Anda di halaman 1dari 14

AJ R:178, J anuary 2002

3

n the last 1015 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

1

H MR imag-
ing 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
8]. More recently, the dependency of molecular
diffusion on the orientation of white matter ber
tracts has elicited great interest in studying the
factors that inuence this dependency and in
spatially mapping these ber tracts using diffu-
sion imaging [921].
In this paper, we briey 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 tech-
niques for acquiring relatively high-resolution
diffusion-sensitive MR images and computer-
based algorithms that allow the generation of
white matter ber tract maps from the tensor
data. We provide an overview of current expe-
rience and some clinical examples that are on-
going 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 ner-
vous system neoplasms (presurgical evalua-
tion), and brain function.

Background

Random motion of water molecules (diffu-
sion) in the presence of a strong magnetic gra-
dient results in MR signal loss as a result of the
dephasing of spin coherence (Fig. 1). The ap-
plication of a pair of strong gradients to elicit
differences in the diffusivity of water mole-
cules among various biologic tissues is known
as diffusion sensitization or diffusion weight-
ing [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 rect-
angular gradients around a 180 radiofre-
quency pulse that is most commonly
implemented on clinical MR scannersthe b
value is determined by the duration (


) and
strength (

G

) of the sensitizing pulsed gradi-
ents, and the time interval between the two
pulsed gradients (


) is determined according
to the equation:
b value =


2

G

2


2

(


/ 3),
where


is the gyromagnetic ration. Thus, the
b value (diffusion sensitization) can be in-
creased by using stronger (

G

) and longer (


)
pulsed gradients or by lengthening the time
between the pulsed gradients (


).
Adding diffusion-sensitizing gradients to an
imaging sequence (spatial encoding) consti-
tutes the basis for diffusion-weighted MR im-
aging. The signal intensity (S) in every voxel
of a diffusion-weighted MR image is inu-
enced by the choice of b value and pulse se-
quence TE and by two parameters intrinsic to
biologic tissues: apparent diffusion coefcient
(ADC), a coefcient that reects molecular
diffusivity in the presence of restrictions, such
as viscosity and spatial barriers; and spinspin
relaxation time (T2). The following formula
describes the relationship between signal in-
tensity in a diffusion-weighted MR image and
the different parameters:
S = S

0

e

b(ADC)

,
where S

0

is the signal intensity at a b value of
0; or the natural logarithm
Ln (S / S

0

) = b(ADC).
Acquiring diffusion-weighted images with
at least two different b values (commonly 20
and 1000 sec/mm

2

) while keeping the TE
xed allows the determination of the ADC
value for each image voxel (Fig. 3). The
lower of the two b values is purposefully se-
lected to be slightly greater than zero to elim-
inate the effects of large vessels and ow.
Assigning a gray scale to the range of ADC
values in the different voxels constitutes an
ADC map (Fig. 4). The ADC map provides
contrast based purely on differences in diffu-

Diffusion Tensor MR Imaging of the Brain and
White Matter Tractography

Elias R. Melhem

1,2

, Susumu Mori

1,2

, Govind Mukundan

1

, Michael A. Kraut

1,2

, Martin G. Pomper

1

, Peter C. M. van Zijl

1,2

Received December 28, 2000; accepted after revision J uly 5, 2001.

1

Department of Radiology and Radiological Sciences, The J ohns Hopkins Medical Institutions, 600 N. Wolfe St., Baltimore, MD 21287-2182. Address correspondence to E. R. Melhem.

2

F. M. Kirby Research Center for Functional Brain Imaging, Kennedy-Krieger Institute, The J ohns Hopkins Medical Institutions, 707 N. Broadway, Baltimore, MD 21287.

AJ R

2002;178:316 0361803X/02/17813 American Roentgen Ray Society

Review

I
D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
j
r
o
n
l
i
n
e
.
o
r
g

b
y

1
9
0
.
1
6
9
.
9
9
.
1
1
5

o
n

0
2
/
2
0
/
1
4

f
r
o
m

I
P

a
d
d
r
e
s
s

1
9
0
.
1
6
9
.
9
9
.
1
1
5
.

C
o
p
y
r
i
g
h
t

A
R
R
S
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
;

a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d


4

AJ R:178, J anuary 2002

Melhem et al.

sivity of water in biologic tissue that is not
contaminated by differences in T2 relaxation
times (T2 shine-through).
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 ber tracts in white matter
and is inuenced by its micro- and macrostruc-
tural features [25]. Of the microstructural fea-
tures, intraaxonal organization appears to be of
greatest inuence on diffusion anisotropy;
other features include density of ber and neu-
roglial cell packing, degree of myelination,
and individual ber diameter. On a macro-
scopic scale, the variability in the orientation
of all white matter tracts in an imaging voxel
inuences the degree of anisotropy assigned to
that voxel [25].
Diffusion anisotropy is characterized by a 3


3 second-rank tensor. A tensor is a mathematic
construct that describes the properties of an el-
lipsoid in three-dimensional space (Fig. 5). In
diffusion tensor imaging, the diffusion proper-
ties of water are measured in the laboratory
frame of reference, using the spatial coordi-
nates

x

,

y

, and

z

(

z

is the axis along the main
magnetic eld). In this laboratory frame, the
tensor matrix has nine nonzero elements, of
which three are the same (symmetric tensor).
The remaining six elements (

D

xx

,

D

yy

,

D

zz

,

D

xy

,

D

xz

, and

D

yz

) for each voxel are calculated
from six images obtained by applying diffu-
sion-sensitizing gradients in at least six non-
colinear directions (for example:

xx

,

yy

,

zz

,

xy

,

xz

, and

yz

) in addition to a nondiffusion-
weighted image. A property of second-rank
tensors is that they can always be diagonal-
ized, leaving only three nonzero elements
along the main diagonal of the tensor
namely, the eigenvalues (


1

,


2

,


3

). The
eigenvalues reect the shape or conguration
of the ellipsoid; and their sum (trace =


1

+


2

+


3

), which is independent of the orientation
of the ellipsoid (rotationally invariant), reects
Fig. 1.Diagramshows loss of intensity of MR signal (S,
solid line) resulting frominefcient rephasing of dephased
spins because of displacement of water molecules (diffu-
sion) between application of bipolar gradients (G). With
decreased diffusion (reduced displacement of water mol-
ecules), rephasing process is more efcient and signal
loss is predominantly caused by T2 decay (dotted line).
Fig. 2.Diagram shows Stejskal-Tanner pulsed bipolar gradient
scheme [24]. This scheme is commonly implemented on clinical MR
scanners for diffusion sensitization. Degree of diffusion sensitization
(b value) is determined by duration () and strength (height) of sensi-
tizing pulsed-gradients (G), and time interval between two pulsed
gradients (). RF =radiofrequency.
D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
j
r
o
n
l
i
n
e
.
o
r
g

b
y

1
9
0
.
1
6
9
.
9
9
.
1
1
5

o
n

0
2
/
2
0
/
1
4

f
r
o
m

I
P

a
d
d
r
e
s
s

1
9
0
.
1
6
9
.
9
9
.
1
1
5
.

C
o
p
y
r
i
g
h
t

A
R
R
S
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
;

a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d


Diffusion Tensor MR Imaging of the Brain

AJ R:178, J anuary 2002

5

the size of the ellipsoid [18]. The mathematic
relationship between the principal coordinates
of the ellipsoid and the laboratory frame is de-
scribed by the eigenvectors (v

1

, v

2

, v

3

). The el-
lipsoids surface represents the root mean
square diffusive displacement of free water in
anisotropic media.
From diffusion tensor imaging, the eigenvec-
tors 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. 6). 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-di-
mensional color maps, three-dimensional white
matter ber tract maps can be created that are
based on similarities between neighboring vox-
els in the shape (quantitative diffusion anisot-
ropy measures) and orientation (principal
eigenvector map) of the diffusion ellipsoid [19,
21] (Fig. 6). The algorithms used to generate
these maps are detailed in the next section.

Methodology

Diffusion tensor MR imaging is the only
noninvasive in vivo method for mapping
white matter ber tract trajectories in the hu-
man brain. However, this process is techni-
cally and mathematically demanding and
requires several advances in the elds of diffu-
sion-weighted MR imaging and data process-
ing [21, 2730]. First, high-quality MR
images are mandatory. Recent MR scanner
hardware improvements, including better sta-
bility and homogeneity of the main magnetic
eld, as well as stronger and faster magnetic
gradients, have been imperative for the acqui-
Fig. 3.Graph of natural logarithmof signal intensity (Ln [S /
S
0
]) fromdiffusion-weighted images with different degrees of
diffusion weighting (b value) allows determination of apparent
diffusion coefcient (ADC) based on linear t of data points. Ab-
solute value of slope of plotted line, and thus ADC, is greater for
cerebrospinal uid (CSF) than for gray matter (GM).
C
Fig. 4.Diffusion tensor MR images in 20-year-old healthy man.
A, Reference T2-weighted spin-echo echoplanar MR image (TR/TE, 5000/92; b value, 0).
B,Diffusion-weighted spin-echo echoplanar MR image (5000/92; b value, 600 sec/mm
2
).
C,Apparent diffusion coefcient map can be calculated fromAand B.
B A
D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
j
r
o
n
l
i
n
e
.
o
r
g

b
y

1
9
0
.
1
6
9
.
9
9
.
1
1
5

o
n

0
2
/
2
0
/
1
4

f
r
o
m

I
P

a
d
d
r
e
s
s

1
9
0
.
1
6
9
.
9
9
.
1
1
5
.

C
o
p
y
r
i
g
h
t

A
R
R
S
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
;

a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d


6

AJ R:178, J anuary 2002

Melhem et al.

sition of high-quality (signal-to-noise ratio
and spatial resolution) diffusion-weighted MR
images in a reasonable time. Furthermore,
several schemes to reduce commonly encoun-
tered artifacts related to motion, eddy cur-
rents, and eld inhomogeneity have also been
critical for the success of diffusion tensor MR
imaging. Second, a robust mathematic frame-
work capable of generating a smooth repre-
sentation of the macroscopic white matter
ber 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 fol-
lowing or tracking of the individual white
matter ber tracts must be created.

Data Acquisition

The process of data acquisition in diffusion
tensor MR imaging consists of diffusion sensi-
tization and spatial encoding. Diffusion tensor
imaging is inuenced by the strength, number,
and orientation (diffusion tensor encoding
scheme) of the sensitizing gradients. Regard-
ing the strength of the diffusion-sensitizing
Fig. 5.Drawings show diffusion of water
molecule.
A, Diffusion in environment with strongly
aligned white matter bers (high anisotropy).
B, Diffusion ellipsoid in three-dimensional
space.
B A
Fig. 6.Diffusion tensor MR images in 30-year-old
healthy man.
A,Fractional anisotropy map describes degree of dif-
fusion 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.
B,Vector map coregistered onto T2-weighted MR im-
age (TR/TE, 5000/92) describes orientation of principal
eigenvector in each voxel.
C,Color-coded white matter ber maps are generated on
basis of fractional anisotropy and vector maps. cc =cor-
pus callosum, slf =superior longitudinal fasciculus, ilf =in-
ferior longitudinal fasciculus, cst =corticospinal tract.
D, Three-dimensional ber tracts are generated on
basis of fractional anisotropy and vector maps.
B A
D C
D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
j
r
o
n
l
i
n
e
.
o
r
g

b
y

1
9
0
.
1
6
9
.
9
9
.
1
1
5

o
n

0
2
/
2
0
/
1
4

f
r
o
m

I
P

a
d
d
r
e
s
s

1
9
0
.
1
6
9
.
9
9
.
1
1
5
.

C
o
p
y
r
i
g
h
t

A
R
R
S
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
;

a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d


Diffusion Tensor MR Imaging of the Brain

AJ R:178, J anuary 2002

7
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 eigen-
vector in each image voxel is indicated by an arrow. On basis of dened thresholds,
tracking (long curved arrows) is along voxels with similar measures of anisotropy and di-
rection of principal eigenvector. Algorithmcan distinguish between tracts A and B be-
cause 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.
Fig. 8.Diffusion tensor MR images in 26-year-old healthy
man. (Reprinted with permission from[82])
A, Three-dimensional white matter ber tracking of fronto-
pontine (FPT, green), corticospinal (CST, red), and temporo-
parietooccipitopontine (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 sep-
aration between tracts traversing cerebral peduncle be-
cause of closeness and similarities in direction and
anisotropy. Another ROI at level of lower pons, below termi-
nation of FPT and TPOPT (purple arrow), allows separation
of tracts.
B, Three-dimensional white matter ber tracking of FPT
(green), CST (red), and TPOPT (blue) tracts coregistered
onto axial T2-weighted MR images (5000/92).
C, Anatomic drawing at level of midbrain shows good
agreement with tracking method (B).
B
A
C
D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
j
r
o
n
l
i
n
e
.
o
r
g

b
y

1
9
0
.
1
6
9
.
9
9
.
1
1
5

o
n

0
2
/
2
0
/
1
4

f
r
o
m

I
P

a
d
d
r
e
s
s

1
9
0
.
1
6
9
.
9
9
.
1
1
5
.

C
o
p
y
r
i
g
h
t

A
R
R
S
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
;

a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d


8

AJ R:178, J anuary 2002

Melhem et al.

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/mm

2

[32]. Regarding the number of direc-
tions and the orientation of the diffusion-sensi-
tizing gradients, more uniform scanning of
three-dimensional space has been shown to re-
sult in a reduction in noise in raw diffusion-en-
coded measurements and to improve the
accuracy of diffusion tensor estimates and -
ber tracking applications. Uniform scanning of
three-dimensional space is achieved by in-
creasing the number of sensitizing gradient di-
rections (up to 642 directions) and optimizing
their spatial orientation [33, 34].
Recently, on the basis of analytic compari-
sons and Monte Carlo simulations, Hasan et al.
[35] have found, using a single-tensor model
of diffusion, that there is no substantial advan-
tage to implementing more than six encoding
directions as long as the orientation of the gra-
dients (diffusion tensor encoding schemes) is
optimized. Those authors also addressed the
inuence of various diffusion tensor encoding
schemes, including heuristic, numerically opti-
mized, and geometric polyhedra, on the accu-
racy of diffusion tensor estimates; they found
that the commonly implemented six-direc-
tional heuristic schemes are suboptimal [35].
A variety of spatial encoding schemes have
been proposed for diffusion tensor MR imag-
ing. The most commonly used in practice are
based on echoplanar readout and include sin-
gle-shot (so-called snapshot) and multishot (or
interleaved, segmented) techniques. The ad-
vantages 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 respira-
tion, cerebrovascular and cerebrospinal uid
ow, eye motion, and involuntary head motion.
In diffusion-sensitive multishot echoplanar im-
aging, 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-in-
duced error [23]. To reduce motion-related arti-
facts in diffusion-sensitive multishot echoplanar
techniques, cardiac gating and motion correc-
tion schemes such as navigator echo have been
implemented [3640]. Navigator echo correc-
tion, which is a nonphase-encoded readout
gradient echo, provides a measure of the mo-
tion-induced phase variations between the ech-
oes 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 reso-
lution 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-
Fig. 9.Schematic shows difference between tracking meth-
ods. Maintaining information on intercepts is achieved by per-
forming tracking in a continuous number eld, which results in
precise white matter ber tracking. Voxels fromwhich tracking
is launched are designated by asterisks, direction of principal
eigenvector in each voxel is designated by an arrow, and vox-
els connected by tracking algorithmare indicated by boxes
with shaded arrows. Dark arrows indicate interrupted trajec-
tory of track. (Reprinted with permission from[19])
A and B, Tracking (long curved arrows) without (A) and with (B)
keeping information on intercepts when tracking leaves voxel.
B A
Fig. 10.Diffusion tensor MR images in 35-year-old
healthy man. Three-dimensional ber tracking of brain-
stemwith ventral (purple) and medial (turquoise) compo-
nents of middle cerebellar peduncles, inferior cerebellar
peduncle (green), superior cerebellar peduncle (yellow),
medial longitudinal fasciculus (orange), and cortico-
pontospinal (red) tracts are superimposed on these two
images. (Reprinted with permission from[82])
A, Mid sagittal T2-weighted MR image (TR/TE, 5000/92) of
brainstemand cerebellum.
B, Photograph of anatomic specimen shows postmortem
dissection of brainstemand cerebellum.
B A
D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
j
r
o
n
l
i
n
e
.
o
r
g

b
y

1
9
0
.
1
6
9
.
9
9
.
1
1
5

o
n

0
2
/
2
0
/
1
4

f
r
o
m

I
P

a
d
d
r
e
s
s

1
9
0
.
1
6
9
.
9
9
.
1
1
5
.

C
o
p
y
r
i
g
h
t

A
R
R
S
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
;

a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d


Diffusion Tensor MR Imaging of the Brain

AJ R:178, J anuary 2002

9

and spin-echo (GRASE), fast spin-echo, and
line scanning [42, 43]. Both GRASE and fast
spin-echo techniques are resistant to distor-
tions from static eld inhomogeneities and
diffusion sensitization gradient-induced eddy
currents compared with single-shot echopla-
nar 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 col-
umn-selective excitation techniques are occa-
sionally 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 inher-
ently 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 im-
ages 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 im-
aging protocol consists of a diffusion-sensitive,
cardiac-gated, navigator echocorrected, two-
dimensional multishot spin-echo echoplanar
readout. The TE is xed at 92 msec, whereas
the TR varies with the heart rate (50006000
msec). The number of echoes acquired in each
TR interval is 18 (1 navigator echo and 17
phase-encoded echoes). The eld of view
ranges from 12



12 to 25


25 cm, depending
on the head size. The number of sampling
points along frequency- and phase-encoding di-
rections ranges from 64 to 128 interpolated to
128 to 256 (zero lling) to maintain a pixel size
of 0.98 mm

2

. Forty interleaved, 3-mm-thick
slices are acquired in the axial plane using the
multislice mode. Diffusion sensitization is per-
formed along six noncolinear directions (heuris-
tic 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/
mm
2
at the maximum gradient strength of 2.1
G/cm. A reference image with low diffusion
weighting (b = 33 sec/mm
2
) is also recorded. A
single set of these seven measurements takes 4
5 min depending on heart rate. These measure-
ments 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 echopla-
nar 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 coreg-
istration process. The entire examination re-
quires approximately 50 min.
Data Correction and Processing
Misregistration of diffusion tensor MR im-
ages has deleterious effects on spatial resolu-
tion 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
eld inhomogeneity resulting from imperfect
shimming and differences in magnetic proper-
ties of adjacent tissues and from diffusion sen-
sitization gradient-induced eddy currents [41,
44, 46]. Unwarping algorithms, using inhomo-
geneity eld maps, are used for the correction of
distortion originating from the static magnetic
eld inhomogeneities [41]. For the correction of
diffusion sensitization gradient-induced eddy
current distortions, two methods are often used.
The rst method entails the application of least
squares straight line ts and cross-correlation
functions to the diffusion-weighted image [46,
47]. The second method uses a one-dimen-
sional inhomogeneity eld map in the read and
phase-encoding directions in each slice and
each diffusion-sensitizing direction step to cor-
rect the distortion [41]. Motion-related mis-
alignment can be corrected by computer-based
algorithms, such as Automated Image Regis-
tration 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 direc-
tions, head position, and rotation) indexes can
be derived that measure distinct intrinsic fea-
tures 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 diffu-
sion coefcients measured along the three prin-
cipal 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 2
h
/ (
m
+
l
),
which measure the relative magnitudes of the
diffusivities along the longest axis of the diffu-
sion ellipsoid and other orthogonal axes. Here
the eigenvalues are sorted in order of decreasing
magnitude (
h
: highest diffusivity;
m
: interme-
diate diffusivity; and
l
: lowest diffusivity). Be-
cause 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 ratiothe ratio of the
Fig. 11.9-year-old boy with cerebral X-linked adrenoleukodystrophy who presented with progressive motor
and visual decits.
A, T2-weighted MR image (TR/TE, 5000/92) shows abnormal symmetric hyperintensity involving peritrigonal
white matter.
B, 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).
B A
D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
j
r
o
n
l
i
n
e
.
o
r
g

b
y

1
9
0
.
1
6
9
.
9
9
.
1
1
5

o
n

0
2
/
2
0
/
1
4

f
r
o
m

I
P

a
d
d
r
e
s
s

1
9
0
.
1
6
9
.
9
9
.
1
1
5
.

C
o
p
y
r
i
g
h
t

A
R
R
S
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
;

a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d

10 AJ R:178, J anuary 2002
Melhem et al.
volume of an ellipsoid to the volume of a
sphere whose radius is the averaged diffusiv-
ityranges between 0 and 1, where 0 means
highest anisotropy and 1 indicates complete
isotropic diffusion. Relative anisotropythe
ratio of the variance of the eigenvalues to their
meanand fractional anisotropythe ratio of
the anisotropic component of the diffusion ten-
sor to the whole diffusion tensorboth vary
between 0 (isotropic diffusion) and 1 (innite
anisotropic diffusion).
Simulations have shown a rapid divergence
of the eigenvalues and invariant anisotropy in-
dexes from true values as the signal-to-noise ra-
tio decreases [49, 50]. In tissues with low-level
anisotropy, this divergence is further exagger-
ated 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 mul-
tiple diffusion-weighted images are tted us-
ing multivariant linear least square tting to
obtain the six elements of the symmetric diffu-
sion tensor. The diffusion tensors at each voxel
are diagonalized to obtain eigenvalues and
eigenvectors for each voxel. The eigenvector
(v
1
) associated with the largest eigenvalue (
l
)
is used to represent the local ber 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 vec-
tor elements of the eigenvector v
1
for each
voxel. The absolute values of the vector ele-
ments are assigned to red (x element), green (y
element), and blue (z element) [12]. If the prin-
cipal eigenvector is aligned along the x-axis,
pure red is assigned to the corresponding
voxel, whereas if the eigenvector is 45 be-
tween 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. 6).
White Matter Fiber Tracking
Methods to reconstruct white matter tracts
can be placed into two broad categories [19, 21,
5154]. Methods in the rst category are based
on line propagation algorithms that use local
tensor information for each step of the propaga-
tion. Simple line propagation techniques that
connect voxels on the basis of discrete number
elds (local principal eigenvector orientation)
are incapable of providing accurate representa-
tion of white matter tracts [19]. Improvements
in line propagation techniques using continu-
ous, rather than discrete, number elds can
provide connections that follow the actual
white matter tract [19, 52]. Furthermore, line
propagation techniques can be modied to
create a smooth (curved) path by interpolat-
ing 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 nd a path
between two predetermined voxels with mini-
mum energy violation. This category includes
the methods of fast marching (Parker GJ et al.,
presented at the International Society of Mag-
netic Resonance meeting, April 2000) and
simulated annealing (Tuch DS et al., presented
at the ISMRM meeting, April 2001).
In this review, ber tracking is performed
using a line propagation technique based on
continuous number elds developed at our
center called ber assignment using continu-
ous tracking [19]. Tracking is launched from
a seed voxel from which a line is propagated in
Fig. 12.6-year-old boy with anterior form of X-linked adrenoleukodystrophy who presented with personality
changes. Diffusion-weighted images show three-dimensional ber tracking of corpus callosumcoregistered onto im-
ages 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 callosumand in both frontal lobes compared with healthy boy.
A and B,Axial (A) and sagittal (B) T2-weighted MR images (TR/TE, 5000/92) of affected boy.
C and D, Axial (C) and sagittal (D) T2-weighted MR images (5000/92) of healthy boy.
B A
D C
D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
j
r
o
n
l
i
n
e
.
o
r
g

b
y

1
9
0
.
1
6
9
.
9
9
.
1
1
5

o
n

0
2
/
2
0
/
1
4

f
r
o
m

I
P

a
d
d
r
e
s
s

1
9
0
.
1
6
9
.
9
9
.
1
1
5
.

C
o
p
y
r
i
g
h
t

A
R
R
S
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
;

a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d

Diffusion Tensor MR Imaging of the Brain
AJ R:178, J anuary 2002 11
both retrograde and antegrade directions ac-
cording to the principal eigenvector at each
voxel. Tracking propagates on the basis of the
orientation of the eigenvector that is associ-
ated with the largest eigenvalue. Tracking is
terminated when it reaches a voxel with frac-
tional anisotropy lower than a threshold of
0.250.35 and when the angle between the
two principal eigenvectors is greater than 35
40 (Fig. 7).
Fractional anisotropy thresholds help to ex-
clude 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 reli-
ability of tracking the corticospinal tract using
different thresholds.
Knowing the ber projections in relation to
anatomic landmarks, we dene multiple re-
gions 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 difcult to separate the different cortico-
pontospinal tracts; placing another region of
interest in the lower pons facilitates their
separation (Fig. 8).
Tracking can be achieved in one of two
ways. The rst 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 in-
terest (if the region of interest contains 10 vox-
els, only 10 tracking results delineate the tract).
In the second method, ber tracking is initiated
from the center of every voxel in the brain, but
only bers passing through the specied re-
gions of interest are retained [51]. In this ap-
proach, multiple tracts penetrate the region of
interest, thus revealing a more comprehensive
structure of the tract.
One of the difculties in tracking a ber
path from the three-dimensional vector matrix
is to translate the discrete vector information to
continuous ber tracts. The ber-assignment-
using-continuous-tracking algorithm main-
tains the information of intercept when the
tracking exits a voxel by performing the track-
ing in a continuous number eld rather than in
a discrete number eld (Fig. 9).
The nal step of this process is coregistering
the three-dimensional white matter ber tracts
on the anatomic T2-weighted images using Au-
tomated Image Registration software.
Applications
Diffusion tensor imaging data promise a
range of applications in clinical medicine. Ini-
tially, diffusion anisotropy was observed in skel-
etal muscle [55]. Since then, diffusion tensor
measures have been used to study brous orga-
nized structures like human and animal spinal
cords [56], myocardium [57, 58], intervertebral
discs [59], and cerebral white matter [60].
Clinically, properties derived from the diffu-
sion tensor like the trace, which reects overall
water content, have been used successfully to
evaluate brain ischemia [61, 62]. Measures of
the diffusion tensor have also been used to in-
vestigate brain development [17, 63] and to aid
in neurosurgical planning. In addition, param-
eters derived from the diffusion tensor, such as
anisotropy indexes, have been used to evaluate
white matter disease in Krabbes disease [64],
cerebral adrenoleukodystrophy [65], AIDS
[66], multiple sclerosis [20, 67, 68], hyperten-
sive encephalopathy [69], age-related changes
[70], schizophrenia [71], Alzheimers disease
[72], ischemic leukoaraiosis [73], and epilepsy
[74]. Other studies have also probed the poten-
tial of diffusion tensor MR imaging in brain tu-
mors [75], migraines [76], and eclampsia [77].
Requisites for sound application of diffu-
sion tensor MR imaging and white matter trac-
tography in central nervous system disease are
the establishment of normative anisotropy val-
ues and white matter tract maps, and the vali-
dation of the normative white matter tract
maps on the basis of detailed anatomic knowl-
edge of these tracts. As illustrated in Figures 6
and 10, 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 difcult be-
cause of a lack of an appropriate gold standard.
In fact, diffusion tensor MR imaging is the
Fig. 13.6-year-old boy with cerebral palsy resulting fromperiventricular leukomalacia who presented with
asymmetric spastic diplegia affecting left side more than right.
AD, Color maps of brainstemwhite matter tracts show decrease in size of corticopontospinal tracts in affected
boy (arrowheads, Aand B) compared with those shown on color maps of brainstemwhite matter tracts in 8-year-
old healthy boy (arrowheads, Cand D). Furthermore, right corticopontospinal tract is more involved than left in
affected boy, which correlates with his neurologic examination.
B A
D C
D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
j
r
o
n
l
i
n
e
.
o
r
g

b
y

1
9
0
.
1
6
9
.
9
9
.
1
1
5

o
n

0
2
/
2
0
/
1
4

f
r
o
m

I
P

a
d
d
r
e
s
s

1
9
0
.
1
6
9
.
9
9
.
1
1
5
.

C
o
p
y
r
i
g
h
t

A
R
R
S
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
;

a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d

12 AJ R:178, J anuary 2002
Melhem et al.
only method available that has the potential for
tracking white matter tracts in vivo. Attempts
at in vitro validation based on white matter his-
tology are limited because of geometric distor-
tions resulting from dissection, freezing,
dehydration, xation, 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 ac-
quisition (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 imper-
ative for proper interpretation of the maps.
Specically, misregistration of diffusion tensor
MR images caused by eddy currents, ghosting
due to motion, and signal loss due to suscepti-
bility variations can all affect the computed
trajectory of the ber tract. Furthermore, the
tracking algorithm may fail to provide an ac-
curate representation of the ber trajectory
when a voxel contains nonuniformly distrib-
uted bers, curved bers, or two or more inter-
digitating ber populations. This failure is
largely because the direction of the measured
principal eigenvector is based on a voxel aver-
age and does not necessarily represent the tra-
jectories of individual microscopic tracts [21].
At our center, diffusion tensor MR imaging
and white matter tractography are being ap-
plied in three major clinical research projects.
The rst project studies children with a cere-
bral form of adrenoleukodystrophy, a genetic
disorder that involves the white matter, adrenal
cortex, and testes [78]; the second studies pa-
tients with spastic cerebral palsy resulting
from periventricular leukomalacia, a domi-
nant pattern of anoxic brain injury in prema-
ture 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 re-
versible and irreversibly damaged white matter,
and categorize affected white matter on the ba-
sis of well-dened histopathologic zones (Fig.
11). Information derived from ADC and frac-
tional anisotropy measures in these patients
may also shed light on axonal and myelin ultra-
structure contributions to the anisotropy phe-
nomena. Preliminary results show that affected
white matter can be divided into three distinct
zones on the basis of differences in diffusion
values that may reect varying degrees of ax-
onal and myelin loss (Table 1). The effect of
white matter lesions on specic commissural
(corpus callosum) (Fig. 12), projectional (cor-
ticospinal), and association tracts is currently
being correlated with neurologic and neuro-
psychologic function. Similar hypotheses are
being tested in patients with periventricular
leukomalacia. The severity of injury to the
periventricular white matter determines the de-
gree of damage to white matter tracts travers-
ing the injured zone (Fig. 13). In particular,
correlations between damage to different trans-
callosal cortical connections and types of neu-
ropsychologic decits are being evaluated. The
severity of spastic diplegia is also being corre-
lated with the degree of injury to traversing cor-
ticospinal tracts.
In children with holoprosencephaly, anoma-
lous white matter tracts are identied in addition
Fig. 14.1-year-old boy with middle hemispheric variant of holoprosencephaly.
A, 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.
B, Coronal fast phase-sensitive inversion-recovery MR image (3300/40; inversion time, 200msec) through frontal horns
shows noncleavage of cerebral hemispheres with continuation of anomalous gray and white matter (arrowheads)
across midline. Note absence of septumpellucidum.
C, 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.
D, 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).
B A
D C
D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
j
r
o
n
l
i
n
e
.
o
r
g

b
y

1
9
0
.
1
6
9
.
9
9
.
1
1
5

o
n

0
2
/
2
0
/
1
4

f
r
o
m

I
P

a
d
d
r
e
s
s

1
9
0
.
1
6
9
.
9
9
.
1
1
5
.

C
o
p
y
r
i
g
h
t

A
R
R
S
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
;

a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d

Diffusion Tensor MR Imaging of the Brain
AJ R:178, J anuary 2002 13
to the well-described noncleavage of cortical
and deep gray matter structures. Rigorous com-
parison with age-matched normative white mat-
ter tract maps is still required for better
understanding of these complex anomalies. De-
ning white and gray matter anomalies may al-
low more precise prediction of developmental
outcome in holoprosencephaly (Fig. 14).
Other applications of diffusion tensor MR
imaging and white matter tractography with
more immediate clinical impact are the eval-
uation of brain tumors and acute stroke. De-
ning the relationship of brain tumors to
eloquent white matter tracts will undoubt-
edly help guide the surgical approach and the
extent of the resection (Fig. 15). The true ex-
tent of inltration of neoplasms along white
matter tracts may be better delineated with
these techniques. With respect to stroke, pre-
liminary results have shown an immediate
increase in fractional anisotropy in regions of
reversible acute ischemia (decreased diffu-
sion), which may be related to reduction in
ow. Further investigation of this nding and
its use in guiding acute stroke management
is warranted.
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. Im-
provements in signal-to-noise ratios continue to
be required for more precise calculation of
anisotropy measures and more accurate white
matter ber tracking. Imaging with isotropic,
high-resolution voxels reduces the effect of vol-
ume averaging on the direction of the principal
eigenvector, which provides a better representa-
tion of the actual orientation of the ber 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 al-
Fig. 15.T2-weighted MR images (TR/
TE, 5000/92) in 48-year-old woman with
left-sided posterior fossa meningioma.
(Reprinted with permission from[82])
A,Mid sagittal image shows levels of axial
sections (B, C, and D)and slightly hyperin-
tense (compared with brain) extraaxial
mass compressing brainstem.
BD, Axial images at level of third ventri-
cle (B), anterior commissure (C), and
brainstem (D) show coregistered left
(red) and right (yellow) corticopontospi-
nal tracts. Note relationship of extraaxial
mass to posteriorly displaced left corti-
cospinal tract.
E, Three-dimensional representation of
left corticospinal tract superimposed
onto sagittal and coronal images better
shows relationship between mass
(green) and tract (red).
B A
D C
E
Note.Data are averages and SDs, in units of 10
3
mm
2
/
sec. NAWM =normal-appearing white matter.
TABLE 1
Isotropic Apparent Diffusion
Coefficient (ADC
i
) and
Fractional Anisotropy (FA)
Values
White Matter Zone ADC
i
FA
NAWM 0.66 (0.12) 0.52 (0.16)
Area A 0.84 (0.16) 0.31 (0.12)
Area B 1.36 (0.19) 0.22 (0.07)
Area C 1.73 (0.15) 0.13 (0.05)
D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
j
r
o
n
l
i
n
e
.
o
r
g

b
y

1
9
0
.
1
6
9
.
9
9
.
1
1
5

o
n

0
2
/
2
0
/
1
4

f
r
o
m

I
P

a
d
d
r
e
s
s

1
9
0
.
1
6
9
.
9
9
.
1
1
5
.

C
o
p
y
r
i
g
h
t

A
R
R
S
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
;

a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d

14 AJ R:178, J anuary 2002
Melhem et al.
lows 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 efciency of the acquisition, a scheme is
needed that is based on real-time navigator
echo correction. The scheme consists of contin-
uously identifying and reacquiring only the cor-
rupted portions of each acquisition. This
scheme improves the efciency 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 re-
dundancy in the acquisition.
Fiber tracking algorithms, such as diffu-
sion spectrum techniques that may resolve -
ber orientation heterogeneity in a voxel, are
needed to accurately dene crossing and dis-
persing white matter tracts [81]. Algorithms
that take into consideration the medium and
minor eigenvectors, in addition to the princi-
pal eigenvector, may better resolve intra-
voxel ber orientation. Objective methods
for comparing white matter tracts in patients
with normative templates, for quantifying the
dimensions of various white matter tracts,
and for correlating these dimensions with
cognitive function, are also required for vali-
dation of this technique.
Finally, an exciting application of this
technique is in the clarication of the tempo-
ral relationships between loci of signal
change in functional MR imaging experi-
ments. Understanding the temporal sequence
of regional activations is an important part of
understanding how, rather than where, a cog-
nitive process is executed. Because of the
substantial disparity between the relatively
slow evolution of functional MR imaging
signal changes and the speed of the underly-
ing neural processes, functional MR imaging
may not be able to differentiate between sec-
ondary and higher order sites in the activa-
tion cascade. Mapping of white matter tracts
may help divide apparently secondary activa-
tion sites into true secondary and higher or-
der sites on the basis of the nature of their
connections to the primary site of activation.
References
1. Carr HY, Purcell EM. Effects of diffusion on free
precession in nuclear magnetic resonance experi-
ments. Phys Rev 1954;94:630638
2. Paulson OB, Hertz MM, Bolwig TG, Lassen NA. Fil-
tration and diffusion of water across the blood-brain
barrier in man. Microvasc Res 1977;13:113124
3. Taylor DG, Bushell MC. The spatial mapping of
translational diffusion coefcients by the NMR im-
aging techniques. Phys Med Biol 1985;30:345349
4. Moseley ME, Kucharczyk J, Mintorovitch J, et al.
Diffusion-weighted MR imaging of acute stroke:
correlation with T2-weighted and magnetic suscep-
tibility-enhanced MR imaging in cats. AJNR
1990;11:423429
5. Mintorovitch J, Moseley ME, Chileuitt L, Shimizu
H, Cohen Y, Weinstein PR. Comparison of diffu-
sion- and T2-weighted MRI for the early detection
of cerebral ischemia and reperfusion in rats. Magn
Reson Med 1991;18:3950
6. Minematsu K, Li L, Fisher M, Sotak CH, Davis MA,
Fiandaca MS. Diffusion-weighted magnetic reso-
nance imaging: rapid and quantitative detection of
focal brain ischemia. Neurology 1992;42:235240
7. van Gelderen P, de Vleeschouwer MHM, des Pres
D, Pekar J, van Zijl PCM, Moonen CTW. Water dif-
fusion and acute stroke. Magn Reson Med 1994;31:
154163
8. Sorensen AG, Buonanno FS, Gonzalez RG, et al.
Hyperacute stroke: evaluation with combined multi-
section diffusion-weighted and hemodynamically
weighted echo-planar MR imaging. Radiology
1996;199:391401
9. Chenevert TL, Brunberg JA, Pipe JG. Anisotropic
diffusion in human white matter: demonstration
with MR techniques in vivo. Radiology 1990;177:
401405
10. Moseley ME, Cohen Y, Kucharczyk J, et al. Diffu-
sion-weighted MR imaging of anisotropic water dif-
fusion in cat central nervous system. Radiology
1990;176:439445
11. Doran M, Hajnal JV, Van Bruggen N, King MD,
Youn IR, Bydder GM. Normal and abnormal white
matter tracts shown by MR imaging using direc-
tional diffusion weighted sequences. J Comput As-
sist Tomogr 1990;14:865873
12. Douek P, Turner R, Pekar J, Patronas N, Le Bihan D.
MR color mapping of myelin ber orientation. J
Comput Assist Tomogr 1991;15:923929
13. Sakuma H, Nomura Y, Takeda K, et al. Adult and
neonatal human brain: diffusional anisotropy and
myelination with diffusion-weighted MR imaging.
Radiology 1991;180:229233
14. Coremans J, Luypaert R, Verhelle F, Stadnik T, Os-
teaux M. A method for myelin ber orientation
mapping using diffusion-weighted MR images.
Magn Reson Imaging 1994;12:443454
15. Nomura Y, Sakuma H, Takeda K, Tagami T, Okuda
Y, Nakagawa T. Diffusional anisotropy of the hu-
man brain assessed with diffusion-weighted MR: re-
lation with normal brain development and aging.
AJNR 1994;15:231238
16. Ono J, Harada K, Mano T, Sakurai K, Okada S. Dif-
ferentiation of dys- and demyelination using diffu-
sional anisotropy. Pediatr Neurol 1997;16:6366
17. Neil JJ, Shiran SI, McKinstry RC, et al. Normal
brain in human newborns: apparent diffusion and
diffusion anisotropy measured by using diffusion
tensor MR imaging. Radiology 1998;209:5766
18. Shimony JS, McKinstry RC, Akbudak E, et al.
Quantitative diffusion-tensor anisotropy brain MR
imaging: normative human data and anatomic anal-
ysis. Radiology 1999;212:770784
19. Mori S, Crain BJ, Chacko VP, van Zijl PCM. Three-
dimensional tracking of axonal projections in the
brain by magnetic resonance imaging. Ann Neurol
1999;45:265269
20. Bammer R, Augustin M, Strasser-Fuchs S, et al.
Magnetic resonance diffusion tensor imaging for
characterizing diffuse and focal white matter abnor-
malities in multiple sclerosis. Magn Reson Med
2000;44:583591
21. Basser PJ, Pajevic S, Pierpaoli C, Duda J, Aldroubi
A. In vivo ber tractography using DT-MRI data.
Magn Reson Med 2000;44:625632
22. Le Bihan D, Breton E, Lallemand D, Aubin MI,
Vignaud J, Laval-Jeantet M. Separation of diffusion
and perfusion in intravoxel incoherent motion MR
imaging. Radiology 1988;168:497505
23. Turner R, Le Bihan D, Maier J, Vavrek R, Hedges
LK, Pekar J. Echo-planar imaging of intravoxel in-
coherent motion. Radiology 1990;177:407414
24. Stejskal EO, Tanner JE. Spin diffusion measure-
ments: spin echoes in the presence of a time-depen-
dent eld gradient. J Chem Phys 1965;42:288292
25. Pierpaoli C, Jezzard P, Basser PJ, Barnett A,
DiChiro G. Diffusion tensor MR imaging of the hu-
man brain. Radiology 1996;201:637648
26. Sorensen AG, Wu O, Copen WA, et al. Human acute
cerebral ischemia: detection of changes in water dif-
fusion anisotropy by using MR imaging. Radiology
1999;212:785792
27. Papadakis NG, Martin KM, Pickard JD, Hall LD,
Carpenter TA, Huang CL. Gradient pre-emphasis
calibration in diffusion-weighted echo-planar imag-
ing. Magn Reson Med 2000;44:616624
28. Basser PJ, Pajevic S. Statistical artifacts in diffusion
tensor MRI (DT-MRI) caused by background noise.
Magn Reson Med 2000;44:4150
29. Skare S, Li TQ, Nordell B, Ingvar M. Noise consid-
erations in the determination of diffusion tensor
anisotropy. Magn Reson Imaging 2000;18:659669
30. Parker GJM, Schnabel JA, Symms MR, Werring DJ,
Barker GJ. Nonlinear smoothing for reduction of
systematic and random errors in diffusion tensor im-
aging. J Magn Reson Imaging 2000;11:702710
31. Melhem ER, Itoh R, Jones L, Barker PB. Diffusion
tensor MR imaging of the brain: effect of diffusion
weighting on trace and anisotropy measurements.
AJNR 2000;21:18131820
32. Burdette JH, Elster AD, Ricci PE. Calculation of ap-
parent diffusion coefcients (ADCs) in brain using
two-point and six-point methods. J Comput Assist
Tomogr 1998;22:792794
33. Jones DK, Horseld MA, Simmons A. Optimal
strategies for measuring diffusion in anisotropic sys-
tems by magnetic resonance imaging. Magn Reson
Med 1999;42:515525
34. Papadakis NG, Xing D, Huang CLH, Hall LD, Car-
penter TA. A comparative study of acquisition
schemes for diffusion tensor imaging using MRI.
J Magn Reson 1999;137:6782
35. Hasan KM, Parker DL, Alexander AL. Comparison
of gradient encoding schemes for diffusion-tensor
MRI. J Magn Reson Imaging 2001;13:769780
36. Brockstedt S, Thomsen C, Wirestam R, Holtas S,
Stahlberg F. Quantitative diffusion coefcient maps
using fast spin-echo MRI. Magn Reson Imaging
1998;16:877886
37. Mori S, van Zijl PC. A motion correction scheme by
D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
j
r
o
n
l
i
n
e
.
o
r
g

b
y

1
9
0
.
1
6
9
.
9
9
.
1
1
5

o
n

0
2
/
2
0
/
1
4

f
r
o
m

I
P

a
d
d
r
e
s
s

1
9
0
.
1
6
9
.
9
9
.
1
1
5
.

C
o
p
y
r
i
g
h
t

A
R
R
S
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
;

a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d

Diffusion Tensor MR Imaging of the Brain
AJ R:178, J anuary 2002 15
twin-echo navigation for diffusion-weighted mag-
netic resonance imaging with multiple RF echo ac-
quisition. Magn Reson Med 1998;40:511516
38. Anderson AW, Gore JC. Analysis and correction of
motion artifacts in diffusion weighted imaging.
Magn Reson Med 1994;32:379387
39. Ordidge RJ, Helpern JA, Qing ZX, Knight RA,
Nagesh V. Correction of motional artifacts in diffu-
sion-weighted MR images using navigator echoes.
Magn Reson Imaging 1994;12:455460
40. Butts K, de Crespigny A, Pauly JM, Moseley M.
Diffusion-weighted interleaved echo-planar imag-
ing with a pair of orthogonal navigator echoes.
Magn Reson Med 1996;35:763770
41. Jezzard P, Barnett AS, Pierpaoli C. Characterization
of and correction for eddy current artifacts in echo
planar diffusion imaging. Magn Reson Med 1998;
39:801812
42. Finsterbusch J, Frahm J. Diffusion-weighted single-
shot line scan imaging of the human brain. Magn
Reson Med 1999;42:772778
43. Alsop DC. Phase insensitive preparation of single-
shot RARE: application to diffusion imaging in hu-
mans. Magn Reson Med 1997;38:527533
44. Itoh R, Melhem ER, Folkers PJ. Diffusion-tensor
MR imaging of the human brain with gradient- and
spin-echo readout: technical note. AJNR 2000;21:
15911595
45. Finsterbusch J, Frahm J. Diffusion tensor mapping
of the human brain using single-shot line scan imag-
ing. J Magn Reson Imaging 2000;12:388394
46. Haselgrove JC, Moore JR. Correction for distortion
of echo-planar images used to calculate the apparent
diffusion coefcient. Magn Reson Med 1996;36:
960964
47. Bastin ME. Correction of eddy current-induced arti-
facts in diffusion tensor imaging using iterative
cross-correlation. Magn Reson Imaging 1999;17:
10111024
48. Woods RP, Cherry SR, Mazziotta JC. Rapid auto-
mated algorithm for aligning and reslicing PET im-
ages. J Comput Assist Tomogr 1992;16:620633
49. Pierpaoli C, Basser PJ. Toward a quantitative assess-
ment of diffusion anisotropy. Magn Reson Med
1996;36:893906
50. Bastin ME, Armitage PA, Marshall I. A theoretical
study of the effect of experimental noise on the mea-
surement of anisotropy in diffusion imaging. Magn
Reson Imaging 1998;16:773785
51. Conturo TE, Lori NF, Cull TS, et al. Tracking neu-
ronal ber pathways in the living human brain. Proc
Natl Acad Sci U S A 1999;96:1042210427
52. Xue R, van Zijl PC, Crain BJ, Solaiyappan M, Mori
S. In vivo three-dimensional reconstruction of rat
brain axonal projections by diffusion tensor imag-
ing. Magn Reson Med 1999;42:11231127
53. Poupon C, Clark CA, Frouin V, et al. Regularization
of diffusion-based direction maps for the tracking of
brain white matter fascicles. Neuroimage 2000;12:
184195
54. Poupon C, Mangin J, Clark CA, et al. Towards infer-
ence of human brain connectivity from MR diffu-
sion tensor data. Med Image Anal 2001;5:115
55. Cleveland GG, Chang DC, Hazelwood CF. Nuclear
magnetic resonance measurements of skeletal mus-
cle: anisotropy of the diffusion coefcient of the in-
tracellular water. Biophys J 1976;16:104105
56. Lutsep HL, Albers GW, De Crespigny A, Kamat
GN, Marks MP, Moseley ME. Clinical utility of dif-
fusion-weighted magnetic resonance imaging in the
assessment of ischemic stroke. Ann Neurol 1997;41:
574580
57. Hsu EW, Muzikant AL, Matulevicius SA, Penland
RC, Henriquez CS. Magnetic resonance myocar-
dial ber-orientation mapping with direct histo-
logical correlation. Am J Physiol 1998;274:
H1627H1634
58. Reese TG, Weisskoff RM, Smith RN, Rosen BR,
Dinsmore RE, Wedeen VJ. Imaging myocardial -
ber architecture in vivo with magnetic resonance.
Magn Reson Med 1995;4:786791
59. Hsu EW, Setton LA. Diffusion tensor microscopy of
the intervertebral disc anulus brosus. Magn Reson
Med 1999;41:992999
60. Peled S, Gudbjartsson H, Westin CF, Kikinis R,
Jolesz FA. Magnetic resonance imaging showing
orientation and asymmetry of white matter ber
tracts. Brain Res 1998;780:2733
61. Moseley ME, Butts K, Yenari MA, Marks M, de
Crespigny A. Clinical aspects of DWI. NMR
Biomed 1995;8:8796
62. Warach S, Gaa J, Siewert B, Weilpolski P, Edelman
RR. Acute human stroke studied by whole brain
echo planar diffusion weighted magnetic resonance
imaging. Ann Neurol 1995;7:231241
63. Huppi PS, Maier SE, Peled S, et al. Microstructural
development of human newborn cerebral white mat-
ter assessed in vivo by diffusion tensor magnetic
resonance imaging. Pediatr Res 1998;44:584590
64. Guo AC, Petrella JR, Kurtzberg J, Provenzale JM.
Evaluation of white matter anisotropy in Krabbe
disease with diffusion tensor MR imaging: initial
experience. Radiology 2001;218:809815
65. Ito R, Melhem ER, Mori S, Eicher FS, Raymond
GV, Moser HW. Diffusion tensor brain MR imaging
in X-linked cerebral adrenoleukodystrophy. Neurol-
ogy 2001;56:544547
66. Pomara N, Crandall DT, Choi SJ, Johnson G, Lim
KO. White matter abnormalities in HIV-1 infection:
diffusion tensor imaging study. Psychiatry Res
2001;106:1524
67. Ciccarelli O, Werring DJ, Wheeler-Kingshott CA, et
al. Investigation of MS normal-appearing brain us-
ing diffusion tensor MRI with clinical correlations.
Neurology 2001;56:926933
68. Filippi M, Cercignani M, Inglese M, Horseld
MA, Comi G. Diffusion tensor magnetic reso-
nance imaging in multiple sclerosis. Neurology
2001;56:304311
69. Schwartz RB, Mulkern RV, Gudbjartsson H, Jolesz
F. Diffusion-weighted MR imaging in hypertensive
encephalopathy: clues to pathogenesis. AJNR 1998;
19:859862
70. Sullivan EV, Adalsteinsson E, Hedehus M, et al.
Equivalent disruption of regional white matter mi-
crostructure in aging healthy men and women. Neu-
roreport 2001;12:99104
71. Foong J, Maier M, Clark CA, Barker GJ, Miller DH,
Ron MA. Neuropathological abnormalities of the
corpus callosum in schizophrenia: a diffusion tensor
imaging study. J Neurol Neurosurg Psychiatry
2000;68:242244
72. Rose SE, Chen F, Chalk JB, et al. Loss of connectiv-
ity in Alzheimers disease: an evaluation of white
matter tract integrity with color-coded MR diffusion
tensor imaging. J Neurol Neurosurg Psychiatry
2000;69:528530
73. Jones DK, Lythgoe D, Horseld MA, Simmons A,
Williams SC, Markus HS. Characterization of white
matter damage in ischemic leukoaraiosis with diffu-
sion tensor MRI. Stroke 1999;30:393397
74. Eriksson SH, Rugg-Gunn FJ, Symms MR, Barker
GJ, Duncan JS. Diffusion tensor imaging in patients
with epilepsy and malformations of cortical devel-
opment. Brain 2001;124:627636
75. Krabbe K, Gideon P, Wagn P, et al. MR diffusion
imaging of human intracranial tumors. Neuroradiol-
ogy 1997;39:483489
76. Chabriat H, Vahedi K, Clark CA, et al. Decreased
hemispheric water mobility in hemiplegic migraine
related to mutation of CACNA1A gene. Neurology
2000;54:510512
77. Schaefer PW, Buonanno FS, Gonzalez RG,
Schwamm LH. Diffusion-weighted imaging dis-
criminates between cytotoxic and vasogenic
edema in a patient with eclampsia. Stroke 1997;
28:10821085
78. Melhem ER, Barker PB, Raymond GV, Moser HW.
X-linked adrenoleukodystrophy: review of genetic,
clinical, and MR imaging characteristics. AJR
1999;173:15751581
79. Melhem ER, Hoon AH, Ferrucci JT III, et al. Brain
MR imaging in periventricular leukomalacia: rela-
tionship between lateral ventricular volume and se-
verity of cognitive and motor impairment. Radiology
2000;214:199204
80. Fitz CR. Holoprosencephaly and related entities.
Neuroradiology 1983;25:225238
81. Wiegell MR, Larsson HBW, Wedeen VJ. Fiber
crossing in human brain depicted with diffusion ten-
sor MR imaging. Radiology 2000;217:897903
82. Stieltjes B, Kaufmann WE, van Zijl PC, et al. Dif-
fusion tensor imaging and axonal tracking in the
human brainstem. Neuroimage 2001;14:723735
Appendix 1appears on the next page.
D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
j
r
o
n
l
i
n
e
.
o
r
g

b
y

1
9
0
.
1
6
9
.
9
9
.
1
1
5

o
n

0
2
/
2
0
/
1
4

f
r
o
m

I
P

a
d
d
r
e
s
s

1
9
0
.
1
6
9
.
9
9
.
1
1
5
.

C
o
p
y
r
i
g
h
t

A
R
R
S
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
;

a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d

16 AJ R:178, J anuary 2002
Melhem et al.
APPENDIX 1. Equations in Diffusion Tensor MR Imaging
Fractional anisotropy:
Relative anisotropy:
Volume ratio:

1
,
2
, and
3
are the eigenvalues of the principal, medium, and minor eigenvectors, respectively.
D is the trace (
1
+
2
+
3
) of the diffusion tensor.
_
D
o
w
n
l
o
a
d
e
d

f
r
o
m

w
w
w
.
a
j
r
o
n
l
i
n
e
.
o
r
g

b
y

1
9
0
.
1
6
9
.
9
9
.
1
1
5

o
n

0
2
/
2
0
/
1
4

f
r
o
m

I
P

a
d
d
r
e
s
s

1
9
0
.
1
6
9
.
9
9
.
1
1
5
.

C
o
p
y
r
i
g
h
t

A
R
R
S
.

F
o
r

p
e
r
s
o
n
a
l

u
s
e

o
n
l
y
;

a
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d