Magnetic Resonance Imaging in Medicine
Magnetic Resonance Imaging in Medicine
Magnetic Resonance Imaging in Medicine
www.iop.org/Journals/PhysEd
Abstract
Over the past twenty years, magnetic resonance imaging (MRI) has become
one of the most important imaging modalities available to clinical medicine.
It offers great technical flexibility, and is free of the hazards associated with
ionizing radiation. In addition to its role as a routine imaging technique with
a growing range of clinical applications, the pace of development in MRI
methodology remains high, and new ideas with significant potential emerge
on a regular basis. MRI is a prime example of the spin-off benefits of basic
science, and is an area of medicine in which physical science continues to
play a major role, both in supporting clinical applications and in developing
new techniques. This article presents a brief history of MRI and an overview
of the underlying physics, followed by a short survey of current and
emerging clinical applications.
Figure 2. Possible orientations of a nucleus with I = 21 in a static magnetic field. The size of the magnetic
moment is well defined, and only two values of the z-component are possible—corresponding to alignment with
or against the applied magnetic field. Nuclei precess about the z-axis, and the x and y components of the moment
of a given nucleus at a given time are random.
Figure 3. Origin of the bulk magnetization vector in a macroscopic sample. Excess magnetic moments aligned in
the positive z-direction sum to give M , while the random x and y components cancel. The excess of a few nuclei
per million has been greatly exaggerated in this diagram.
where B0 is the flux density of the applied electromagnetic waves. The effect of this
magnetic field and the gyromagnetic ratio, γ , irradiation can be described in either quantum
is characteristic of the nucleus (for hydrogen, mechanical or classical terms. In the quantum
γ = 42.57 MHz T−1 , so in a magnetic field model, if the energy of the RF photons is equal
of strength 1 T hydrogen nuclei precess at a to the difference between the two energy levels,
rate of 42.57 million revolutions per second). nuclei in the lower level may gain energy by
Thus the z-component of the magnetic moment absorbing photons and those in the higher level
is well-defined and may adopt one of only two may be induced to emit photons and so lose energy.
values while, in the classical model, the x These processes cause nuclei to flip between
and y components vary rapidly (ω0 lies in the the two orientations, modifying the population
range from megahertz to hundreds of megahertz) distribution and hence the magnitude and direction
and the position (or phase) of the magnetic of M .
moment on the precessional cone at a given time However, considering the colossal number
is random. This can also be understood in of nuclei present in a macroscopic sample, it is
quantum mechanical terms as a manifestation of convenient to adopt an entirely classical model.
Heisenberg’s uncertainty principle. In this approach, NMR is described in terms of
Spins in the two orientations have different interactions between the bulk magnetization and
energies, the difference being E = h̄ω0 , where the magnetic field component of the RF wave, B1 .
h̄ = h/2π (h is Planck’s constant). Because If the RF frequency, and hence the frequency at
of this, the populations of the two states differ which B1 rotates in the xy plane, matches the
slightly, with a few more nuclei per million precessional frequency of the magnetic moments,
aligned with B0 than against. This is sufficient ω0 , a resonant condition is achieved in which M
to endow the sample as a whole with a small bulk tips, or nutates, from the z-axis towards the xy
magnetization, M , lying along the positive z-axis plane. Nutation continues until B1 is removed, so
(see figure 3). There is no net magnetization in the amplitude and duration of B1 can be tailored to
the xy plane because of the random phases of the nutate M through specific angles. An important
moments. special case is that in which M is nutated entirely
The next step in the NMR experiment is into the transverse plane, leaving no magnetization
to irradiate the sample with radiofrequency (RF) along the z-axis. An RF wave of sufficient duration
Figure 4. Special cases of an RF pulse: (a) 90◦ pulse, (b) 180◦ inversion pulse, (c) 180◦ refocusing pulse.
Rotation of B1 and precession of M about B0 have been omitted for clarity.
Table 1. Relaxation times of water protons in brain tissues measured in a healthy volunteer at 1.5 T (courtesy of
Dr Marcus Newbold, GKT School of Medicine, King’s College London).
Tissue T1 (mean ± SD) (ms) T2 (mean ± SD) (ms)
Grey matter 1078.5 ± 52.6 100.4 ± 12.3
White matter 684.2 ± 20.8 79.2 ± 5.0
Cerebrospinal fluid (CSF) 3958.9 ± 305.5 914.6 ± 422.4
and amplitude to achieve this is known, for obvious precessing transverse magnetization, Mxy :
reasons, as a 90◦ pulse. Similarly, a 180◦ pulse
inverts M or, if applied after a 90◦ pulse, rotates Mz = M(1 − e−t/T1 )
M through 180◦ in the transverse plane (figure 4).
These pulses are the basic ‘building blocks’ of Mxy = Me−t/T2 .
NMR, and by stringing them together in different Thus the NMR signal detected by the RF coil
combinations an essentially limitless range of rapidly decays due to T2 relaxation. Relaxation
experiments can be performed. behaviour is strongly dependent on the physico-
Whichever model is adopted, the RF chemical environment of the nucleus, and hence
frequency required to induce NMR is equal to the on the tissue type in which the nuclei are located,
nuclear Larmor frequency, ω0 = γ B0 . and this is an important source of image contrast
A 90◦ pulse generates net transverse magneti- in biomedical MRI. Table 1 shows typical in vivo
zation, which precesses about the direction of B0 . relaxation time values for brain tissues.
This precessing magnetization can be detected by
means of electromagnetic induction, by placing a From NMR to MRI
tuned antenna close to the sample. In the context The NMR experiment described thus far results in
of biomedical NMR, such an antenna is usually acquisition of an undifferentiated signal, with no
known as an RF coil, and similar coils are used to means of incorporating spatial information so as
transmit RF and to detect the emitted NMR signal. to produce an image. Study of a specific tissue can
Following excitation, nuclei return to their only be achieved by excising the tissue from the
initial population distribution via a variety of body, or—if the tissue is superficial—by placing a
relaxation processes, categorized according to small RF coil, known as a surface coil, against the
whether they cause loss of energy from the body.
spin system or simply exchange of energy The method of spatial encoding used
between spins. These two phenomena are universally in MRI is due to Lauterbur. As
known as spin–lattice and spin–spin relaxation, with many of the great ideas of science, it is at
and are characterized by the relaxation times once simple and extremely powerful. Temporary
T1 and T2 , respectively. T1 relaxation results imposition of an additional static magnetic field,
in exponential recovery of z-magnetization, Mz , which lies parallel to B0 but varies linearly in
while T2 processes cause exponential decay of the strength with position along the x, y or z-axis
Figure 5. Effect of a magnetic field gradient along the Figure 6. Use of a magnetic field gradient for selective
x-axis. (a) Without the gradient, B is independent of excitation of a transaxial slice. If a gradient is applied
x. (b) In the presence of the gradient, B varies linearly along the z-direction together with an RF pulse of
with x. Note that the direction of the B vector is the bandwidth δω magnetization will be excited within a
same in both cases; only the amplitude varies in case slice of thickness δz = δω/γ Gz . If performance is
(b). ideal, magnetization outside this slice will be left along
the z-axis and will produce no signal.
(or indeed some direction oblique to these axes)
(figure 5) results in a linear variation in the Larmor
frequency as a function of position:
ω(x) = γ B = γ (B0 + Gx x)
where the strength of the gradient (in mT m−1 ) is
Gx (in this case applied along the x-axis). This
frequency variation can be used in three ways to
achieve spatial localization.
Figure 7. Frequency encoding. If a gradient is applied
1. Slice selection. If the gradient is applied at during signal acquisition, the resulting NMR signal
the same time as the initial RF pulse, and the will contain components with a range of frequencies.
pulse is modified to contain a narrow band of Fourier transformation of this signal yields a projection
frequencies rather than a single frequency, a through the object. A greatly simplified object is
thin slice of spins can be selectively excited. shown in this diagram.
Using this approach, MRI becomes a two-
dimensional tomographic imaging technique 3. Phase encoding. By applying a gradient for
(figure 6). a short time between excitation and signal
2. Frequency encoding. If the gradient acquisition, and repeating the experiment
is applied during signal acquisition, the a number of times with different gradient
resonance frequency of spins contributing amplitudes, it is possible to generate a set
to the signal becomes a linear function of of data that may be Fourier transformed to
their position along the gradient direction. yield spatial information along the direction
The signal can be decomposed using Fourier perpendicular to the frequency encoding axis.
methods to determine the amount of signal at Phase encoding is one of the most difficult
each frequency and hence at each position, concepts in MRI, and a full explanation is
resulting in a projection through the sample beyond the scope of this article. Suffice
(figure 7). it to say that the process is mathematically
At this point, it is perhaps worth eliminating equivalent to frequency encoding, and is
a common misconception. Many people who really a ‘trick’ to get around the fact that
are learning about MRI imagine that images are frequency encoding cannot be performed
formed by scanning line by line through the along two axes simultaneously.
object, obtaining signal from each point in space
sequentially. We can see from the preceding In conventional MRI, all three methods are
description that this is not the case. The signal applied to generate multiple two-dimensional
acquired originates from the entire object under slices through the object or patient under study.
study, and spatial information is obtained only To allow time for this process, acquisition of the
by breaking that signal down into its different NMR signal is delayed for up to a few hundred
frequency components. milliseconds after excitation. This is achieved by
(a) (b)
Figure 8. Flexibility of imaging plane and image weighting in MRI. (a) Transaxial T1 -weighted image of the
head at the level of the orbits. There is good soft tissue contrast, and fluid-filled structures, such as the eyes and
the ventricles in the brain, appear dark. (b) Heavily T2 -weighted parasagittal image of the head. Note the intense
signal from CSF around the surface of the brain and from the contents of the eye, except for the dark lens
(pictures courtesy of Guy’s and St Thomas’ NHS Trust).
using magnetic field gradients and RF pulses to acquisition of an echo, magnetization in different
produce what is known as an echo at the desired parts of the sample will be undergoing T2 decay
time. Unlike x-ray CT, MRI can image slices in at different rates depending on its environment
any desired plane by appropriate use of gradients. (i.e. the tissue in which it is located). Thus by
Alternatively, the technique can be modified to varying this interval, known as the echo time, TE ,
produce three-dimensional volumes of data that it is possible to vary the degree of T2 weighting—
can be viewed using special software or post- i.e. the extent to which differences in T2 affect
processed to generate arbitrary slices. the appearance of the image. In addition, as
The images generated in MRI are made noted above it is necessary to repeat acquisition
up of voxels (three-dimensional pixels), and the a number of times to allow phase encoding.
brightness of each voxel is dependent on the During the interval between repetitions, known as
intensity of signal from the corresponding location the repetition time, TR , magnetization undergoes
in the object. A typical image consists of 256×256 differential T1 recovery. By varying this interval,
or, increasingly, 512×512 voxels. This means that the user can alter the extent of T1 weighting.
the in-plane resolution is typically 0.5–2.0 mm. T1 and T2 weighting result in very different
The voxel is usually considerably larger in the image appearances. T1 -weighted images provide
third dimension, as the slice thickness is typically excellent soft tissue contrast, and material with
1–5 mm. For specialist (usually non-medical) long T1 , such as cerebrospinal fluid (CSF) in the
applications, NMR microscopes are available that ventricles of the brain, appears dark. This type
can achieve a resolution of a few micrometres. of imaging is often used to depict anatomical
As described so far, an MR image of the structures. On T2 -weighted images, tissues with
body is essentially a map of water distribution. increased water content appear bright. This
Such an image would be of limited clinical value, includes CSF, but also pathological processes
since water density varies relatively little between such as neoplasm, inflammation, ischaemia and
tissues. One of the major strengths of MRI is degenerative changes. Thus T2 -weighted images
the ability to manipulate image contrast. The two are often used to highlight areas of disease.
simplest ways of doing this follow readily from Typical T1 - and T2 -weighted images are shown in
the discussion above, and involve weighting the figure 8.
image according to tissue relaxation times. Many other types of image contrast may
During the interval between excitation and be introduced by appropriate adaptation of the
MRI instrumentation
The appearance of a clinical MR scanner is
dominated by the magnet. Until recent years
Figure 10. Low-field (0.2 T) open MR system using a
these were almost invariably large and forbidding, permanent magnet (picture courtesy of Siemens
with a bore that totally enclosed the patient Medical Solutions).
and frequently disappeared into the wall of the
examination room. Now, developments in magnet is needed. Newer high-temperature superconduc-
technology have led to designs that increase both tors have had a limited impact on MRI thus far, and
patient acceptance and the range of potential superconductivity is maintained by immersing the
applications. Also, whereas most MR magnets turns of the magnet in a bath of liquid helium (at
once operated at a field strength of 0.5–1.5 T, the 4.2 K). Modern superconducting MR systems are
market has now polarized between low (0.2–0.5 T) far more compact and patient-friendly than older
and high (1–3 T) field strength systems (with a designs (figure 9).
few research systems at 4–8 T). The lower field Lower field systems are usually based
systems are cheaper and, being more open and around resistive electromagnets or permanent
more patient-friendly, are arguably more suitable magnets. These are typically more open
for paediatric and interventional work. Higher than superconducting systems, allowing more
field systems produce better quality images and interaction with the patient (figure 10).
are appropriate for more specialized work in areas As well as the magnet, the performance of
such as neurology and cardiology. an MR scanner depends on the design of the
High fields are achieved using superconduct- gradient and RF systems, and increasingly on
ing electromagnets: once the magnet is on field, the performance of the computer system used to
current flows without loss and no power supply control data acquisition and image processing.
Clinical applications
Because of its excellent depiction of soft tissues,
MRI has long been the modality of choice for
examination of the central nervous system (brain
and spine), with indications including suspected
tumours, slipped disks, multiple sclerosis and
degenerative diseases (see figure 8). It now
also has a major role in orthopaedics and
musculoskeletal imaging (figure 11).
Cardiac MRI is an area of increasing
importance, including investigation of both
congenital abnormalities and acquired conditions
such as coronary heart disease. The high
resolution and contrast between tissues possible
using MRI allow image processing techniques to
be applied to extract features of interest (figure 12).
Improvements in the speed of MRI in recent Figure 13. MR angiogram showing the abdominal
years have made abdominal imaging possible. aorta and its various branches, including the renal
Figure 13 shows an MR angiogram of the kidneys, arteries that feed the kidneys with blood. The aorta
acquired using an injection of contrast agent to forks into the right and left common iliac arteries, each
highlight flowing blood. Images such as this can of which in turn divides into external and internal
branches supplying the legs and pelvis (picture
be used to investigate narrowing of the arteries due courtesy of Guy’s and St Thomas’ Hospital NHS
to fatty plaques. Trust).
Figure 14. Brain activation due to a simple right-handed motor task investigated using fMRI. The images
show activation in the part of the left hemisphere responsible for controlling the right hand (picture courtesy of
Dr Andrew Simmons, Institute of Psychiatry, King’s College London).
Functional MRI (fMRI) allows parts of the of conditions such as epilepsy and Alzheimer’s
brain involved in processing sensory data (e.g. disease, in which the relative levels of the
from the eyes and ears) or involved in motor compound visible in the spectrum are found to
tasks (such as moving the fingers) to be identified change (figure 15).
and studied. The technique uses the fact that Carrying out surgical procedures in or close
activation of part of the brain results in increased to an MR scanner might seem a dangerous or at
delivery of oxygenated blood to that area, and the best an unpromising idea. The high magnetic
magnetic properties of this blood allow the area field turns conventional surgical instruments into
to be identified using MRI (figure 14). The fMRI deadly missiles, while the restricted bore of
technique is helping to improve our understanding conventional MR systems offers little access
of the basic functioning of the brain, and in to the patient. However, the development of
individual patients it can be used to map the special tools and more open magnet geometries
function of important parts of the brain prior to has led to the development of interventional
surgery. MRI, in which minimally invasive procedures
Proton magnetic resonance spectroscopy or even open surgery can be performed inside
(MRS) uses minute differences in the resonance the MR scanner. Figure 16 shows an artist’s
frequencies of protons in different chemical impression of an interventional MR room, similar
environments to perform noninvasive chemical to a suite soon to be installed at the author’s
analysis. NMR spectroscopy considerably pre- institution. Developments like this have led some
dates MRI and has wide applications outside commentators to speculate that MR scanners may
medicine. However, MRS is now starting to have a prominent place in the operating theatre of
have a role in the diagnosis and understanding the future.