J Biomed Res. 2013 Nov; 27(6): 439–451.

Published online 2013 Sep 28. doi: 10.7555/JBR.27.20130138

PMCID: PMC3841469

Multimodality image fusion for diagnosing

coronary artery disease
Marina Piccinelli✉ and Ernest Garcia
Author information ► Article notes ► Copyright and License information ►
This article has been cited by other articles in PMC.
Go to:

Abstract
Coronary artery disease (CAD) is one of the leading causes of death in the US and a substantial health-care
burden in all industrialized societies. In recent years we have witnessed a constant strive towards the
development and the clinical application of novel or improved detection methods as well as therapies.
Particularly, noninvasive imaging is a decisive component in the cardiovascular field. Image fusion is the
ability of combining into a single integrated display the anatomical as well as the physiological data
retrieved by separated modalities. Clinical evidence suggests that it represents a promising strategy in CAD
assessment and risk stratification by significantly improving the diagnostic power of each modality
independently considered and of the traditional side-by-side interpretation. Numerous techniques and
approaches taken from the image registration field have been implemented and validated in the context of
CAD assessment and management. Although its diagnostic power is widely accepted, additional technical
developments are still needed to become a routinely used clinical tool.

Keywords: image fusion, CAD diagnosis, computed tomography angiography, nuclear imaging

INTRODUCTION
Cardiovascular diseases (CVD) are the leading cause of death in most Western societies. Statistics from
recent updates from the American Heart Association[1],[2] report that an estimated 82.6 millions of
American adults, i.e. >1 in 3, have one or more types of CVD. Coronary artery disease (CAD) is estimated
to cause alone about 1 every five deaths and results in an estimated total annual cost of US $165.4 billion.
Although its mortality rate has decreased in the last decade, CAD prevalence is projected to increase over
the next 20 years remaining a significant health-care burden for all industrialized and developing
countries[3].

As a result of the profound economical and social impact of CAD prevalence, in conjunction with the
persistence of many of its risk factors (e.g. obesity, diabetes, hypertension, sedentary life), in recent years we
have witnessed a constant strive towards the development and the clinical application of novel or improved
detection methods as well as therapies. Particularly, noninvasive imaging is a decisive component in the
cardiovascular field, affecting all stages of disease management: from a correct and possibly early diagnosis,
to guidance through available therapeutic solutions, to the selection of the patients that would benefit from
those choices.
Today, most common noninvasive imaging modalities to investigate CAD include myocardial perfusion
imaging (MPI), performed by means of single photon emission computed tomography (SPECT) or positron
emission tomography (PET), and computed tomography angiography (CTA). All modalities have their own
weaknesses and strengths and which test will be initially performed on a particular patient depends mostly
on the symptoms and the pretest likelihood of CAD. CTA is primarily considered an anatomical study to
image the coronary tree that supplies blood to the myocardium, potentially capable of replacing invasive
coronary angiography (CA); SPECT and PET are highly reliable physiological tests conveying information
about myocardial perfusion (blood flow) and viability. A substantial amount of literature indicates that a
comprehensive technique for CAD assessment should provide the complementary information about
coronary morphology and myocardial function. The concept of ‘one-stop-shop’ -i.e. one imaging modality
that could alone assess all aspects of ischemic disease - has intrigued clinicians for some time[4], but despite
that technical improvements are continuously being implemented, each technique will likely keep bringing
its own specific advantages, at least in the cardiac field.

Instead, both cardiologist and researchers in the field agree that image fusion, i.e. the possibility of
combining into a single integrated display the clinical data independently retrieved by each modality,
represents a new promising strategy in cardiovascular diagnosis and risk stratification. The integration is
achieved by way of software methods globally catalogued under the term of image registration techniques
that align two different types of images of the cardiovascular system obtained either sequentially and, to
some extent, simultaneously, into a single framework. In its most successful cardiac application, image
fusion overlays anatomical and functional datasets: for example, CTA and SPECT may be combined, so that
each hypoperfused region can be directly linked to a specific stenosis along the coronary tree. In the fused
image the two studies work synergistically increasing not only the diagnostic power of each modality, but
also of their more traditional side-by-side interpretation[5].

In the following sections the more widely used noninvasive imaging modalities for CAD assessment will be
presented, and advantages and disadvantages of each approach will be outlined putting forward the rationale
for multimodality image fusion. A general description of the image registration framework and its essential
technical requirements will be introduced. Main techniques and results of image fusion and its clinical
potential will be reported. This review will conclude underlying the existing limitations of image fusion
approaches in the reality of clinical settings and patient well-being.

NON-INVASIVE IMAGING MODALITIES FOR CAD

ASSESSMENT
Computed tomography angiography

Although the gold standard for the characterization of coronary morphology remains invasive CA, contrast-
enhanced, electrocardiographically gated CTA is the most promising non-invasive modality to visualize both
calcified and non-calcified plaques and to estimate the degree of stenosis with a minimum pre-requisite of
64-multidetector row device. Given the tremendous advances in terms of temporal and spatial resolution,
CTA has been given significant attention in recent years; a number of single center and multicenter studies
have been published detailing its capability in correctly diagnosing CAD with respect to CA[6],[7],[8],[9].
In Fig. 1, a normal case and a case with left anterior descending artery disease are depicted by means of CA,
reformatted CTA and CTA-derived volume rendering technique.
Fig. 1

A normal case (A) and a case with LAD disease (B) visualized by means of reformatted CTA (top), CA views (bottom,
left) and CTA-derived volume rendering technique (bottom, right).

The main advantage of CTA is the excellent negative predictive value (NPV) - that is the ability of ruling out
the presence of disease - in symptomatic patients with low pre-test likelihood of CAD, but even in situations
of acute chest pain[10]–[13]. The main disadvantage is the suboptimal positive predictive value (PPV) - that
is the ability to correctly diagnose the presence of disease - for both >50% and >70% stenoses. The low PPV
is due to the frequent overestimation of luminal narrowing especially in situation of densely calcified
plaques. Calcifications induce blooming artifacts in the images, which in turn prevent the correct and
reliable assessment of the degree of stenosis; most commonly this results in overestimation of the severity of
the disease or in technically inadequate studies that prevent the accurate assessment of certain vascular
segments. Additionally, since CTA cannot evaluate ischemic burden, even if obstructive disease is detected
the functional relevance of a measured stenosis cannot be estimated[14]–[17] and consequently the actual
need for revascularization assessed. This is particularly true for intermediate lesions for which the questions
on whether CA would be needed and catheterization would be effective[18] may remain unanswered on the
basis of the CTA test alone. There are situations where CTA alone could answer the clinical question. Two
examples are patients with normal coronaries and high-risk patients with CTA-based diagnosis of 2- or 3-
vessel coronary disease and involvement of the proximal left anterior descending (LAD) coronary artery will
most likely benefit from revascularization[19].

Apart from the aforementioned limitations in evaluating densely calcified plaques and intermediate lesions,
CTA necessitates the administration of an intravenous iodinated contrast, which is particularly
contraindicated in patients with renal dysfunction. In patients with high heart rate, poor quality images may
also result from difficulties in performing gated acquisitions. The radiation exposure is also a limiting factor
to the widespread use of CTA, although recent developments and new protocols allow acquisitions with low
radiation dose[20].

Myocardial perfusion imaging: SPECT and PET

Rest and stress myocardial perfusion imaging, utilizing SPECT and PET radionuclide techniques, are today
the most efficient noninvasive methodology for the detection and quantification of obstructive CAD.
Presence of reversible perfusion defects of tracer accumulation are interpreted as a sign of exercise-induced
hypo-perfusion, while fixed defects at rest and stress as proxies for myocardial scar[21],[22]. The strength of
nuclear cardiology resides in the extensive amount of clinical and scientific publications supporting the use
of ischemia measurements in guiding patient management and therapeutical decisions. Additionally, in the
last decade, a number of software packages have been developed for the fully automated and objective
quantification of myocardial perfusion images and extraction of clinical data. These automated software
packages have allowed the standardization and widespread utilization of nuclear cardiology as the most
reliable diagnostic test for CAD disease evaluation and patient risk stratification[23],[24],[25]. In Fig. 2, the
active view ECTb display of a patient is shown[23]. the quantitative analysis includes: planar projections,
oblique slices, polar maps including normal database comparison, 3D displays, 17-segment summed score
polar maps, LV volume/time curve and regional thickening map.
Fig. 2

Active view display ECTb23.

Go to:

SPECT
Single photon emission computed tomography is the most frequently performed MPI technique in the US
for CAD assessment. Mean sensitivity and specificity for the detection of >50% stenosis are, respectively,
86% and 74%[26]. Robust data support its prognostic value[27]–[30]: the basic assumption is that risk of
adverse events is directly linked to the size of the perfusion defect, which in turn reflects the extent of
obstructive CAD; the larger the defect, the worse the prognosis. Annual rate of events for patients with
normal stress scans is 0.8% while it increases to annually 5.6% for abnormal scan.

The main disadvantages of SPECT imaging are that attenuation artifacts can be interpreted as perfusion
defects and that SPECT commonly underestimates CAD extent in case of multi-vessel disease and LAD
involvement, particularly in patients with diabetes. The latter is due to a mechanism called “balanced
ischemia” that reduces contrast heterogeneity between normally and abnormally perfused territories,
affecting the test ability to distinguish between them. MPI tests reveal ischemia in areas supplied by the
most severely stenosed vessels and no detection of subclinical atherosclerosis is possible. This is indeed
expected since MPI is designed to extract functional information, i.e. flow-limiting stenoses.

A crucial drawback when dealing with MPI interpretation is the lack of anatomical information. Myocardial
distribution territories are traditionally assigned to their subtending arteries by the use of the 17-segment
model as proposed by the American Heart Association[31]; it is, however, documented that in 17% to 50%
of patients the standard modeling does not correspond to the actual coronary anatomy[32], further
complicating the identification of the stenosis responsible for a specific perfusion defect.

Go to:

PET
Positron emission tomography (PET) shares many aspects with SPECT in terms of pros and cons. It can be
considered as an alternative test although until recently its diffusion as a standard clinical study was
hampered by the limited availability of devices and the complex production of proper radiopharmaceuticals.
With respect to SPECT, PET is characterized by a higher spatial and temporal resolution and benefits from
depth-independent attenuation correction techniques, which makes it the gold standard of myocardial
perfusion imaging. Its sensitivity and specificity are slightly better (90% and 89%, respectively[9]), as it is
the capability of detecting patients with multi-vessel disease. As opposed to SPECT, PET allows the accurate
quantitative measurement of absolute myocardial blood flow at rest and stress (measured in mL/g/min); the
coronary flow reserve can further be computed as the ratio between myocardial blood flow at stress and the
blood flow at rest. The clinical use of these measures has been shown to have strong additional prognostic
power in risk stratification especially in difficult cases with multiple vessel disease[33]–[35].

Image fusion: more than the sum

The routine application of CTA and MPI for the assessment of CAD provide complementary diagnostic
information for a comprehensive picture of disease severity. Another attribute of combining the use of these
two techniques is that their weaknesses affect different aspects of the complex task of CAD detection and
management. Whenever one of the two approaches, anatomical or functional, fails to be decisive in
detecting disease or guiding clinical decisions, the other one can complement the diagnosis with additional
information and help solving borderline findings. Densely calcified vessels will not affect image quality in
SPECT or PET studies; multi-vessel disease or left main coronary artery stenosis, which is often misread
with MPI, is, on the other hand, well diagnosed by means of CTA. Intermediate lesions can be difficult to
classify on CTA unless functional formation on the subtending myocardial perfusion is provided, while even
in the presence of a defect, MPI scans may not clearly identify the vessels responsible for the hypo-
perfusion, unless patient-specific coronary tree anatomy is given.

The excellent ability of this noninvasive integrated information to guide CAD management and limit the use
of CA to the patients that will truly benefit from revascularization techniques was demonstrated in a number
of studies[36],[37]. Patients will benefit from invasive procedures only when flow-limiting stenoses are
present and ischemia proven; on the opposite, aggressive treatment of non-flow-limiting stenoses - which is
not without periprocedural risks - will not significantly change patient's prognosis or symptoms[38]. In their
work, Gaemperli and co-workers reported that by CTA-MPI integration all flow-limiting stenoses were
identified, but also that a consistent number of cases that did not show flow-limiting stenoses or myocardial
ischemia eventually underwent unnecessary revascularization. The authors concluded that non-invasive
combined imaging approach could prevent unnecessary procedures and guide patient selection.

Thus, image fusion does not just deal with the convenience of having an additional cardiovascular test in
ambiguous cases (it is estimated that 15-30% of patients will require multimodality imaging tests[38]), but,
with the actual merging in the same visualization, display the two types of information simultaneously.
Traditionally, multimodal imaging studies have been viewed and analyzed separately in a side-by-side
fashion and mentally combined together to clear doubts or inconsistencies. In a study specifically designed
to address this issue[5], the authors examined the incremental diagnostic value provided by fusion of SPECT
and CTA studies with respect to MPI evaluation alone and CTA-MPI side-by-side approach. After the initial
diagnosis with only the MPI scans physicians were presented with the CTA datasets, which allowed them to
modify their reports in 50% of the cases; when the fused displays were provided, reviewers changed their
decisions in 28% of the cases. The confidence in interpretation was higher with software-integrated images
with decreased number of misinterpretations and better accuracy, particularly in the LAD territory (Fig. 3).

Fig. 3

Example of fusion display versus side-by-side interpretation for a patient with multi-vessel coronary artery disease.

Analogously, Gaemperli and coworkers[39] demonstrated that in one third of their cardiac patients the fused
CTA-SPECT analysis provided additional diagnostic information with respect to the side-by-side
interpretation of stand-alone CTA and SPECT acquisitions. Particularly, cases with lesions in the distal and
diagonal segments benefited by the integrated display. The ability of fused images to correctly establish the
hemodynamic significance of an intermediate lesion on CTA was further underlined as it may help avoiding
unnecessary interventional procedures.

Go to:

MEDICAL IMAGE REGISTRATION, THEORY AND METHODS

Image fusion is the procedure of integrating relevant information from two or more images into a single one
so that the result will be more informative than any of the input singularly considered. From the technical
point of view, image fusion applies techniques from the more general field of image registration[40],[41],
[42]. The final goal is the spatial alignment of two images so that anatomical or functional features can be
directly co-localized in both datasets point-by-point. The increasing use of radiological images obtained with
different modalities and at different points in time has made image registration a very active field in medical
image processing. Particularly, automated registration techniques have been the focus of considerable
investigation[43],[44]. Interactive alignment is still an option in research applications, often used as
reference for validation purposes, but lacks reproducibility and can be critically time-consuming when
translated to a clinical setting.

Three main components have to be defined for a registration procedure to be correctly outlined: a
transformation model, a similarity metric and an optimization method. The process of image registration will
find the optimal geometric transformation that maximizes the correspondences across the two images, the
floating image - the one that has to be registered - and the reference image. All three components are
application-dependent and a number of choices exist.

The transformation model defines how the coordinates between the images are related. The rigid body
transformation is the most common choice: it assumes that the distance between any two points in the body
remains the same irrespective of the image modality. The only possible geometric transformations between
the two representations are rotations and translations along X, Y and Z for a total number of six parameters -
scaling may also be included adding three more parameters. Eventually, a single coordinate transformation
maps each position in the floating image to the anatomically correspondent point in the reference one. Non-
rigid transformations are employed when the hypothesis of rigidity cannot be applied, which typically
happens when breathing motions or tissue movements between the two acquisitions cannot be disregarded.
In this case, a global mapping between floating and reference image cannot be written; instead a rigid
registration is commonly used as a starting estimate and the non-rigid component is modeled as a
deformation field that represents the displacement vector required to align the datasets for each point of the
image lattice. More physical models have also been broadly investigated to mathematically characterize how
tissues may move; common examples are fluid[45], diffusion[46] and elastic models[47]. The complexity of
non-rigid transformations with respect to the rigid ones is greatly increased from both the theoretical and
computation point of view, but rapid progresses have been made in the field and many applications benefit
from the higher accuracy granted by these types of models[48].

The similarity metric quantifies the degree of alignment between two given images; during the registration
process, which usually employs iterative algorithms, the similarity measure increases until a maximum is
reached. A large variety of metrics have been proposed, mostly because the quantification of “matching”
depends on the type of information that will be used to actually align the data. Two main approaches are
available: feature-based and voxel-based. In the feature-based approach a pre-processing step is required to
extract from the images anatomic landmarks or geometric characteristics, such as lines or edges or organ
contours, and the similarity measure quantifies the “distance” between corresponding landmarks or
surfaces[49]. Among the most used numerical techniques are the iterative closest point (ICP) algorithm for
matching clouds of points[50] and the “head and hat” method. The head and hat method was specifically
applied to intermodality registration of a high-resolution anatomical dataset with a low-resolution functional
one[51], as it commonly occurs in image fusion for CAD assessment[52],[53]. The great advantage of
feature-based approach is its suitability for both intra and intermodality registration, but requires a
preliminary segmentation step tailored to each modality and high accuracy to avoid segmentation errors to
adversely affect the registration results. Voxel-based approaches directly use the intensity values from the
two datasets: in case of intramodality registration, they directly compare the datasets voxel by voxel, or
compute the sum of intensity differences (SSD[45]) or the normalized cross correlation (NCC[45]). For the
more recurrent multimodality registration, techniques from the information theory framework have been
applied: image registration should in fact align images so that the content of shared information is
maximized in the combined image. To this task, metrics to quantify information in a given image have been
used, such as Shannon-Wiener entropy and the joint entropy measures[54], mutual information[55] (MI) and
normalized mutual information[56]. Methods maximizing MI have shown the most promising results[57],
[58]: MI is a measure of how well one image can “explain” the other in terms of the information it contains;
no assumption is made on image intensities, which makes it particularly effective in multimodality imaging.

Once the transformation model and the similarity metric have been defined, an optimization method must be
selected. An optimization method is numerical routine that iteratively searches the space of the
transformation parameters, evaluates the agreement between the transformed image and the reference one at
that particular step and stops when some predefined level of matching has been reached. The computational
burden can become easily cumbersome depending on the size of the images, the number of the
transformation parameters and the method selected. Many different algorithms are available ranging in
speed and reliability: Powell's method, simplex algorithm, gradient descent method and least squares
approaches[59]. To improve both speed and accuracy of the selected algorithm, the coarse-to-fine approach
is often selected, where instead of the original images, coarser ones are initially used to obtain an optimal
transform; then the resolution is progressively refined and the previously identified transform used as an
initialization for the new registration process.

Go to:

IMAGE FUSION IMPEMENTATIONS

Over the last decade, many studies and research groups have tried, tested and validated different image
fusion procedures in the context of clinical CAD assessment and management.

The initial attempts to implement image fusion and test its potential appear in the works of Pfeifer et al.[60]
and Faber et al.[61] and they coincide with the ability of reconstructing 3D models of the coronary trees and
of the LV epicardial surface from respectively biplane CA and perfusion studies. Given that the two models
are semi-automatically obtained[62],[63], Faber and colleagues describe a technique to unify them based on
an optimization approach. A cost function consisting in the distance of the LAD to the interventricular
groove and the left circumflex artery (LCX) to the atrioventricular groove is minimized using a conjugate
gradient technique. The method was validated in an animal study and showed an excellent spatial
correspondence between stenosed artery and abnormally perfused areas. Schindler and coworkers[64]
designed a methodology to retrieve the 3D coronary tree from two angiographic views specifically selected
to allow optimal visualization of the vessels and minimization of foreshortening. The vessels were manually
segmented by following their course on the images, placing fiduciary points and interpolating the final paths
with Bezier curves. After an initial manual masking to remove extra cardiac activity, automatic
quantification of MPI-SPECT was performed to obtain a triangulated surface of the left ventricular (LV)
epicardium; the surface was color-coded with the perfusion information. The fusion procedure was
performed in three successive steps the angiographic data were transformed to the MPI coordinate system;
information about the calibration of the angiographic images allowed the coronary tree to be scaled to match
the size of the ventricular surface; a final 3D translation matched the coronary tree to the left ventricle
surface. The authors determined that in 74% of the cases the stenosis manually identified by an expert user
on the CA corresponded to perfusion defects. In a more recent paper, Faber et al.[65] analogously extracted
the coronary tree anatomy from CA and registered it to MPI-SPECT. The nuclear image analysis was
automatically performed by means of the ECTb[23], a software package for objective and reliable
quantification of MPI images that allow the extraction, among many other features, of the LV epicardial
surfaces. The angiographic images were manually segmented with an ad hoc software tool to extract the
coronaries centerline; 3D models of the coronary trees were eventually created with either constant radius or
by applying edge-detection techniques on the original images once the centerlines have been delineated. The
fusion process envisaged three phases: an initial rough alignment of the coronary tree with the SPECT
image, followed by a more refined one and a final non-linear warping of the coronary models to adapt to the
LV surface. The registration was performed by initially aligning the left anterior descending artery (LAD)
and the posterior descending artery (PDA) with the anterior and posterior interventricular grooves,
respectively[66], and successively applying the ICP algorithm to further improve the alignment. In the final
step, the distance between each coronary point and the LV surface was computed and the coronary points
translated by the same distance onto the surface. As a validation of the complete procedure, two areas were
computed on the LV surface: one from the quantification of the perfusion abnormalities - the myocardial
‘blacked-out’ area - and one to estimate the myocardial area at risk due to the presence of the blockage; good
agreement in the overlap of these areas confirmed the accuracy of the registration procedure. In Fig. 3-E, an
example of the fused display obtained with this procedure is shown. This possibility of performing a
quantitative image fusion can be an additional asset to the physicians' understanding of difficult cases and
their confidence in making therapeutical decisions as compared to more qualitative approaches that mostly
rely on efficient visualization techniques of both datasets[67].

The previous examples are mostly application of feature-based registration. In this approach, the original
images are pre-processed to segment objects or identify anatomical landmarks that are in turn used to align
the datasets. Although the coronary segmentation may require some adjustments, the general approach may
be applied to non-invasive CTA acquisitions with relatively small changes.

A voxel-based approach was instead developed by Slomka et al.[68] for the automated registration of CTA
and SPECT images. Pre-processing of the original images was still performed: coronary tree anatomy was
extracted using vendors' software tools in a semi-automated fashion; LV myocardium and blood pool were
automatically identified in MPI images and new volumetric dataset with masked myocardium and blood
pool was created; the CTA image was smoothed with a Gaussian filter to remove noise. The whole images -
CTA and preprocessed MPI - both oriented along the transverse axis, were used as inputs to the registration
framework. The rigid transformation was applied, the SSD was used as a similarity metric and gradient
descent as optimization method. To avoid unreasonable solutions and speed the process the coarse-to-fine
approach was implemented. The final optimal transformation was eventually used to register the coronary
tree on the MPI surfaces.

In all these implementations, rigid transformations were used to control the registration process. Since we
are dealing with the beating heart and in many situations with gated MPI acquisitions, this may lead to
inaccurate results. Woo and coworkers[69] developed a registration method to account for phase mismatch
between CTA and SPECT. A fully automated segmentation of LV myocardium and blood pool was
performed for all gated datasets and new masked volumetric datasets created. All new MPI volumes were
rigidly registered to the CT volume - after Gaussian smoothing. This process is followed by an energy
functional with a piecewise constant image model for the three segmented regions, (myocardium, blood pool
and extra cardiac structures) was constructed and minimized by means of gradient descent algorithm using a
region-based SSD as similarity measure. The frame that best matched the CT was identified and all the
remaining frames were warped by means of thin-plate spline algorithms and summed to create a better static
image with less motion artifact. This was finally fused with the CTA and the CTA-derived coronary
anatomy.

Image registration algorithms should take advantage of all the available information present in the input
images. Given the characteristics of the MPI images, the majority of algorithms employs only the LV
position and orientation to guide the registration process. In an article by Faber et al.[70] the right ventricle
(RV) was also included in a voxel-based registration algorithm to fuse CTA and SPECT acquisitions. Since
multimodality image registration is particularly challenging due to the high differences in image intensities
and features, in this work, pre-processing steps were applied to the CTA volume to make it look more
similar to the MPI so that mutual information could be used to guide more reliably the alignment. CTA
images were manually segmented by means of an in-house software program: endocardial and epicardial
borders for both LV and RV delineated. The LV chamber was automatically identified and reoriented into
short axis slices for both CTA and SPECT[23]. The two LV volumetric masks were rigidly registered by
maximizing their product over all possible 3D rotations and translations. This initial transformation was then
applied to the full CTA binary mask, smoothed with a Gaussian filter and multiplied to the original MPI
image to remove all SPECT extra cardiac activity. The final registration step maximized mutual information
between masked MPI and smoothed CTA binary images and mapped the SPECT image to the CTA short
axis space; the presence of both RV and LV in the input image should reduce problems of symmetry often
encountered when only the LV is used. Should CTA-derived coronary anatomy be available, it would be
directly fused on the MPI images and surfaces. Fig. 4 shows one case example of image fusion obtained
with this methodology.
Fig. 4

One case example of image fusion based on Faber et al.[70] performed with ECTb[23].

As more information is included into the modeling, such as anatomy or tissue deformation, the complexity
of the algorithms and numerical techniques required to solve the registration problem increases consistently.
While the accuracy of the integration may benefit from the more advanced approaches and their application
is greatly desired, the final translation into a clinical environment may be affected in terms of computational
requirements and time constraints.

Go to:

LIMITATIONS
Although significant amount of work has been carried out in the specific field of multimodality image
registration for CAD assessment and its additional diagnostic value is widely accepted and demonstrated,
image fusion techniques have not completely been translated into clinical routine. The main reason resides
in the fact that important parts of the registration techniques proposed and even successfully validated still
require intensive manual interaction. This is particularly true for the anatomical information extraction.
Differently from nuclear cardiology that has witnessed and benefited from the development of robust
software tools for image quantification, automatic segmentation of the coronary artery tree, the myocardium
and even the blood pools from anatomical acquisitions are far from established. Additional work will be
needed in the image processing field to allow a straightforward and seamless inclusion of image fusion into
the routinely used clinical tools.

Hybrid PET/CT and SPECT/CT scanners are considered promising solutions to these issues since both
functional and anatomical datasets are acquired at the same moment and within the same device no
registration is in theory needed. In practice, respiratory and heart motion artifacts will still be present and
require software approaches to be controlled and minimized. Additionally, while allowing immediate image
fusion and direct visualization, hybrid systems will lack the strength of quantitative evaluations that greatly
increase image fusion potentiality to solve borderline cases.

A crucial limitation for the multimodality fusion of stand-alone acquisitions is the cumulative radiation dose
the patient would receive with the sequential imaging. Although this is still under debate, new devices and
new protocols in both the computed tomography and nuclear cardiology fields are continuously being tested
with the final goal of reducing radiation while still maintaining efficiency and accuracy.

Go to:

CONCLUSIONS
Clinical evidence suggests that image fusion can be a reliable and useful tool in the hands of clinicians for a
more accurate diagnosis of CAD, specifically for ambiguous and borderline cases. Additional technical
developments in the extraction of anatomical information are still in order for the whole procedure to
become fully automated and enter the clinical practice.

Go to:

References
1. Go AS, Mozaffarian D, Roger VL, Benjamin EJ, Berry JD, Borden WB, et al. Heart Disease and Stroke Statistics - 2013
Update: A Report From the American Heart Association. Circulation. 2013;127:e6–e245. [PMC free article] [PubMed]

2. Roger VL, Go AS, Lloyd-Jones DM, Adams RJ, Berry JD, Brown TM, et al. Heart Disease and Stroke Statistics - 2011
Update: A report From the American Heart Association. Circulation. 2011;123:e18–e209. [PMC free article] [PubMed]

3. Heidenreich PA, Trogdon JG, Khavjou OA, Butler J, Dracup K, Ezekowitz MD, et al. Forecasting the Future of
Cardiovascular Disease in the United States: a policy statement from the American Heart Association. Circulation.
2011;123:933–44. [PubMed]

4. Kramer CM. Integrated approach to ischemic heart disease. The one-stop shop. Cardiac Magnetic Resonance Imag.
1998;16:267–76. [PubMed]

5. Santana CA, Garcia EV, Faber TL, Sirineni GKR, Esteves FP, Sanyal R, et al. Diagnostic performance of fusion of
myocardial perfusion imaging (MPI) and computed tomography coronary angiography. J Nucl Cardiol. 2009;16:201–
11. [PMC free article] [PubMed]

6. Budoff MJ, Dowe D, Jollis JG, Gitter M, Sutherland J, Halamert E, et al. Diagnostic performance of 64-multidetector
row coronary computed tomographic angiography for evaluation of coronary artery stenosis in individuals without
known coronary artery disease: results from the prospective multicenter ACCURACY (Assessment by Coronary
Computed Tomographic Angiography of Individuals Undergoing Invasive Coronary Angiography) trial. J Am Coll
Cardiol. 2008;52:1724–32. [PubMed]

7. Meijboom WB, Meijs MF, Schuijf JD, Cramer MJ, Mollet NR, van Mieghem CA, et al. Diagnostic accuracy of 64-slice
computed tomography coronary angiography: a prospective, multicenter, multivendor study. J Am Coll Cardiol.
2008;52:2135–44. [PubMed]

8. Miller JM, Rochitte CE, Dewey M, Arbab-Zadeh A, Niinuma H, Gottlieb I, et al. Diagnostic performance of coronary
angiography by 64-row CT. N Engl J Med. 2008;359:2324–36. [PubMed]

9. Di Carli MF, Hachamovitch R. New technology for noninvasive evaluation of coronary artery disease. Circulation.
2007;115:1464–80. [PubMed]

10. Blankstein R, Di Carli MF. Integration of coronary anatomy and myocardial perfusion imaging. Nat Rev Cardiol.
2010;7:226–36. [PubMed]

11. Beller GA. Recent advances and future trends in multimodality cardiac imaging. Heart Lung Circ. 2010;19:193–
209. [PubMed]

12. Achenbach S, Friedrich MG, Nagel E, Kramer CM, Kaufmann PA, Farkhooy A, et al. CV Imaging: what's new in
2012? J Am Coll Cardiol Img. 2013;6:714–34.

13. Litt HI, Gatsonis C, Snyder B, Singh H, Miller CD, Entrikin DW, et al. CT Angiography for safe discharge of patients
with possible acute coronary syndromes. N Engl J Med. 2012;366:1393–403. [PubMed]

14. Gaemperli O, Schepis T, Valenta I, Koepli P, Husmann L, Scheffel H, et al. Functionally relevant coronary artery
disease: comparison of 64-section CT angiography with myocardial perfusion SPECT. Radiology. 2008;248:414–23.
[PubMed]

15. Hacker M, Jakobs T, Hack N, Nikolaou K, Becker C, von Ziegler F, et al. Sixty-four slice spiral CT angiography does
not predict the functional relevance of coronary artery stenoses in patients with stable angina. Eur J Nucl Med Mol
Imaging. 2007;34:4–10. [PubMed]

16. Rispler S, Keidar Z, Ghersin E, Roguin A, Soil A, Dragu R, et al. Integrated single-photon emission computed
tomography and computed tomography coronary angiography for the assessment of hemodynamically significant
coronary artery lesions. J Am Coll Cardiol. 2007;49:1059–67. [PubMed]

17. Di Carli MF, Dorbala S, Curillova Z, Kwong RJ, Goldhaber SZ, Rybicki FJ, et al. Relationship between CT coronary
angiography and stress perfusion imaging in patients with suspected ischemic heart disease assessed by integrated
PET-CT imaging. J Nucl Cardiol. 2007;14:799–809. [PubMed]
18. Hachamovitch R, Hayes SW, Friedman JD, Cohen I, Berman DS. Comparison of the short-term survival benefit
associated with revascularization compared with medical therapy in patients with no prior coronary artery disease
undergoing stress myocardial perfusion single photon emission computed tomography. Circulation. 2003;107:2900–
7. [PubMed]

19. Min JK, Berman DS, Dunning A, Achenbach S, Al-Mallah M, Budoff MJ, et al. All-cause mortality benefit of
coronary revascularization vs. medical therapy in patients without known coronary artery disease undergoing
coronary computed tomographic angiography: results from CONFIRM (COronary CT Angiography EvaluatioN For
Clinical Outcomes: An InteRnational Multicenter Registry) Eur Heart J. 2012;33:3088–97. [PMC free article] [PubMed]

20. McCullough CH, Primak AN, Braun N, Koefler J, Yu L, Christner J. Strategies for reducing radiation dose in CT.
Radiol Clin North Am. 2009;47:27–40. [PMC free article] [PubMed]

21. Garcia EV. Quantitative nuclear cardiology: we are almost there! J Nucl Cardiol. 2012;19:424–37. [PubMed]

22. Klocke FJ, et al. ACC/AHA/ASNC guidelines for the clinical use of cardiac radionuclide imaging-executive summary:
a report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines
(ACC/AHA/ASNC committee to revise the 1995 guidelines for the clinical use of cardiac radionuclide imaging)
Circulation. 2003;108:1404–18. [PubMed]

23. Garcia EV, Faber TL, Cooke CD, Folks RD, Chen J, Santana C. The increasing role of quantification in nuclear
cardiology: The Emory approach. J Nucl Cardiol. 2007;14:420–32. [PubMed]

24. Germano G, Kavanagh PB, Slomka PJ, Van Kriekinge SD, Pollard G, Berman DS. Quantitation in gated perfusion
SPECT imaging: The Cedars-Sinai approach. J Nucl Cardiol. 2007;14:433–54. [PubMed]

25. Ficaro EP, Lee BC, Kritzman JN, Corbett JR. Corridor4DM: The Michigan method for quantitative nuclear
cardiology. J Nucl Cardiol. 2007;14:455–65. [PubMed]

26. Underwood SR, et al. Myocardial perfusion scintigraphy: the evidence. Eur J Nucl Med Mol Imaging.
2004;31:261–91. [PMC free article] [PubMed]

27. Berman DS, Hachamovitch R. Risk assessment in patients with stable coronary artery disease: incremental value
of nuclear imaging. J Nucl Cardiol. 1996;3(S):41–9. [PubMed]

28. Hachamovitch R, Di Carli MF. Methods and limitations of assessing new noninvasive tests. Part II: outcomes-based
validation and reliability assessment of noninvasive testing. Circulation. 2008;117:2793–801. [PubMed]

29. Hachamovitch R, Hayes SW, Friedman JD, Cohen I, Berman DS. A prognostic score for prediction of cardiac
mortality risk after adenosine stress myocardial perfusion scintigraphy. J Am Coll Cardiol. 2005;45:722–9. [PubMed]

30. Hachamovitch R, Hayes S, Friedman JD, Cohen I, Shaw LJ, Germano G, Berman DS. Determinants of risk and its
temporal variation in patients with normal stress myocardial perfusion scans: what is the warranty period of a normal
scan? J Am Coll Cardiol. 2003;41:1329–40. [PubMed]

31. Cerqueira MD, Weissman NJ, Dilsizian V, et al. Standardized myocardial segmentation and nomenclature for
tomographic imaging of the heart: a statement for healthcare professionals from the Cardiac Imaging Committee of
the Council on Clinical Cardiology of the American Heart Association. Circulation. 2002;105:539–42. [PubMed]

32. Kalbfleisch H, Hort W. Quantitative study on the size of coronary artery supplying areas postmortem. Am Heart J.
1977:183–8. [PubMed]

33. Sheth T, Yusuf S. Enhancing risk prediction with PET coronary flow reserve. J Am Coll Cardiol Img. 2012;5:1035–6.
[PubMed]

34. Fiechter M, Ghadri JR, Gebhard C, Fuchs TA, Pazhenkottil AP, Nkoulou RN, Herzog BA, Wyss CA, Gaemperli O,
Kaufmann PA. Diagnostic value of 13N- ammonia myocardial perfusion PET: added value of myocardial flow reserve. J
Nucl Med. 2012;53:1230–4. [PubMed]

35. Murthy VL, Naya M, Foster CR, et al. Association between coronary vascular dysfunction and cardiac mortality in
patients with and without diabetes mellitus. Circulation. 2012;126:1858–68. [PMC free article] [PubMed]
36. Gaemperli O, Husmann L, Schepis T, Koepfli P, Valenta I, Jenni W, et al. Coronary CT angiography and myocardial
perfusion imaging to detect flow-limiting stenoses: a potential gatekeeper for coronary revascularization? Eur Heart J.
2009;30:2921–9. [PubMed]

37. Rispler S, Keidar Z, Ghersin E, Roguin A, Soil A, Dragu R, et al. Integrated single-photon emission computed
tomography and computed tomography coronary angiography for the assessment of hemodynamically significant
coronary artery lesions. J Am Coll Cardiol. 2007;49:1059–67. [PubMed]

38. Garcia EV, Gropler RJ. Eighth nuclear cardiology invitational conference Annapolis, Maryland, 2008. J Nucl Cardiol.
2008;15(6):e37–50. [PubMed]

39. Gaemperli O, Schepis T, Valenta I, Husmann L, Scheffel H, Duerst V, et al. Cardiac image fusion fro stand-alone
SPECT and CT: clinical experience. J Nucl Med. 2007;48:696–703. [PubMed]

40. Smolka PJ, Baum RP. Multimodality image registration with software: state of the art. Eur J Nucl Med Mol
Imaging. 2009;1(S):44–55. [PubMed]

41. Slomka PJ. Software approach to merging molecular with anatomic information. J Nucl Med. 2004;45(S):36–45.
[PubMed]

42. Hutton BF, Braun M, Thurfjell L, Lau DYH. Image registration: an essential tool for nuclear medicine. Eur J Nucl
Med. 2002;29:559–77. [PubMed]

43. Maintz JB, Viergenev MA. A survey of medical image registration. Med Image Anal. 1998;2:1–36. [PubMed]

44. Hill DLG, Batchelor PG, Holden M, Hawkes DJ. Medical image registration. Phys Med Biol. 2001;46:R1–R45.
[PubMed]

45. Beg MF, Miller MI, Younes ATL. Computing metrics via geodesics on flows of diffeo-morphisms. Int J Comput Vis.
2005;61:139–57.

46. Thirion JP. Image matching as a diffusion process: An analogy with Maxwell's demons. Med Image Anal.
1998;2:243–60. [PubMed]

47. Bajcsy R, Kovacic S. Multiresolution elastic matching. Comput Vis Graph Image Process. 1989;46:1–21.

48. Rueckert D, Aljabar P. Nonrigid registration of medical images: theory, methods and applications. IEEE Signal
Process Mag. 2010;7:113–9.

49. Audette MA, Ferrie FP, Peters TM. An algorithmic overview of surface registration techniques for medical imaging.
Med Image Anal. 2000;4:201–217. [PubMed]

50. Besl PJ, McKay ND. A method for registration of 3D shapes. IEEE Trans PAMI. 1992;14:239–56.

51. Pelizzari CA, Chen GTY, Spelbring DR, Weichselbaum RR, Chen CT. Accurate three-dimensional registration of CT,
PET and/or MR images of the brain. J Comput Assist Tomogr. 1989;13:20–6. [PubMed]

52. Faber TL, McColl RW, Opperman RM, Corbett JR, Peshock RM. Spatial and temporal registration of cardiac SPECT
and MR images: methods and evaluation. Radiology. 1991;179:857–61. [PubMed]

53. Scott AM, Macapinlac HA, Divgi CR, Zhang JJ, Kalaigian H, Pentlow K, et al. Clinical validation of SPECT and CT/MRI
image registration in radiolabeled monoclonal antibody studies of colorectal carcinoma. J Nucl Med. 1994;35:1976–
84. [PubMed]

54. Shannon CE. The mathematical theory of communication (Part I and II) Bell Syst Tech J. 1948;27

55. Wells WM, Viola P, Atsumi H, Nakajima S, Kikinis R. Multi-modal volume registration by maximization of mutual
information. Med Image Anal. 1996;1:35–51. [PubMed]

56. Studholme C, Hill DL, Hawkes DJ. Automated 3-D registration of MR and CT images of the head. Med Image Anal.
1996;1:163–175. [PubMed]
57. Maes F, Collignon A, Vandermeulen D, Marchal G, Suetens P. Multimodality image registration by maximization of
mutual information. IEEE Trans Med Imaging. 1997;16:187–98. [PubMed]

58. Pluim JP, Maintz JB, Viergever MA. Mutual-information-based registration of medical images: a survey. IEEE Trans
Med Imaging. 2003;22:986–1004. [PubMed]

59. Press WH, Teukolsky SA, Vetterling WT, Flannery BP. 3nd edn. Cambridge: Cambridge University Press; 2007.
Numerical recipes in C++

60. Peifer JW, Garcia EV, Cooke CD, Klein JL, Folks R, Ezquerra NF. 3-D Registration and Visualization of Reconstructed
Coronary Arterial Trees on Myocardial Perfusion Distributions. Visualization in Biomedical Computing, Richard A Robb
Editor Proc, SPIE. 1992;1808:225–34.

61. Faber TL, Klein JL, Folks RD, Hoff JG, Euhevg JW, Malveaux EJ, et al. Automatic Unification of Three-Dimensional
Cardiac Perfusion with Three-dimensional Coronary Artery Anatomy. Computers in Cardiology. 1996;0276-6547:333–
6.

62. Garcia EV, Cooke CD, Van Train KF, Folks R, Peifer J, DePuey EG, et al. Technical aspects of myocardial SPECT
imaging with technetium-99m sestamibi. Am J Cardiol. 1990;190:23E–31E. [PubMed]

63. Peifer JW, Ezquerra NF, Cooke CD, Mullick R, Klein L, Hyche ME, et al. Visualization of multimodality cardiac
imagery. IEEE Trans Biomed Eng. 1990;37:744–56. [PubMed]

64. Schindler TH, Magosaki N, Jeserich M, Oser U, Krause T, Fischer R, et al. Fusion imaging: combined visualization of
3D reconstructed coronary artery tree and 3D myocardial scintigraphic image in coronary artery disease. Int J Card
Imaging. 1999;15:357–68. [PubMed]

65. Faber TL, Santana CA, Garcia EV, Candell-Riera J, Folks RD, Peifer JW, et al. Three-dimensional fusion of coronary
arteries with myocardial perfusion distribution: clinical validation. J Nucl Med. 2004;45:745–53. [PubMed]

66. Faber TL, Chiron F, Ezquerra NF, et al. Registration of multimodal 3D cardiac information using the iterative closest
point approach. In: Wilson DC, Tagare HD, Bookstein FL, et al., editors. Mathematical Modeling, Estimation, and
Imaging. Bellingham, WA: SPIE; 2000. pp. 233–41. Proceedings SPIE 4121.

67. Gaemperli O, Schepis T, Kalff V, Namdar M, Valenta I, Stefani L, et al. Validation of a new cardiac image fusion
software for three-dimensional integration of myocardial perfusion SPECT and stand-alone 64-slice CT angiography.
Eur J Nucl Med Mol Imaging. 2007;34:1097–106. [PubMed]

68. Slomka PJ, Cheng VY, Dey D, Woo J, Ramesh A, van Kriekinge S, et al. Quantitative analysis of myocardial perfusion
SPECT anatomically guided by coregistered 64-slice coronary CT angiography. J Nucl Med. 2009;50:1621–30. [PMC
free article] [PubMed]

69. Woo J, Slomka PJ, Dey D, Cheng VY, Hong BW, Ramesh A, et al. Geometric feature-based multimodal image
registration of contrast-enhanced cardiac CT with gated myocardial perfusion SPECT. Med Phys. 2009;36:5467–79.
[PMC free article] [PubMed]

70. Faber TL, Santana CA, Piccinelli M, Nye JA, Votaw JR, Garcia EV, et al. Automatic alignment of myocardial
perfusion images with contrast-enhanced cardiac computed tomography. IEEE Trans Nuclear Science. 2011;58:2296–
302. [PMC free article] [PubMed]