AJR 2002; 178:3-16
© American Roentgen Ray Society
Diffusion Tensor MR Imaging of the Brain and White Matter Tractography
Elias R. Melhem1,2,
Susumu Mori1,2,
Govind Mukundan1,
Michael A. Kraut1,2,
Martin G. Pomper1 and
Peter C. M. van Zijl1,2
1
Department of Radiology and Radiological Sciences, The Johns Hopkins Medical
Institutions, 600 N. Wolfe St., Baltimore, MD 21287-2182.
2
F. M. Kirby Research Center for Functional Brain Imaging, Kennedy-Krieger
Institute, The Johns Hopkins Medical Institutions, 707 N. Broadway, Baltimore,
MD 21287.
Received December 28, 2000;
accepted after revision July 5, 2001.
Address correspondence to E. R. Melhem.
Introduction
In the last 10-15 years, MR imaging techniques have been increasingly
applied to the study of molecular displacement (diffusion) in biologic tissue
[1,
2]. The ability to spatially
map the diffusion of free water protons in vivo using 1H MR imaging
and the observation that the diffusion of free water protons is reduced in
acutely infarcted brain tissue are responsible for the widespread use of these
techniques in clinical imaging
[3,4,5,6,7,8].
More recently, the dependency of molecular diffusion on the orientation of
white matter fiber tracts has elicited great interest in studying the factors
that influence this dependency and in spatially mapping these fiber tracts
using diffusion imaging
[9,10,11,12,13,14,15,16,17,18,19,20,21].
In this paper, we briefly describe the tensor theory used to characterize
molecular diffusion in white matter and how the tensor elements are measured
experimentally using diffusion-sensitive MR imaging. We then review techniques
for acquiring relatively high-resolution diffusion-sensitive MR images and
computer-based algorithms that allow the generation of white matter fiber
tract maps from the tensor data. We provide an overview of current experience
and some clinical examples that are ongoing in our center. Finally, we discuss
the possible future role of these white matter maps in the assessment of white
matter diseases, congenital brain malformations, central nervous system
neoplasms (presurgical evaluation), and brain function.
Background
Random motion of water molecules (diffusion) in the presence of a strong
magnetic gradient results in MR signal loss as a result of the dephasing of
spin coherence (Fig. 1). The
application of a pair of strong gradients to elicit differences in the
diffusivity of water molecules among various biologic tissues is known as
diffusion sensitization or diffusion weighting
[22,
23]. The degree of diffusion
weighting is described by the b value, a parameter that is determined by the
type of sensitizing gradient scheme implemented in the MR experiment. For the
Stejskal-Tanner spin-echo scheme
[24]
(Fig. 2)—a pulsed pair of
approximately rectangular gradients around a 180° radiofrequency pulse
that is most commonly implemented on clinical MR scanners—the b value is
determined by the duration (
) and strength (G) of the
sensitizing pulsed gradients, and the time interval between the two pulsed
gradients (
) is determined according to the equation:
where 
is the gyromagnetic ration. Thus, the b value (diffusion
sensitization) can be increased by using stronger (G) and longer
() pulsed gradients or by lengthening the time between the pulsed
gradients ().
View larger version (15K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 1. —Diagram shows loss of intensity of MR signal (S, solid
line) resulting from inefficient rephasing of dephased spins because of
displacement of water molecules (diffusion) between application of bipolar
gradients (G). With decreased diffusion (reduced displacement of
water molecules), rephasing process is more efficient and signal loss is
predominantly caused by T2 decay (dotted line).
|
|
Adding diffusion-sensitizing gradients to an imaging sequence (spatial
encoding) constitutes the basis for diffusion-weighted MR imaging. The signal
intensity (S) in every voxel of a diffusion-weighted MR image is influenced by
the choice of b value and pulse sequence TE and by two parameters intrinsic to
biologic tissues: apparent diffusion coefficient (ADC), a coefficient that
reflects molecular diffusivity in the presence of restrictions, such as
viscosity and spatial barriers; and spin—spin relaxation time (T2). The
following formula describes the relationship between signal intensity in a
diffusion-weighted MR image and the different parameters:
where S0 is the signal intensity at a b value of 0; or the natural
logarithm
Acquiring diffusion-weighted images with at least two different b values
(commonly 20 and 1000 sec/mm2) while keeping the TE fixed allows
the determination of the ADC value for each image voxel
(Fig. 3). The lower of the two
b values is purposefully selected to be slightly greater than zero to
eliminate the effects of large vessels and flow. Assigning a gray scale to the
range of ADC values in the different voxels constitutes an ADC map (Fig.
4A,4B,4C).
The ADC map provides contrast based purely on differences in diffusivity of
water in biologic tissue that is not contaminated by differences in T2
relaxation times (T2 shine-through).
View larger version (14K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 3. —Graph of natural logarithm of signal intensity (Ln [S /
S0]) from diffusion-weighted images with different degrees of
diffusion weighting (b value) allows determination of apparent diffusion
coefficient (ADC) based on linear fit of data points. Absolute value of slope
of plotted line, and thus ADC, is greater for cerebrospinal fluid (CSF) than
for gray matter (GM).
|
|
In white matter, the diffusion of free water molecules is not the same in
all directions of a three-dimensional space (anisotropy)
[9,
10]. Diffusion anisotropy is
predominantly caused by the orientation of fiber tracts in white matter and is
influenced by its micro- and macrostructural features
[25]. Of the microstructural
features, intraaxonal organization appears to be of greatest influence on
diffusion anisotropy; other features include density of fiber and neuroglial
cell packing, degree of myelination, and individual fiber diameter. On a
macroscopic scale, the variability in the orientation of all white matter
tracts in an imaging voxel influences the degree of anisotropy assigned to
that voxel [25].
Diffusion anisotropy is characterized by a 3 x 3 second-rank tensor.
A tensor is a mathematic construct that describes the properties of an
ellipsoid in three-dimensional space (Fig.
5A,5B).
In diffusion tensor imaging, the diffusion properties of water are measured in
the laboratory frame of reference, using the spatial coordinates x,
y, and z (z is the axis along the main magnetic field).
In this laboratory frame, the tensor matrix has nine nonzero elements, of
which three are the same (symmetric tensor). The remaining six elements
(Dxx, Dyy, Dzz,
Dxy, Dxz, and Dyz)
for each voxel are calculated from six images obtained by applying
diffusion-sensitizing gradients in at least six non-colinear directions (for
example: xx, yy, zz, xy, xz, and yz) in addition to a
non—diffusion-weighted image. A property of second-rank tensors is that
they can always be diagonalized, leaving only three nonzero elements along the
main diagonal of the tensor—namely, the eigenvalues
(
1, 2, 3). The
eigenvalues reflect the shape or configuration of the ellipsoid; and their sum
(trace = 1 + 2 + 3),
which is independent of the orientation of the ellipsoid (rotationally
invariant), reflects the size of the ellipsoid
[18]. The mathematic
relationship between the principal coordinates of the ellipsoid and the
laboratory frame is described by the eigenvectors (v1,
v2, v3). The ellipsoid's surface represents the root
mean square diffusive displacement of free water in anisotropic media.
From diffusion tensor imaging, the eigenvectors can be characterized in
each voxel. Several measures of diffusion anisotropy, including fractional
anisotropy, relative anisotropy, and volume ratio, can be calculated on the
basis of formulas that incorporate the tensor elements (Appendix 1) to
generate quantitative brain maps
[26] (Fig.
6A,6B,6C,6D).
A more powerful approach than just using anisotropy maps is to include the
knowledge of the eigenvalues and eigenvectors to generate white matter color
maps, in which the intensity represents anisotropy and the color represents
direction. In addition to the two-dimensional color maps, three-dimensional
white matter fiber tract maps can be created that are based on similarities
between neighboring voxels in the shape (quantitative diffusion anisotropy
measures) and orientation (principal eigenvector map) of the diffusion
ellipsoid [19,
21] (Fig.
6A,6B,6C,6D).
The algorithms used to generate these maps are detailed in the next
section.
View larger version (116K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 6A. —Diffusion tensor MR images in 30-year-old healthy man.
Fractional anisotropy map describes degree of diffusion anisotropy in each
voxel. In white matter, where anisotropy is high, bright end of gray scale is
assigned; in gray matter, where anisotropy is low, dark end of gray scale is
assigned.
|
|
View larger version (178K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 6B. —Diffusion tensor MR images in 30-year-old healthy man. Vector
map coregistered onto T2-weighted MR image (TR/TE, 5000/92) describes
orientation of principal eigenvector in each voxel.
|
|
View larger version (58K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 6C. —Diffusion tensor MR images in 30-year-old healthy man.
Color-coded white matter fiber maps are generated on basis of fractional
anisotropy and vector maps. cc = corpus callosum, slf = superior longitudinal
fasciculus, ilf = inferior longitudinal fasciculus, cst = corticospinal
tract.
|
|
Methodology
Diffusion tensor MR imaging is the only noninvasive in vivo method for
mapping white matter fiber tract trajectories in the human brain. However,
this process is technically and mathematically demanding and requires several
advances in the fields of diffusion-weighted MR imaging and data processing
[21,
27,28,29,30].
First, high-quality MR images are mandatory. Recent MR scanner hardware
improvements, including better stability and homogeneity of the main magnetic
field, as well as stronger and faster magnetic gradients, have been imperative
for the acquisition of high-quality (signal-to-noise ratio and spatial
resolution) diffusion-weighted MR images in a reasonable time. Furthermore,
several schemes to reduce commonly encountered artifacts related to motion,
eddy currents, and field inhomogeneity have also been critical for the success
of diffusion tensor MR imaging. Second, a robust mathematic framework capable
of generating a smooth representation of the macroscopic white matter fiber
tract direction from discrete, coarsely sampled, and voxel-averaged diffusion
tensor data has to be developed. Third, a computer-based algorithm that allows
the reliable following or tracking of the individual white matter fiber tracts
must be created.
Data Acquisition
The process of data acquisition in diffusion tensor MR imaging consists of
diffusion sensitization and spatial encoding. Diffusion tensor imaging is
influenced by the strength, number, and orientation (diffusion tensor encoding
scheme) of the sensitizing gradients. Regarding the strength of the
diffusion-sensitizing gradients, Melhem et al.
[31] have shown an effect of b
value on the measures of diffusivity and anisotropy in different anatomic
locations in the brain. On the other hand, the impact of the number of b
values on these measures has been minimal in the range between 0 and 1000
sec/mm2 [32].
Regarding the number of directions and the orientation of the
diffusion-sensitizing gradients, more uniform scanning of three-dimensional
space has been shown to result in a reduction in noise in raw
diffusion-encoded measurements and to improve the accuracy of diffusion tensor
estimates and fiber tracking applications. Uniform scanning of
three-dimensional space is achieved by increasing the number of sensitizing
gradient directions (up to 642 directions) and optimizing their spatial
orientation [33,
34].
Recently, on the basis of analytic comparisons and Monte Carlo simulations,
Hasan et al. [35] have found,
using a single-tensor model of diffusion, that there is no substantial
advantage to implementing more than six encoding directions as long as the
orientation of the gradients (diffusion tensor encoding schemes) is optimized.
Those authors also addressed the influence of various diffusion tensor
encoding schemes, including heuristic, numerically optimized, and geometric
polyhedra, on the accuracy of diffusion tensor estimates; they found that the
commonly implemented six-directional heuristic schemes are suboptimal
[35].
A variety of spatial encoding schemes have been proposed for diffusion
tensor MR imaging. The most commonly used in practice are based on echoplanar
readout and include single-shot (so-called snapshot) and multishot (or
interleaved, segmented) techniques. The advantages of multishot echoplanar
imaging compared with single-shot are greater spatial resolution, greater
signal-to-noise ratio, and less susceptibility-related distortion. The major
disadvantage is longer data acquisition time, which makes multishot echoplanar
techniques more susceptible to artifacts related to respiration,
cerebrovascular and cerebrospinal fluid flow, eye motion, and involuntary head
motion. In diffusion-sensitive multishot echoplanar imaging, even small bulk
motions can cause ghosting artifacts because of discontinuities between the
echoes sampled at different times in k-space. In single-shot echoplanar
imaging sequences, these discontinuities in k-space are eliminated because the
entire set of echoes is acquired with the same motion-induced error
[23]. To reduce motion-related
artifacts in diffusion-sensitive multishot echoplanar techniques, cardiac
gating and motion correction schemes such as navigator echo have been
implemented
[36,37,38,39,40].
Navigator echo correction, which is a non—phase-encoded readout gradient
echo, provides a measure of the motion-induced phase variations between the
echoes in each shot and corrects for small amounts of motion
[40]. Diffusion-sensitive
multishot echoplanar techniques with cardiac gating and navigator echo
correction provide images with reduced motion artifacts and great spatial
resolution that are essential for evaluating the micro-structures of white
matter in vivo [40,
41].
Other spatial encoding schemes used in diffusion tensor imaging include
gradient-and spin-echo (GRASE), fast spin-echo, and line scanning
[42,43].
Both GRASE and fast spin-echo techniques are resistant to distortions from
static field inhomogeneities and diffusion sensitization gradient-induced eddy
currents compared with single-shot echoplanar techniques. However, GRASE and
fast spin-echo suffer from a low signal-to-noise ratio and relatively long
acquisition times [44].
Diffusion tensor line scanning or column-selective excitation techniques are
occasionally used in imaging children because of the inherent insensitivity of
those techniques to variations in the phase of the MR signal induced by
physiologic motion
[42,45].
Again, the main disadvantages of this technique are an inherently low
signal-to-noise ratio (because only a single column contributes to the MR
signal) and relatively long acquisition times.
In this review, brain diffusion tensor MR images are generated on a 1.5-T
scanner equipped with a 2.3 G/cm triple-axis gradient system and a head coil
operating in receive mode. The imaging protocol consists of a
diffusion-sensitive, cardiac-gated, navigator echo—corrected,
two-dimensional multishot spin-echo echoplanar readout. The TE is fixed at 92
msec, whereas the TR varies with the heart rate (5000-6000 msec). The number
of echoes acquired in each TR interval is 18 (1 navigator echo and 17
phase-encoded echoes). The field of view ranges from 12 x 12 to 25
x 25 cm, depending on the head size. The number of sampling points along
frequency- and phase-encoding directions ranges from 64 to 128 interpolated to
128 to 256 (zero filling) to maintain a pixel size of 0.98 mm2.
Forty interleaved, 3-mm-thick slices are acquired in the axial plane using the
multislice mode. Diffusion sensitization is performed along six noncolinear
directions (heuristic encoding schemes: [1, 0, 0]: [0, 1, 0]: [0, 0, 1]: [0,
2-1/2, 2-1/2]: [2-1/2, 0, 2-1/2]:
[2-1/2, 2-1/2, 0]) using diffusion weighting of b = 600
sec/mm2 at the maximum gradient strength of 2.1 G/cm. A reference
image with low diffusion weighting (b = 33 sec/mm2) is also
recorded. A single set of these seven measurements takes 4-5 min depending on
heart rate. These measurements are repeated at least six times to increase the
signal-to-noise ratio. The images are checked and the acquisition is repeated
when motion-related degrading artifacts are found. In addition, double-echo
(TEs of 22 and 92 msec) two-dimensional multishot spin-echo echoplanar MR
imaging is performed for anatomic guidance. The diffusion tensor and
double-echo MR imaging protocols are matched for spatial orientation and
resolution to facilitate the coregistration process. The entire examination
requires approximately 50 min.
Data Correction and Processing
Misregistration of diffusion tensor MR images has deleterious effects on
spatial resolution and on the accuracy of diffusivity and anisotropy estimates
[46]. Misregistration is
caused by variations in geometric distortions and by misalignment resulting
from minimal motion between the various measurements. The geometric
distortions are caused by static field inhomogeneity resulting from imperfect
shimming and differences in magnetic properties of adjacent tissues and from
diffusion sensitization gradient-induced eddy currents
[41,44,46].
Unwarping algorithms, using inhomogeneity field maps, are used for the
correction of distortion originating from the static magnetic field
inhomogeneities [41]. For the
correction of diffusion sensitization gradient-induced eddy current
distortions, two methods are often used. The first method entails the
application of least squares straight line fits and cross-correlation
functions to the diffusion-weighted image
[46,47].
The second method uses a one-dimensional inhomogeneity field map in the read
and phase-encoding directions in each slice and each diffusion-sensitizing
direction step to correct the distortion
[41]. Motion-related
misalignment can be corrected by computer-based algorithms, such as Automated
Image Registration software
[48], that take into account
differences in contrast and spatial resolution between the coregistered
images.
From the diffusion tensor MR imaging data, quantitative, directionally
invariant (relatively independent of the diffusion gradient directions, head
position, and rotation) indexes can be derived that measure distinct intrinsic
features of water diffusion in biologic tissue. The most fundamental
quantitative measures are the three principal diffusivities (eigenvalues) of
the diffusion tensor, which are the principal diffusion coefficients measured
along the three principal coordinates of the ellipsoid in each voxel. From the
eigenvalues, several indexes that measure the degree of diffusion anisotropy
in tissues have been proposed. The most intuitive and simplest indexes are
ratios of the principal diffusivities
[49], such as the
dimensionless anisotropy ratio h / l or
2h / (m + l), which
measure the relative magnitudes of the diffusivities along the longest axis of
the diffusion ellipsoid and other orthogonal axes. Here the eigenvalues are
sorted in order of decreasing magnitude (h: highest
diffusivity; m: intermediate diffusivity; and
l: lowest diffusivity). Because simulations have revealed
that these sorted indexes are statistically biased by noise contamination
[49], other nonsorted indexes,
such as volume ratio, relative anisotropy, and fractional anisotropy, have
been proposed [49] (Appendix
1). Volume ratio—the ratio of the volume of an ellipsoid to the volume
of a sphere whose radius is the averaged diffusivity—ranges between 0
and 1, where 0 means highest anisotropy and 1 indicates complete isotropic
diffusion. Relative anisotropy—the ratio of the variance of the
eigenvalues to their mean—and fractional anisotropy—the ratio of
the anisotropic component of the diffusion tensor to the whole diffusion
tensor—both vary between 0 (isotropic diffusion) and 1 (infinite
anisotropic diffusion).
Simulations have shown a rapid divergence of the eigenvalues and invariant
anisotropy indexes from true values as the signal-to-noise ratio decreases
[49,
50]. In tissues with low-level
anisotropy, this divergence is further exaggerated in anisotropy indexes,
requiring sorting of the eigenvalues. Indexes such as the
"lattice" anisotropy index that include all the information of the
eigenvectors (both shape and orientation of the ellipsoid) and do not require
sorting have been shown to reduce the effect of noise on the measurement
[49].
In this review, motion-related misalignment of the diffusion tensor MR
images is corrected off-line using Automated Image Registration. Subsequently,
the voxel intensities of the multiple diffusion-weighted images are fitted
using multivariant linear least square fitting to obtain the six elements of
the symmetric diffusion tensor. The diffusion tensors at each voxel are
diagonalized to obtain eigenvalues and eigenvectors for each voxel. The
eigenvector (v1) associated with the largest eigenvalue
(1) is used to represent the local fiber direction.
Fractional anisotropy maps are calculated from the eigenvalues on the basis of
standard formulas (Appendix 1). White matter color maps are created on the
basis of the three vector elements of the eigenvector v1 for each
voxel. The absolute values of the vector elements are assigned to red
(x element), green (y element), and blue (z
element) [12]. If the
principal eigenvector is aligned along the x-axis, pure red is
assigned to the corresponding voxel, whereas if the eigenvector is 45°
between the x- and y-axes, yellow (red plus green) is
assigned to the voxel. The intensity of color in each voxel is gauged by the
degree of fractional anisotropy (Fig.
6A,6B,6C,6D).
White Matter Fiber Tracking
Methods to reconstruct white matter tracts can be placed into two broad
categories [19,
21,
51,52,53,54].
Methods in the first category are based on line propagation algorithms that
use local tensor information for each step of the propagation. Simple line
propagation techniques that connect voxels on the basis of discrete number
fields (local principal eigenvector orientation) are incapable of providing
accurate representation of white matter tracts
[19]. Improvements in line
propagation techniques using continuous, rather than discrete, number fields
can provide connections that follow the actual white matter tract
[19,
52]. Furthermore, line
propagation techniques can be modified to create a smooth (curved) path by
interpolating the vector of the principal axis or the whole diffusion tensor
at each coordinate as a line is propagated
[21,
51].
Methods in the second category are based on global energy minimization to
find a path between two predetermined voxels with minimum energy violation.
This category includes the methods of fast marching (Parker GJ et al.,
presented at the International Society of Magnetic Resonance meeting, April
2000) and simulated annealing (Tuch DS et al., presented at the ISMRM meeting,
April 2001).
In this review, fiber tracking is performed using a line propagation
technique based on continuous number fields developed at our center called
"fiber assignment using continuous tracking"
[19]. Tracking is launched
from a seed voxel from which a line is propagated in both retrograde and
antegrade directions according to the principal eigenvector at each voxel.
Tracking propagates on the basis of the orientation of the eigenvector that is
associated with the largest eigenvalue. Tracking is terminated when it reaches
a voxel with fractional anisotropy lower than a threshold of 0.25-0.35 and
when the angle between the two principal eigenvectors is greater than
35-40° (Fig. 7).
View larger version (120K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 7. —Schematic shows white matter tracking used by algorithm.
Degree of diffusion anisotropy is indicated by gray scale (white is highest),
and direction of principal eigenvector in each image voxel is indicated by an
arrow. On basis of defined thresholds, tracking (long curved arrows)
is along voxels with similar measures of anisotropy and direction of principal
eigenvector. Algorithm can distinguish between tracts A and B because they are
separated by voxels with low anisotropy, and between tracts A and C because of
differences in direction of principal eigenvectors. Asterisks indicate
starting points of tracking.
|
|
Fractional anisotropy thresholds help to exclude gray matter and to segment
white matter tracts that are separated by gray matter
(Fig. 7). In the situation in
which two white matter tracts are close
(Fig. 7), the angle between
the principal eigenvectors of the adjacent white matter voxels becomes an
important criterion for adequate segmentation. Our choice of fractional
anisotropy and angle thresholds is based on experiments that evaluated the
reliability of tracking the corticospinal tract using different
thresholds.
Knowing the fiber projections in relation to anatomic landmarks, we define
multiple regions of interest. The use of multiple regions of interest allows
separation of white matter tracts that are adjacent to each other in one
anatomic location and distant in another. For example, at the level of the
cerebral peduncle, it is difficult to separate the different
corticopontospinal tracts; placing another region of interest in the lower
pons facilitates their separation (Fig.
8A,8B,8C).
View larger version (115K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 8A. —Diffusion tensor MR images in 26-year-old healthy man.
(Reprinted with permission from
[82]) Three-dimensional white
matter fiber tracking of frontopontine (FPT, green), corticospinal
(CST, red), and temporoparietooccipitopontine (TPOPT, blue)
tracts coregistered onto sagittal T2-weighted MR images (TR/TE, 5000/92).
Tracking is based on two regions of interest (ROIs). ROI placed at level of
midbrain (blue arrow) does not allow separation between tracts
traversing cerebral peduncle because of closeness and similarities in
direction and anisotropy. Another ROI at level of lower pons, below
termination of FPT and TPOPT (purple arrow), allows separation of
tracts.
|
|
View larger version (141K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 8B. —Diffusion tensor MR images in 26-year-old healthy man.
(Reprinted with permission from
[82]) Three-dimensional white
matter fiber tracking of FPT (green), CST (red), and TPOPT
(blue) tracts coregistered onto axial T2-weighted MR images
(5000/92).
|
|
Tracking can be achieved in one of two ways. The first method consists of
initiating tracking from each voxel included in the region of interest. This
method can delineate a limited number of branching patterns of the tract of
interest (if the region of interest contains 10 voxels, only 10 tracking
results delineate the tract). In the second method, fiber tracking is
initiated from the center of every voxel in the brain, but only fibers passing
through the specified regions of interest are retained
[51]. In this approach,
multiple tracts penetrate the region of interest, thus revealing a more
comprehensive structure of the tract.
One of the difficulties in tracking a fiber path from the three-dimensional
vector matrix is to translate the discrete vector information to continuous
fiber tracts. The fiber-assignment-using-continuous-tracking algorithm
maintains the information of intercept when the tracking exits a voxel by
performing the tracking in a continuous number field rather than in a discrete
number field (Fig.
9A,9B).
View larger version (104K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 9A. —Schematic shows difference between tracking methods.
Maintaining information on intercepts is achieved by performing tracking in a
continuous number field, which results in precise white matter fiber tracking.
Voxels from which tracking is launched are designated by asterisks, direction
of principal eigenvector in each voxel is designated by an arrow, and voxels
connected by tracking algorithm are indicated by boxes with shaded arrows.
Dark arrows indicate interrupted trajectory of track. (Reprinted with
permission from [19]) Tracking
(long curved arrows) without (A) and with (B) keeping
information on intercepts when tracking leaves voxel.
|
|
View larger version (117K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 9B. —Schematic shows difference between tracking methods.
Maintaining information on intercepts is achieved by performing tracking in a
continuous number field, which results in precise white matter fiber tracking.
Voxels from which tracking is launched are designated by asterisks, direction
of principal eigenvector in each voxel is designated by an arrow, and voxels
connected by tracking algorithm are indicated by boxes with shaded arrows.
Dark arrows indicate interrupted trajectory of track. (Reprinted with
permission from [19]) Tracking
(long curved arrows) without (A) and with (B) keeping
information on intercepts when tracking leaves voxel.
|
|
The final step of this process is coregistering the three-dimensional white
matter fiber tracts on the anatomic T2-weighted images using Automated Image
Registration software.
Applications
Diffusion tensor imaging data promise a range of applications in clinical
medicine. Initially, diffusion anisotropy was observed in skeletal muscle
[55]. Since then, diffusion
tensor measures have been used to study fibrous organized structures like
human and animal spinal cords
[56], myocardium
[57,
58], intervertebral discs
[59], and cerebral white
matter [60].
Clinically, properties derived from the diffusion tensor like the trace,
which reflects overall water content, have been used successfully to evaluate
brain ischemia [61,
62]. Measures of the diffusion
tensor have also been used to investigate brain development
[17,
63] and to aid in
neurosurgical planning. In addition, parameters derived from the diffusion
tensor, such as anisotropy indexes, have been used to evaluate white matter
disease in Krabbe's disease
[64], cerebral
adrenoleukodystrophy [65],
AIDS [66], multiple sclerosis
[20,
67,
68], hypertensive
encephalopathy [69],
age-related changes [70],
schizophrenia [71],
Alzheimer's disease [72],
ischemic leukoaraiosis [73],
and epilepsy [74]. Other
studies have also probed the potential of diffusion tensor MR imaging in brain
tumors [75], migraines
[76], and eclampsia
[77].
Requisites for sound application of diffusion tensor MR imaging and white
matter tractography in central nervous system disease are the establishment of
normative anisotropy values and white matter tract maps, and the validation of
the normative white matter tract maps on the basis of detailed anatomic
knowledge of these tracts. As illustrated in Figures
6A,6B,6C,6D
and
10A,10B,
it is feasible to map anisotropy values and white matter tracts for both the
infra- and supratentorial compartments of the brain in healthy volunteers.
However, the process of anatomic validation has been more difficult because of
a lack of an appropriate gold standard. In fact, diffusion tensor MR imaging
is the only method available that has the potential for tracking white matter
tracts in vivo. Attempts at in vitro validation based on white matter
histology are limited because of geometric distortions resulting from
dissection, freezing, dehydration, fixation, cutting into thin slices, and
thawing of the histologic samples
[21]. The radiologist
interpreting these maps needs to have a detailed knowledge of the location,
origin, and termination of the different white matter tracts in the central
nervous system. Furthermore, a thorough understanding of the effects of
artifacts originating from image acquisition (diffusion tensor MR imaging),
data processing, and the tracking algorithm on the accuracy of the geometry
(shape) and topology (branching) of the white matter tracts is imperative for
proper interpretation of the maps. Specifically, misregistration of diffusion
tensor MR images caused by eddy currents, ghosting due to motion, and signal
loss due to susceptibility variations can all affect the computed trajectory
of the fiber tract. Furthermore, the tracking algorithm may fail to provide an
accurate representation of the fiber trajectory when a voxel contains
nonuniformly distributed fibers, curved fibers, or two or more interdigitating
fiber populations. This failure is largely because the direction of the
measured principal eigenvector is based on a voxel average and does not
necessarily represent the trajectories of individual microscopic tracts
[21].
View larger version (135K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 10A. —Diffusion tensor MR images in 35-year-old healthy man.
Three-dimensional fiber tracking of brainstem with ventral (purple)
and medial (turquoise) components of middle cerebellar peduncles,
inferior cerebellar peduncle (green), superior cerebellar peduncle
(yellow), medial longitudinal fasciculus (orange), and
corticopontospinal (red) tracts are superimposed on these two images.
(Reprinted with permission from
[82]) Mid sagittal T2-weighted
MR image (TR/TE, 5000/92) of brainstem and cerebellum.
|
|
View larger version (92K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 10B. —Diffusion tensor MR images in 35-year-old healthy man.
Three-dimensional fiber tracking of brainstem with ventral (purple)
and medial (turquoise) components of middle cerebellar peduncles,
inferior cerebellar peduncle (green), superior cerebellar peduncle
(yellow), medial longitudinal fasciculus (orange), and
corticopontospinal (red) tracts are superimposed on these two images.
(Reprinted with permission from
[82]) Photograph of anatomic
specimen shows postmortem dissection of brainstem and cerebellum.
|
|
At our center, diffusion tensor MR imaging and white matter tractography
are being applied in three major clinical research projects. The first project
studies children with a cerebral form of adrenoleukodystrophy, a genetic
disorder that involves the white matter, adrenal cortex, and testes
[78]; the second studies
patients with spastic cerebral palsy resulting from periventricular
leukomalacia, a dominant pattern of anoxic brain injury in premature infants
[79]; and the third studies
infants with holoprosencephaly, a congenital brain malformation attributed to
errors in ventral induction
[80].
In adrenoleukodystrophy, diffusion tensor MR imaging may help identify
affected white matter that is not evident on conventional MR imaging,
differentiate between potentially reversible and irreversibly damaged white
matter, and categorize affected white matter on the basis of well-defined
histopathologic zones (Fig.
11A,11B).
Information derived from ADC and fractional anisotropy measures in these
patients may also shed light on axonal and myelin ultrastructure contributions
to the anisotropy phenomena. Preliminary results show that affected white
matter can be divided into three distinct zones on the basis of differences in
diffusion values that may reflect varying degrees of axonal and myelin loss
(Table 1). The effect of white
matter lesions on specific commissural (corpus callosum) (Fig.
12A,12B,12C,12D),
projectional (corticospinal), and association tracts is currently being
correlated with neurologic and neuropsychologic function. Similar hypotheses
are being tested in patients with periventricular leukomalacia. The severity
of injury to the periventricular white matter determines the degree of damage
to white matter tracts traversing the injured zone (Fig.
13A,13B,13C,13D).
In particular, correlations between damage to different transcallosal cortical
connections and types of neuropsychologic deficits are being evaluated. The
severity of spastic diplegia is also being correlated with the degree of
injury to traversing corticospinal tracts.
View larger version (96K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 11A. —9-year-old boy with cerebral X-linked adrenoleukodystrophy
who presented with progressive motor and visual deficits. T2-weighted MR image
(TR/TE, 5000/92) shows abnormal symmetric hyperintensity involving
peritrigonal white matter.
|
|
View larger version (142K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 11B. —9-year-old boy with cerebral X-linked adrenoleukodystrophy
who presented with progressive motor and visual deficits. Fractional
anisotropy map shows central zone (asterisks) of marked hypointensity
(marked decrease in anisotropy) and peripheral zone (arrowheads) of
mild hypointensity (mild decrease in anisotropy).
|
|
View larger version (93K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 12A. —6-year-old boy with anterior form of X-linked
adrenoleukodystrophy who presented with personality changes.
Diffusion-weighted images show three-dimensional fiber tracking of corpus
callosum coregistered onto images of affected boy and of 8-year-old healthy
boy. In affected boy, substantial decrease is seen in white matter tracts
crossing genu and anterior body of corpus callosum and in both frontal lobes
compared with healthy boy. Axial (A) and sagittal (B)
T2-weighted MR images (TR/TE, 5000/92) of affected boy.
|
|
View larger version (91K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 12B. —6-year-old boy with anterior form of X-linked
adrenoleukodystrophy who presented with personality changes.
Diffusion-weighted images show three-dimensional fiber tracking of corpus
callosum coregistered onto images of affected boy and of 8-year-old healthy
boy. In affected boy, substantial decrease is seen in white matter tracts
crossing genu and anterior body of corpus callosum and in both frontal lobes
compared with healthy boy. Axial (A) and sagittal (B)
T2-weighted MR images (TR/TE, 5000/92) of affected boy.
|
|
View larger version (96K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 12C. —6-year-old boy with anterior form of X-linked
adrenoleukodystrophy who presented with personality changes.
Diffusion-weighted images show three-dimensional fiber tracking of corpus
callosum coregistered onto images of affected boy and of 8-year-old healthy
boy. In affected boy, substantial decrease is seen in white matter tracts
crossing genu and anterior body of corpus callosum and in both frontal lobes
compared with healthy boy. Axial (C) and sagittal (D)
T2-weighted MR images (5000/92) of healthy boy.
|
|
View larger version (97K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 12D. —6-year-old boy with anterior form of X-linked
adrenoleukodystrophy who presented with personality changes.
Diffusion-weighted images show three-dimensional fiber tracking of corpus
callosum coregistered onto images of affected boy and of 8-year-old healthy
boy. In affected boy, substantial decrease is seen in white matter tracts
crossing genu and anterior body of corpus callosum and in both frontal lobes
compared with healthy boy. Axial (C) and sagittal (D)
T2-weighted MR images (5000/92) of healthy boy.
|
|
View larger version (105K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 13A. —6-year-old boy with cerebral palsy resulting from
periventricular leukomalacia who presented with asymmetric spastic diplegia
affecting left side more than right. Color maps of brainstem white matter
tracts show decrease in size of corticopontospinal tracts in affected boy
(arrowheads, A and B) compared with those shown on
color maps of brainstem white matter tracts in 8-year-old healthy boy
(arrowheads, C and D). Furthermore, right
corticopontospinal tract is more involved than left in affected boy, which
correlates with his neurologic examination.
|
|
View larger version (88K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 13B. —6-year-old boy with cerebral palsy resulting from
periventricular leukomalacia who presented with asymmetric spastic diplegia
affecting left side more than right. Color maps of brainstem white matter
tracts show decrease in size of corticopontospinal tracts in affected boy
(arrowheads, A and B) compared with those shown on
color maps of brainstem white matter tracts in 8-year-old healthy boy
(arrowheads, C and D). Furthermore, right
corticopontospinal tract is more involved than left in affected boy, which
correlates with his neurologic examination.
|
|
View larger version (106K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 13C. —6-year-old boy with cerebral palsy resulting from
periventricular leukomalacia who presented with asymmetric spastic diplegia
affecting left side more than right. Color maps of brainstem white matter
tracts show decrease in size of corticopontospinal tracts in affected boy
(arrowheads, A and B) compared with those shown on
color maps of brainstem white matter tracts in 8-year-old healthy boy
(arrowheads, C and D). Furthermore, right
corticopontospinal tract is more involved than left in affected boy, which
correlates with his neurologic examination.
|
|
View larger version (110K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 13D. —6-year-old boy with cerebral palsy resulting from
periventricular leukomalacia who presented with asymmetric spastic diplegia
affecting left side more than right. Color maps of brainstem white matter
tracts show decrease in size of corticopontospinal tracts in affected boy
(arrowheads, A and B) compared with those shown on
color maps of brainstem white matter tracts in 8-year-old healthy boy
(arrowheads, C and D). Furthermore, right
corticopontospinal tract is more involved than left in affected boy, which
correlates with his neurologic examination.
|
|
In children with holoprosencephaly, anomalous white matter tracts are
identified in addition to the well-described noncleavage of cortical and deep
gray matter structures. Rigorous comparison with age-matched normative white
matter tract maps is still required for better understanding of these complex
anomalies. Defining white and gray matter anomalies may allow more precise
prediction of developmental outcome in holoprosencephaly (Fig.
14A,14B,14C,14D).
View larger version (90K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 14A. —1-year-old boy with middle hemispheric variant of
holoprosencephaly. Axial T2-weighted MR image (TR/TE, 3000/100) above level of
lateral ventricles shows site of noncleavage of cerebral hemispheres.
Arrowheads indicate white matter.
|
|
View larger version (152K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 14B. —1-year-old boy with middle hemispheric variant of
holoprosencephaly. Coronal fast phase-sensitive inversion-recovery MR image
(3300/40; inversion time, 200 msec) through frontal horns shows noncleavage of
cerebral hemispheres with continuation of anomalous gray and white matter
(arrowheads) across midline. Note absence of septum pellucidum.
|
|
View larger version (114K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 14C. —1-year-old boy with middle hemispheric variant of
holoprosencephaly. Fractional anisotropy map in axial plane, above level of
lateral ventricles, helps differentiate between white (high anisotropy,
arrowheads) and gray (low anisotropy, asterisks) matter
crossing midline.
|
|
View larger version (88K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 14D. —1-year-old boy with middle hemispheric variant of
holoprosencephaly. White matter color map provides information about direction
of white matter tracts, with red assigned to tracts running across midline
(x-axis, arrowheads) and blue assigned to tracts running
perpendicular to image section (z-axis, asterisks).
|
|
Other applications of diffusion tensor MR imaging and white matter
tractography with more immediate clinical impact are the evaluation of brain
tumors and acute stroke. Defining the relationship of brain tumors to eloquent
white matter tracts will undoubtedly help guide the surgical approach and the
extent of the resection (Fig.
15A,15B,15C,15D,15E).
The true extent of infiltration of neoplasms along white matter tracts may be
better delineated with these techniques. With respect to stroke, preliminary
results have shown an immediate increase in fractional anisotropy in regions
of reversible acute ischemia (decreased diffusion), which may be related to
reduction in flow. Further investigation of this finding and its use in
guiding acute stroke management is warranted.
View larger version (131K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 15A. —T2-weighted MR images (TR/TE, 5000/92) in 48-year-old woman
with left-sided posterior fossa meningioma. (Reprinted with permission from
[82]) Mid sagittal image shows
levels of axial sections (B, C, and D) and slightly hyperintense (compared
with brain) extraaxial mass compressing brainstem.
|
|
View larger version (137K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 15B. —T2-weighted MR images (TR/TE, 5000/92) in 48-year-old woman
with left-sided posterior fossa meningioma. (Reprinted with permission from
[82]) Axial images at level of
third ventricle (B), anterior commissure (C), and brainstem
(D) show coregistered left (red) and right (yellow)
corticopontospinal tracts. Note relationship of extraaxial mass to posteriorly
displaced left corticospinal tract.
|
|
View larger version (133K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 15C. —T2-weighted MR images (TR/TE, 5000/92) in 48-year-old woman
with left-sided posterior fossa meningioma. (Reprinted with permission from
[82]) Axial images at level of
third ventricle (B), anterior commissure (C), and brainstem
(D) show coregistered left (red) and right (yellow)
corticopontospinal tracts. Note relationship of extraaxial mass to posteriorly
displaced left corticospinal tract.
|
|
View larger version (135K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 15D. —T2-weighted MR images (TR/TE, 5000/92) in 48-year-old woman
with left-sided posterior fossa meningioma. (Reprinted with permission from
[82]) Axial images at level of
third ventricle (B), anterior commissure (C), and brainstem
(D) show coregistered left (red) and right (yellow)
corticopontospinal tracts. Note relationship of extraaxial mass to posteriorly
displaced left corticospinal tract.
|
|
View larger version (82K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 15E. —T2-weighted MR images (TR/TE, 5000/92) in 48-year-old woman
with left-sided posterior fossa meningioma. (Reprinted with permission from
[82]) Three-dimensional
representation of left corticospinal tract superimposed onto sagittal and
coronal images better shows relationship between mass (green) and
tract (red).
|
|
The Future
White matter tractography based on diffusion tensor imaging is a rapidly
evolving technology in central nervous system imaging, with many challenges
and exciting new applications. Improvements in signal-to-noise ratios continue
to be required for more precise calculation of anisotropy measures and more
accurate white matter fiber tracking. Imaging with isotropic, high-resolution
voxels reduces the effect of volume averaging on the direction of the
principal eigenvector, which provides a better representation of the actual
orientation of the fiber tract within a voxel and the accuracy of the tracking
algorithm [21]. Pulse sequence
innovations are needed, such as an optimized diffusion-sensitive
three-dimensional echoplanar technique that allows isotropic high-resolution
imaging with an improved signal-to-noise ratio compared with two-dimensional
acquisitions. Another critical issue that limits the routine clinical use of
this technique is the relatively long acquisition time resulting from the many
averages or excitations required to enhance image signal-to-noise ratio and
built-in acquisition redundancies needed to replace motion-corrupted images.
To improve the efficiency of the acquisition, a scheme is needed that is based
on "real-time" navigator echo correction. The scheme consists of
continuously identifying and reacquiring only the corrupted portions of each
acquisition. This scheme improves the efficiency of scanning by eliminating
the need for a manual check of each acquisition before deciding whether more
scans are needed and by decreasing the degree of redundancy in the
acquisition.
Fiber tracking algorithms, such as diffusion spectrum techniques that may
resolve fiber orientation heterogeneity in a voxel, are n
Top
Introduction
Background
