N., Yoshikawa, H. & Tamura, S. (2001). Segmentation
of avascular necrosis of the femoral head using 3-D MR images, Computerized Medical
Imaging and Graphics 25(6): 511521.
400 Medical Imaging

Medical Imaging

Uploaded by

Copyright:

Available Formats

Medical Imaging

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Medical Imaging

Uploaded by

Copyright:

Available Formats

MEDICAL IMAGING

Edited by Okechukwu Felix Erondu

is the directional energetic radiance at the spatial position r

is a position on the boundary and

is an outer normal unit vector. The

is the roughness at the resolution . w

f (s, t)ds , (9)

is extracted from the complete 3D utrasound image.

. This segmentation must be

and a narrow band giving

, b) denition of the narrow

=0.49mS/cm. The question is how to include this

z -direction. Likewise the material complex permittivity to be sought

z ) and transverse (t) components as:

as r . For a two-dimensional case (29a) where

phase shifted (quadrature)

(since inputs are equiprobable)

(; x)p(; x)d(x) , (15)

are the fundamental solutions for the potential and

is the coefcient vector, and X is a matrix of N rows and (K +

g. The bias-corrected image is then computed

, and letting w denote the weight vector (w

, h, X; )) is a partition function; (y, h, X; ) R is a

L() can be obtained with

)) as shown in Fig.4 (b).

, X; )) is a partition function; (L, X; ) R is a potential

denotes all possible state transitions from l

of the input sequence is

, we tuned this parameter and trained the

T and a 3 by 3 rotation matrix

You might also like