Michael E. Moseley, Haleh S. Asgari, MS
PhD #{149} Yoram Cohen, #{149} Michael F. Wendland,
Diffusion-weighted Water Diffusion The diffusion behavior of intracranial water in the cat brain and spine was examined with the use of diffusion-weighted magnetic resonance (MR) imaging, in which the direction of the diffusion-sensitizing gradient was varied between the x, y, and z axes of the magnet. At very high diffusion-sensitizing gradient strengths, no clear evidence of anisotropic water diffusion was found in either cortical or subcortical (basal ganglia) gray matter. Signal intensities clearly dependent on onentation were observed in the contical and deep white matter of the brain and in the white matter of the spinal cord. Greater signal attenuation (faster diffusion) was observed when the relative orientation of white matter tracts to the diffusionsensitizing gradient was parallel as compared to that obtained with a perpendicular alignment. These effects were seen on both premortem and immediate postmortem images obtained in all axial, sagittal, and coronal views. Potential applications of this MR imaging technique include the stereospecific evaluation of white matter in the brain and spinal cord and in the characterization of demyelinating and dysmyelinating diseases.
PhD #{149} John Kucharczyk, PhD #{149} Jay Tsumuda,
PhD MD
MR Imaging In Cat Central
T
HE
signal
intensity
in
From the (M.E.M.,
D.N.)
and
Departments J.K., J.M.,
Pharmaceutical
proton
ty
of Radiology, H.S.A., M.F.W., Chemistry
Box J.T.,
(Y.C.,
J.M.), University of California, 500 Parnassus Aye, San Francisco, CA 94143. From the 1989 RSNA scientific assembly. Received December 1 1, 1989;
Index terms: Brain, anatomy, 10.92 #{149} Brain, MR studies, 10.1214 #{149} Brain, white matter Magnetic resonance (MR), experimental #{149} Magnetic resonance (MR), technology Radiology
1990;
176:439-445
revision
requested
January
Mintorovitch, Norman, MD
PhD
#{149}
of Anisotropic Nervous System’
magnetic resonance (MR) imaging depends on Ti and T2 relaxation processes, proton density, and molecular motions due to pulsatile flow and convective or diffusion processes. Although measurement of diffusion with the use of optimized pulsed-gradient spin-echo (StejskalTanner) MR imaging and the assessment of anisotropic restricted diffusion were performed more than 20 years ago (1,2), in vivo diffusion MR imaging, which can map microscopic motion of water protons (motional lengths on the order of microns), has been the subject of only a few recent studies (3-8). The original Stejskal-Tanner sequence and the diffusion-weighted MR imaging sequence adopted for our study are shown in Figure 1. Because the two diffusion-sensitizing gradient pulses are symmetric (same length, amplitude, and position with respect to the RF pulses) and sepamated by a 180#{176} RF pulse, all spin dephasing caused by the first diffusionsensitizing gradient pulse will be refocused by the second diffusionsensitizing gradient pulse for stationany spins. Moving spins (because of flow, perfusion, and diffusion) will not refocus and will attenuate the signal. From the original StejskalTanner pulse sequence, shown in Figure la, the observed echo intensi-
0628,
#{149} Jan
#{149} David
24,
1990;
revision received February 26; accepted April 20. Y.C. supported in part by grant ROlGM34767 from the National Institutes of Health and by the Fulbright Scholarship administered by the USA-Israel Education Foundation. Address reprint requests to M.E.M. C RSNA, 1990
S(TR,
TE,
can
Ci)
be expressed
as
follows: S(TR,
TE,
xli
Ci)
= S(a,
0, 0)
exp(-TE/T2)
-2exp[-(TR-TE/2)/T1] + exp[-TR/T1]
x
exp[-’y2#{246}2Gi2(i
-
#{244}/3)D],
where S(c#{176}, 0, 0) is the signal at a repetition time (TR) of and an echo time (TE) of 0 msec; Ti and T2 are the longitudinal and transverse relaxation times, respectively; ‘y is the gyromagnetic ratio; #{244} and Ci are the du-
90
180
[1
H---
5
Echo
0-
I
TE
90
Echo
180
Sift.
::::i omueoo
TE
b.
Figure 1. The Stejskal-Tanner pulsed-gradient spin-echo sequence for measuring diffusion (a) and the pulse sequence used to acquire the diffusion-weighted image (b). G is the amplitude of diffusion-sensitizing gradient pulses that can, in principle, be applied to any direction and in our study was varied along x, y, and z axes of a superconducting magnet. Parameters #{246} and i refer to the duration
and
separation
of
the
matched
diffu-
sion-sensitizing gradient pair, respectively. The diffusion-sensitizing gradient pair G was placed on each of the section (Slice) and frequencyand phase-encoding axes in separate experiments to assess diffusional anisotropy. Rf = radio frequency.
Abbreviations: RF = radio TE = echo
CSF frequency, time.
cerebrospinal TR repetition
fluid, time,
439
ration and the amplitude of the diffusion-sensitizing gradient pulses, respectively; is the interval between the leading edges of the diffusion-sensitizing gradient pulses; and D is the water apparent self-diffusion coefficient. The first two terms in this equation represent Ti and T2 melaxation processes, with the third representing the diffusion-dependent term. Division of the two images obtained with use of the same MR imaging parameters with and without diffusion-sensitizing gradient pulses allows determination of the apparent diffusion coefficient according to the following equation (1,2,7): ln S(TE,
Gi)/S(TE, =
-‘y22Gi2(
0) -
#{244}/3)D -bD,
where the gradient attenuation factor (3,4) b is defined as ‘y2#{244}2Gi2( t5/3). This equation suggests a major advantage of measuring diffusion with the use of MR imaging methods for in vivo applications in that the direction of the diffusion-sensitizing gradient pulses can be controlled and apparent diffusion along the respective direction can be measured. By variation of the time separation (is) between the diffusion-sensitizing gradient pulses, the mean path length can be estimated with the Emstein equation and the measured diffusion coefficient (9). One potentially important in vivo application of diffusion imaging is the detection of microscopic motion in the various intracranial fluid cornpantments. As the strength and duration of the diffusion-sensitizing gradient pulses are increased, the apparent diffusion of the more freely diffusing protons, such as those found in cemebrospmnal fluid (CSF) and pulsatile blood, is manifested as signal loss on the observed magnitude-calculated, diffusion-weighted image, leaving only slower-diffusing proton motions to contribute to image intensities. This has been of use in the observation of gray matter and white matter structures devoid of hypenintense CSF and blood signals. In another application, heavily diffusion-weighted MR imaging has been found to be very efficient in the early detection of cerebral ischemia (8). The ability to measure the diffusion tensor (the magnitude as a function of direction) allows assessment of the anisotropy of diffusion. The presence of nonrandom barriers will impart anisotnopic proton diffusion, resulting in a slower apparent diffusion coefficient for protons along a -
440
#{149} Radiology
more restricted direction compared with a direction in which the spins can translate freely. Stejskal-Tannem MR diffusion methods have been invaluable in determining the anisotropic (differing diffusional rates along one or more of the three magnet axes) diffusion behavior of small molecules in ordemed or oriented matter (9-il) and of tissues in vitro (12-14). Strong anisotropy in the motion of water molecules has also been demonstrated in bovine tendons, presumably because of the regular parallel molecular amrangement of the collagen molecule surfaces (15). Anisotropy of diffusion, although not measured, has been suggested by Thomsen et al (7) as a possible explanation for the large regional differences in in vivo apparent diffusion coefficients of human white matter. In the present study, we examined the relationship between the apparent diffusion behavior of water protons in normal gray and white matter of cat brain and spinal cord with the diffusion-gradient strength and, principally, the diffusion-gradient direction.
Figure 2. Demonstration of heavy diffusion-weighted anisotropy is shown from images of fresh sugar cane (1,800/80, 3-mm section, 80-mm field of view, two signals averaged) with a b value of 1,413 sec/mm2. Image A was acquired with diffusion gradient applied along z axis (out of plane on A). The hypointensity
with
AND
Nonfasting mongrel studied to date) weighing sedated injection
METHODS cats
(30 have been 2.5-3.5 kg were
with a 1-2 mg/kg of acepromazine,
with l%-2% isoflurane, with a Drager respirator maintain normal Pao2
intramuscular anesthetized and ventilated (Telford, Pa) to and Paco2 levels.
Special care was taken to ensure that the cats were firmly restrained and positioned in the head coil and probe. Pancuronium bromide (Pavulon; Organon, West Orange, NJ) (0.5 mL/kg) was administered for
intravenously muscle
random
paralysis motion.
all cats ing the
every to reduce The
core
applied
stalk
bundles.
was maintained at 37#{176}Cby placanimal on a thermally regulated
for
comparison
and
and
cyclohexane,
magnets, equip smaller
although
inserts for
are
it is not
lange-bore
clinical
magnets
gradient
inserts.
Diffusion-
sion-weighted echo images mm sections, field of view)
diastole singleand
and and axial
to
pulses
(half-
(#{244}) of 20 msec, a difseparation time (z) of 40 in sec/
gra-
This
ing for
and were
ces), resulting 16 minutes. In separate of the
applied
pulses at these sections were calculated to be respectively (3,4). similar for all axial,
coronal
images.
averaged
phase-encoding
with
for
steps
(128
in a total
Four each
on eight
of
X 256
128
matni-
collection
experiments,
the
time
of
direction
diffusion-sensitizing
gradi-
ent pulses was varied between the x, y, and z axes of the superconducting magnet, with the z axis taken as the B0 field direction and the x and y axes taken as the horizontal and vertical magnet direc-
end multidiffu-
(Stejskal-Tanner) spin(TRITE 1,100-1,800/80, 1-mm section gaps, 80-mm were acquired with diffu-
scale. Diffusion phanthe sugar cane in water, pure corn oil,
value takes into account the shape of the diffusion-sensitizgradient pulses. The residual b values the section-selection and frequency-
signals
weighted MR imaging was performed with a low-pass bird-cage proton imaging coil with an 8.5-cm inner diameter (19). Candiac-gated (to end systole) and nongated section coronal, sagittal,
is shown at a lower
half-sine
small-bore
difficult
in
from applied
msec, and applied diffusion-sensitizing gradients of 0-5.6 G/cm, resulting dient b values (3,4) of up to 1,413 mm2.
commercial-
relatively
orientation
gradient
durations
with
sagittal,
only
(honi-
respectively.
shielded
gradient
vascu-
x axis
is displayed
relative intensity tubes placed above image are of pure
encoding gradient and fields of view 14 and 32 sec/mm2, These values were
available
fibrous
the
fiber
(X4) torn each
water blanket within the magnet. A CS! 2.0-1 unit (GE Medical Systems, Fremont, Calif) equipped with selfshielded gradient coils (Acustar; GE Medical Systems) (±20 G/cm, 15-cm free bore) was used (16-18). These strong, selfly
along
The
fusion-gradient
of
the
sugar cane can thus be determined knowledge of the direction of the gradient. Image D (b 0 sec/mm2)
sine)
further
temperature
gradient
sion-sensitizing
45 minutes
in
zontal on image, arrow) shows markedly slower diffusion (hyperintensity) in sugar cane bundles when fiber direction is perpendicular to applied gradient direction. The subtraction image C (B minus A) mdicates the presence of large anisotropy only in
MATERIALS
observed
lar bundles indicates fast diffusion when fiber direction is parallel to the direction of applied gradient. Corresponding image B
3-
tions, respectively. diffusion-sensitizing were used, resulting
Although very strong gradient pulses in b values much
August
1990
Figure 3. Coronal diffusion-weighted spin-echo images of a cat showing the directional dependence of increasing diffusion-sensitizing gradient strengths. MR parameters are similar to those given for Figure 2 (1,800/80, four signals averaged, 3-mm section, 80-mm field of view). Images A-C were obtained at b ‘ 0, 628, and 1,413 sec/mm2 along the x axis (arrow), while images D-F were obtained similarly along the z axis. Note that these images are not windowed, showing image intensities down to noise levels. Phantoms are of corn oil, water, and cycbohexane, respectively. All images were cardiac gated. Anisotropy, which can be seen at a b value of 628 sec/mm2, is readily apparent at a b value of 1,413 sec/mm2. Because of dynamic range, images C and F are displayed at X2 intensities.
ed images of fresh sugar cane were obtained with the parameters listed above and are shown in Figure 2. With the diffusion-sensitizing gradient applied parallel to the orientation of the vascular bundles (the fibrous sieve tubes or columns) of the sugar cane (the z axis of the magnet; Fig 2a), the apparent water diffusion mate was observed to be larger than the rate observed with the gradient applied perpendicular to the z axis (x and y axes of the magnet; Fig 2b) by at least a factor of five. This is seen as water-proton hypointensity (Fig 2a) and hypenintensity (Fig 2b), respectively, from the fibrous bundles. Subtraction (Fig 2c) of the direction-dependent diffusion-weighted images illustrates the presence of anisotropic diffusion, emphasizing the contnibution of directional information to the image contrast. Regions of isotropic (no anisotropy) diffusion subtract out (such as for the 5-mm outer diameter tubes of water, corn oil, or cycbohexane positioned above the sugar cane and for the sugar cane parenchyma or pith; Fig 2c). Figure 3 shows the effect of both strength (b 0, 628 and 1,413 sec/ mm2) and direction (x and z axes) of the diffusion-sensitizing gradient pulses on image contrast in cardiacgated coronal images from a representative cat. At b values of 1,413 sec/ mm2 (5.6 C/cm), fast diffusing protons in CSF and ocular fluid, which are normally hypenintense on T2weighted images, were not visible on the diffusion-weighted images. At these high b values, gray and white matter tissue contrast was found to be very dependent on the direction of the diffusion-sensitizing gradient pulses because of the directional dependence of white matter and the lack of corresponding dependence of gray matter regions. From regional signal intensities measured at b 0, 628, and 1,413 sec/mm2, apparent diffusion coefficients for gray and white matter regions have been estimated from a number of the cats studied. In cortical and subcontical gray matter and in basal ganglia, the mean apparent coefficient were observed to be 0.67 X i0cm2/sec ± 0.10 X i0, 0.75 X i0cm2/sec ± 0.10 X i0, 0.69 X i0cm2/sec ± 0.09 X i0 averaged for the x, y, and z directions, respectively. Within a given region of cortical white matter (corona radiata), however, the apparent diffusion coefficients varied considerably and were calculated to be as low as 0.45 X i0 cm2/sec ± 0.06 X 1O (in the x direction) and as high as 1.0 X iO ‘
higher
than
those
quency-encoding gradient pulses quence, it can terms pulses
changing gradients The in
this
in be
between and the
an important age acquisitions
from
the
the
and section-selection the MR imaging argued that the
(1). were
frequency(swapping
To address repeated and axes).
secross
Ti, T2, and heavy
ten
and
diffusion
chloride.
shown
images
with
weighting.
in a diffusion-weighted
atively region of the
faster apparent diffusion within or voxel will result in attenuation voxel signal and, consequently,
image,
nela
which
the
signal
(on
intensity
the diffusion effect, requires images of varying b values sualization of anisotnopic on diffusion-weighted
up
sec/mm2 tion-selection gradients).
to
1,413
sec/mm2
inin
solely
two (3,4,7).
or more The vi-
and
to
is easier which are
14-30
(the
residual b values from secand frequency-encoding These images with low b val-
ues are images
referred where
Volume
176
b
to in our study 0 sec/mm2).
#{149} Number
2
as “b
direction of tubes
along of wa-
solvents
cycbo-
isotropic
estimated
b
from
and methyhene eddy
current
simple
one-
line shapes of water aca variable delay following gradient pulses 20 msec).
of 5.6
x-,
G/cm
RESULTS
devoid of CSF and blood signals (which are very hyperintense on T2-weighted and pure diffusion images). Except where noted, all images were acquired without cardiac gating. In our study, images were obtained at b values
gradient in a series
gradient
z-axes duration,
were
the gradient in eight incre-
Residual spectral after
25#{176}C, the measured diffusion coefficients of methylene chloride, acetone, cyclohexane, benzene, and water were all within 5% of values in the published literature (9 and reference cited therein), regardless of the gradient direction chosen. No changes in the position, intensity, or line width of the water line shapes could be detected at delays longer than 3 msec after a gradient pulse of 5.6 C/cm along any axis. From this assessment of the gradient recovery time, the minimum delay after the diffusion-sensitizing gradient pulses in the spin-echo diffusion pulse sequence was set at 6 msec. To test the utility of MR imaging in the assessment of anisotropic waten diffusion, axial diffusion-weightAt
is due
detail images,
images
acetone,
pulse quined
of
of the same parameters the cat model. Images
by varying sec/mm2
pure,
reliability
an
hypointensity from that voxel. Conversely, slower apparent diffusion will produce a smaller signal attenuation, resulting in regional relative hypenintensity in those fast-diffusion voxels. The image,
use with
benzene, were
observed
of a “pure-diffusion” incoherent motion)
the
effects
y-, and (pulse
and
measurements,
with used
ments) and the x, y, and z axes
this, imwith inter-
images
accuracy
acquired (0-1,413
hexane,
magnitude
the
diffusion
were factor
play
Thus,
calculation travoxel
check
obtained as those
phase-encoding
diffusion-weighted are
To
fre-
the diffusion-sensitizing other gradients may robe
study
inherent
0”
Radiology
. 441
5.
4.
Figures
4, 5. (4) Coronal nongated, diffusion-weighted, spin-echo images of a cat at the level of the inferior colliculi. MR parameters are similar to those of Figure 2 (1,800/80, four signals averaged, 3-mm section, 80-mm field of view). Gradient strengths are the same as those used for Figure 2 (b . 0 sec/mm2 in A and 1,413 sec/mm2 in B-D). The direction of the applied diffusion-sensitizing gradient was varied among the x axis (left-to-right direction in B, white arrow), y axis (vertical direction in C, arrow), and z axis (out-of-plane direction in D). When white matter tracts are oriented parallel to the direction of applied diffusion gradient, fast directional (relatively unhindered or unrestricted) diffusion of water protons is indicated by regions of hypointensity. Thus, in B, transhemispheric fibers, such as the corpus callosum (open arrow) and posterior commissure, have low signal intensity. Water diffusion perpendicular to applied gradient direction is slower (more hindered by the structure and orientation of white-matter bundles), which accounts for the high signal intensity of the corona radiata in B (solid black arrow), and corpus callosum and posterior commissure in D. No clear relationship between white matter signal intensity and gradient direction can be discerned in C, because the plane of applied gradient is oblique to the orientation of most white matter tracts at this level of the neuraxis. (5) Subtraction of the axial y axis (A) and x axis (B) direction-dependent diffusion-weighted images produces images that show little signal from gray matter regions (C [A minus B], D [B minus A]). Subtraction images clearly demonstrate the presence and orientation of white matter. Absolute image intensities of subtraction images, produced from diffusion-weighted images, do contain Ti and T2 information but are useful in visualization of directional effects.
cm2/sec ± 0.1 X i0(z direction). The corpus calbosum was also observed to be an ordered structure from the pronounced dependence of the apparent diffusion coefficient on the z and x axes (0.35 X iO cm2/sec ± 0.07 X i0and 1.3 X 10 cm2/sec ± 0.12 X i0-, respectively). The profound orientational dependence of white matter signal intensity can be observed on coronal diffusion-weighted images from a different cat acquired along all three magnet axes together with the comesponding spin-echo b = 0 image (Fig 4). This dependence of white matter
ages
signal
isotropy
intensity
can
also
be
observed
the corresponding nongated, diffusion-weighted images at one axial level of the cat shown in Figure 4 and is even more pronounced on the respective subtraction images (Fig 5). To assess the contribution of possible artifacts, additional experiments (Fig 6) were performed. To assess the effect of the cardiac cycle on image intensity, gated (to end diastole [Fig 6a] and to end systole [Fig 6b]), nongated (Fig 6c), and postmortem (Fig 6d) imin
442
#{149} Radiology
1,413
dress
along the x direction with all other parameters (eg, TR) remaining constant. No apparent differences between cardiac-gated images and nongated images were observed. The
tion
were
obtained
at b values
of
sec/mm2
postmortem
diffusion-weighted
age displayed isotropy
the
observed
image;
the
only
matter
hypenintensity
on
pattern the
difference
of an-
was due
which cardiac-gated, weighted images quency-encoding
were sitizing (magnitude
in
to
Figure
gray
7, in
direction).
gradient
z axes
fre-
interchanged
These
sagittal junction)
along
in Figure images
cord
spinal
of
(near
the
shown
in
8. post-
cerviFigures
10 are in agreement with the observations made above. Fast-diffusing water protons in CSF and blood appear hypointense. Furthermore, gray matter in the spinal cord shows no
clear
directional
white
whereas be
im-
ages, in which the cross terms are different, produced similar contrast charactemistics, implying that the anisotropic contrast seen on the image is due primarily to the diffusion-sensitizing gradient pulses. To further ad-
the
dependence,
matter
predicted
entation
and
pulses
are shown
and
mortem
can
obtained with the diffusion-sengradient pulses held constant and
x and
9 and
postmor-
diffusionwith phaseaxes
sion-sensitizing
cothoracic
tern pathophysiobogic processes (8,20). The effect of the normal gradient pulses and the cross term between them and the diffusion-sensitizing gradient pulses on the observed anis shown
postmortem
Axial
premortem
contribution of moto cardiac-gated, diffuimages, pre- and images with the diffu-
relative
sion-weighted
im-
same
the
artifacts
in
applied
by
This
anisotropic
ubamby
well
visualized
to
effect when
diffusion-weighted
rectional
(D in Fig
subtracted
intensity
anatomic
the diffusion-sensitizing
relation
dient.
are
signal its
on-
direction
of gra-
is partic-
diimages
two
9).
DISCUSSION On
the
presented
(1,2),
basis
of principles originally Stejskal and Tanner orthogonal direction-depenby
August
1990
pattern of diffusional the axes.
6.
dent diffusion gradients have been used to probe anisotropic influences on molecular diffusion in a variety of model systems (9-11) and tissues in vitro (12,14). To accurately study ondening in matter, the residual “background” gradients (magnet homogeneity, section-selection gradients, frequency-encoding gradients), must be small or negligible in effective strength (the b value) compared with the strength of the applied diffusionsensitizing gradient pair in the pulsed-gradient experiment. In addition, the suppression of eddy currents must be sufficiently thorough to avoid image distortions and gradient mismatch. One test of this latter prerequisite is that isotropic diffusion must, by definition, subtract out from diffusion-weighted images. This has been clearly demonstrated here with the phantoms studied and images obtained. Figure 2, in particulam, is a good example of gradient behavion and of the large amount of ondening (the degree of water proton diffusional anisotropy) observed in the oriented fibers in most fibrous fruits and vegetables. Diffusional anisotropy of small molecules, such as water protons, can be induced by any oriented lattice that presents a nonmandom array of barriers that hinders diffusion or motion in one direction relative to another. Anisotropy, however, can also be caused by “restricted” or cornpletely hindered diffusion, such as that, for example, imposed by cellVolume
176
#{149} Number
2
membrane barriers in elongated cells. Because the diffusion coefficient of free, unhindered water at 25#{176}C-37#{176}C is about 2-4 X i0 cm2/ sec (3,4,7,9) and the spin diffusion time is approximately #{244}/3 or 30 msec, the root mean square path length of proton diffusion along any one axis can be calculated to be 12-16 trn. In biologic sytems, restricted megional (intracellular) proton diffusion will occur if the cellular diameten is less than 12-16 jzm. Restriction may cause the apparent diffusion coefficient of intracellular protons to be smaller than that for protons in the extracellular or interstitial compartments; consequently, the resulting diffusion-weighted image intensity will be less attenuated compared with that in an area in which free diffusion prevails. A quantitative assessrnent of the effect of restriction can be derived from MR diffusion measurements in which the spin diffusion time (i 6/3) is systematically varied (2,12,14,21). The same amgument applies to restricted or hindemed diffusion of water protons through ordered barriers, in which the mean free path length will be longer in one direction than another. One consistent observation from the in vivo diffusion-weighted irnages is that the apparent diffusion rates of water protons in the white matter of the brain and spinal cord are fundamentally different from -
-
anisotropy
is retainedon
swapping
of
that of the gray matter. Cray matter diffusion does not show diffusional directional dependence, whereas white matter apparent diffusion is very anisotropic. It can be argued that factors other than orientationdependent diffusional anisotropy in white matter are responsible for the observed contrast that is charactenistic when the direction of diffusionsensitizing gradient pulse is varied. Effects of perfusion or physiologic pulsations or vibrations can be ruled out as the major cause of this apparent anisotropic diffusion in white matter by the observations that the degrees of anisotropy are similar for gated, nongated, pmemortern, and irnmediate postmortem images (Figs 6, 8-10). Diffusion-sensitizing gradient mismatch and other gradient effects would have resulted in observable artifacts in the subtraction images as well as incomplete subtraction of the phantom signal, which diffuse isotropically. The similarity in the images obtained with switched axes demonstrates the relative negligible effect of the cross terms between the phaseand frequency-encoding gradient pulses and the diffusion-sensitizing gradient pulses (Fig 7). The in vivo directional diffusionweighted images (Figs 3-10) indicate that water proton diffusion anisotropy is best observed in the large-diarneter, fast-conducting motor and somatosensory nerve fibers that form Radiology
#{149} 443
9.
8.
Figures 8, 9. (8) Premortem (A and C) and postmortem (B and D) diffusion-weighted images at a b value of 1,413 sec/mm2. The cardiac-gated premortem pattern of anisotropy (with the direction of the diffusion-sensitizing gradient along the x axis in A [arrow] and along the z axis in C) is retained post mortem. No windowing of the images was performed. The postmortem hypenintensity is thought to be due to pathophysiologic changes (8,20). (9) Axial diffusion-weighted spin-echo images of immediate postmortem cat spine near the cervicothoracic junction (1,800/80, four signals averaged, 3-mm section, 60-mm field of view; original magnification, X2) for b 0 sec/mm2 (A) and 1,413 sec/mm2 (B, C). The direction of the applied diffusion gradient was varied between the z axis (along the long axis of spinal cord in B) and the y axis (vertical direction in C, arrow). Fast-diffusing protons of CSF in the subarachnoid space (black arrow in A), which are hypermntense with T2 weighting, are hypointense on all diffusion-weighted images. The H-shaped central gray area has a slightly lower signal intensity than does CSF on the T2-weighted image (A). Same area has a variable appearance on the diffusion-weighted images because the myelinated nerve fibers, which enter spinal cord via dorsal roots and exit via ventral roots, also run through the central gray area. Heavily myelinated large-diameter axons, including the proprioceptive fibers in the dorsal columns, appear hypointense when the diffusion-gradient direction is parallel to the long axis of the cord (in the z direction in B). White matter appears bright when the gradient direction is in the transverse plane (C). Subtraction of the two diffusion-weighted images (D) shows the effect of diffusional anisotropy, reflecting the predominantly cephalocaudal orientation of white matter in the spinal cord.
the posterior limb of the internal capsule. In addition to the heavily myelinated propnioceptive axons, optic and olfactory nerves, corpus calbosum, and anterior and posterior cornmissures appear to be associated with significant anisotropic effects. This ordering or hindering of water diffusion or motion probably occurs along the axis of the individual neurofibnils as well as along the axis of the myelm sheath itself. Thus, it seems possible to determine a priori the onientation
of
the
white
matter
tracts
from
knowledge of the direction dependence of the apparent diffusion. Perhaps the clearest correlation between anisotropy observed on MR images and the underlying anatomy can be seen in the altered signal intensities in spinal cord white matter as the dinection of the diffusion gradient was systematically varied. Apparent diffusion
was
faster
(yielding
signal
hy-
pointensity) along the axis of the spinal cord when a z-direction gradient was employed, whereas a diffusion gradient applied perpendicular to the axis of the cord (x-direction of 444
#{149} Radiology
gradient)
detected
slowed
diffusion
(high signal intensity) in white matter. Heavily diffusion-weighted MR imaging is a new and unique tool for the noninvasive assessment of white matter orientation. Measurements of the directional-apparent diffusion coefficients and the translational path lengths may be used to derive information concerning barmier sizes and directional diffusion, from which the barrier or lattice dimensions can be quantified. By oblique application of the diffusion gradients, off-axis onentations can also be studied. Clinical applications of gradient direction-selective diffusion MR imaging
are
immediately
larger
be achieved
gradient
clinically
of
MR
imaging
technique
the
strengths
through
can
the
local
orientation
of
through ton
coils.
Utilization
apparent
matter
improve assessment
disorders,
infamcts,
of water prohas the potenour underof demy-
diffusion
elination
of this to determine
white
measurement
tial to greatly standing and
white
matter
involving white and neonatal brain and development. U
neoplasms
matter
tracts,
spinal
cord
References 1.
Stejskal
EO,
Tanner
measurements: ence 2.
obvious.
While the threshold and sensitivity of the anisotropic effect is inherently rebated to the available gradient strength of the diffusion gradient pair in the sequence, irrespective of whether a Stejskal-Tanner spin-echo (1,2), steady-state free precession (5,22,23), or echoplanar (6) approach is used,
use
3.
Spin field
tropic,
restricted Phys
diffusion 1965;
pres-
gradient.
and
flow.
43:3597-3603.
D, Breton
E, Lallemand
ier P. Cabanis E, Laval-Jeantet aging of intravoxel incoherent application neurologic
the
1965; 42:288-292. Use of spin echo in pulsed gradient to study aniso-
Chem Bihan
diffusion in
of a time-dependent
Chem Phys Stejskal EQ. magnetic-field
Le
JE.
spin-echoes
to diffusion disorders.
D, GrenM. MR motions:
im-
and perfusion Radiology 1986;
in
161:401-407.
4.
Le Bihan D, Breton E, Lallemand M-I. Vignaud J. Laval-Jeantet
M.
tion voxel
in intraimaging.
diology
of diffusion incoherent 1988;
and perfusion motion MR
D,
Aubin
SeparaRa-
168:497-505.
August
1990
13.
14.
Hansen mobility
JR. in
Pulsed muscle
NMR study of water and brain tissues. Bio-
chem Biophys Acta 1971; 230:482-486. Cleveland GG, Chang DC, Hazelwood CF. Rorschach HE. Nuclear magnetic resonance meaurements of skeletal muscle anisotropy
of the
intracellular
diffusion
water.
coefficient
Biophys
of the
J 1976;
16:
1043-1053.
15.
Fullerton
GD,
entation and
Cameron
IL, Ord
of tendons its
effect
ology
in
on
1985;
T2
the
VA.
On-
magnetic
relaxation
field
times.
Radi-
155:433-435.
16.
Roemen PB, Edelstein WA, Hickey JS. Self-shielded gradient coils (abstn). In: Book of abstracts: Society of Magnetic Resonance in Medicine 1986. Berkeley, Calif: Society of Magnetic Resonance in Medicine, 1986; 1067-1068.
17.
Mansfield ic screening
P. Chapman of coils
B. Active static and
for
magnettime-de-
pendent magnetic field generation NMR imaging. J Phys E Sci Instrum 18.
19:540-545. Turner R, Bowley
of switched
a. Figure
Phys
b.
19.
10. Sagittal diffusion-weighted images (94-mm field of view) show white matter anisotropy in spinal cord and cerebellum. (a) Image was obtained with the diffusion gradient applied along the z axis (parallel to the spinal cord axis, arrow). (b) Image was obtained with the gradient applied along the y axis (horizontal direction, arrow). Differences in signal intensity between these images and the images in Figure 7 arise only from white matter contributions
to
signal
intensity
and
mirror
the
observations
in
Figure
E Sci
Hayes An
9.
Merboldt MH,
K-D, Hanicke
W,
“diffusion” suits
in
using
Magn 6.
Bruhn
R, Le
Echo-planar aging
M.
MRI
human
brain:
new
CE-FAST
Med
1989;
Bihan
D,
stracts:
Tesla
Society
Medicine Magnetic
and
of 9.
re-
sequence.
9:423-429. J, Pekar
Delannoy
diffusion
at 2.0
J, Gyngell
Frahm
Deimling
the
a modified
Reson
Turner
H,
and (abstr).
perfusion In:
of Magnetic
J.
of Ab-
Resonance Society 1989;
C, Henriksen
P.
of 11.
Thomsen
vivo measurement in the human
nance
Ring
In-
of water self-diffusion by magnetic reso-
brain
imaging.
0,
Acta
Radiol
1987;
spin-echo
sion.
Prog
28:353-
ischemia
Y, Mintorovitch J, et of regional cerebral
ME, Cohen detection in cats:
comparison
of diffusion-
176
#{149} Number
2
studies NMR
21.
14:330-346.
transform
of molecular 1987;
12.
whole
ME,
of acute
stroke:
weighted
and
imaging.
JE.
Transient
im-
63:
with
T2-
susceptibility-en-
AJNR
1990;
diffusion
by
to NMR field
1985;
correlation magnetic
MR
NMR
J. Mintorovitch MR imaging
Kucharczyk
partitioned
JF, et al. radio-
body
Reson
Diffusion-weighted
pulsed
19:1-
for
I, et al.
plication
diffu-
Schenk
homogeneous
at 1.5 T. J Magn
429. Tanner
19:876-879.
WA,
622-626. Moseley
tern
pulsed-gradi-
Spectroscopy
Lindbloms
G.
for
of dynamics
studies
Scand Tanner
Cooper
NMR
diffusion:
of membrane
1981; JE.
1976; RL,
sion
22.
a method
and
mesophase
lipids.
Acta
Le Bihan tion cession.
Chem 23.
B35:61-62. Measurements
in a sys-
permeable
barriers:
measurements
ap-
with
J Chem
gradient.
11:423-
Phys
a
1978;
of self-diffu-
systems by magneticspin echo methods. In: CG, eds. Magnetic resoscience. Chemical
D.
imaging
using
Magn
incoherent steady
Reson
Med
mo-
state
free
1988;
7:346.
pre-
Le Bihan D, Turner R, Macfall JR. Effects of intravoxel incoherent motions (IVIM) in steady-state
Res-
Intravoxel
aging: imaging.
free
application Magn
precession
(SSFP)
to molecular Reson
Med
im-
diffusion 1989;
10:324.
So-
16-29. Chang
CJ, Ancker-Johnson 1974;
Volume
1990;
spectroscopy.
screening
gradients.
69:1748-1754.
ciety,
Moseley al. Early
and
nance in colloid and interface Washington, DC: American
361.
8.
ent
Med
Fourier
sion in colloidal field-gradient, ing HA, Wade
139.
7.
Reson P.
structure
in
1989. Berkeley, Calif: Resonance in Medicine,
Magn Stilbs
MRI
45.
10.
im-
Book
T2-weighted
1986;
highly coil
hanced 5.
Instrum
efficient
aging
Passive
field
CE, Edelstein
frequency
20.
RM.
magnetic
in 1986;
in biophysical
DB.
Young
B. systems.
AC,
Restricted
Martin
diffu-
Biophys
14:161-177.
Radiology
#{149} 445
PhD #{149} Yoram Cohen, #{149} Michael F. Wendland,
Diffusion-weighted Water Diffusion The diffusion behavior of intracranial water in the cat brain and spine was examined with the use of diffusion-weighted magnetic resonance (MR) imaging, in which the direction of the diffusion-sensitizing gradient was varied between the x, y, and z axes of the magnet. At very high diffusion-sensitizing gradient strengths, no clear evidence of anisotropic water diffusion was found in either cortical or subcortical (basal ganglia) gray matter. Signal intensities clearly dependent on onentation were observed in the contical and deep white matter of the brain and in the white matter of the spinal cord. Greater signal attenuation (faster diffusion) was observed when the relative orientation of white matter tracts to the diffusionsensitizing gradient was parallel as compared to that obtained with a perpendicular alignment. These effects were seen on both premortem and immediate postmortem images obtained in all axial, sagittal, and coronal views. Potential applications of this MR imaging technique include the stereospecific evaluation of white matter in the brain and spinal cord and in the characterization of demyelinating and dysmyelinating diseases.
PhD #{149} John Kucharczyk, PhD #{149} Jay Tsumuda,
PhD MD
MR Imaging In Cat Central
T
HE
signal
intensity
in
From the (M.E.M.,
D.N.)
and
Departments J.K., J.M.,
Pharmaceutical
proton
ty
of Radiology, H.S.A., M.F.W., Chemistry
Box J.T.,
(Y.C.,
J.M.), University of California, 500 Parnassus Aye, San Francisco, CA 94143. From the 1989 RSNA scientific assembly. Received December 1 1, 1989;
Index terms: Brain, anatomy, 10.92 #{149} Brain, MR studies, 10.1214 #{149} Brain, white matter Magnetic resonance (MR), experimental #{149} Magnetic resonance (MR), technology Radiology
1990;
176:439-445
revision
requested
January
Mintorovitch, Norman, MD
PhD
#{149}
of Anisotropic Nervous System’
magnetic resonance (MR) imaging depends on Ti and T2 relaxation processes, proton density, and molecular motions due to pulsatile flow and convective or diffusion processes. Although measurement of diffusion with the use of optimized pulsed-gradient spin-echo (StejskalTanner) MR imaging and the assessment of anisotropic restricted diffusion were performed more than 20 years ago (1,2), in vivo diffusion MR imaging, which can map microscopic motion of water protons (motional lengths on the order of microns), has been the subject of only a few recent studies (3-8). The original Stejskal-Tanner sequence and the diffusion-weighted MR imaging sequence adopted for our study are shown in Figure 1. Because the two diffusion-sensitizing gradient pulses are symmetric (same length, amplitude, and position with respect to the RF pulses) and sepamated by a 180#{176} RF pulse, all spin dephasing caused by the first diffusionsensitizing gradient pulse will be refocused by the second diffusionsensitizing gradient pulse for stationany spins. Moving spins (because of flow, perfusion, and diffusion) will not refocus and will attenuate the signal. From the original StejskalTanner pulse sequence, shown in Figure la, the observed echo intensi-
0628,
#{149} Jan
#{149} David
24,
1990;
revision received February 26; accepted April 20. Y.C. supported in part by grant ROlGM34767 from the National Institutes of Health and by the Fulbright Scholarship administered by the USA-Israel Education Foundation. Address reprint requests to M.E.M. C RSNA, 1990
S(TR,
TE,
can
Ci)
be expressed
as
follows: S(TR,
TE,
xli
Ci)
= S(a,
0, 0)
exp(-TE/T2)
-2exp[-(TR-TE/2)/T1] + exp[-TR/T1]
x
exp[-’y2#{246}2Gi2(i
-
#{244}/3)D],
where S(c#{176}, 0, 0) is the signal at a repetition time (TR) of and an echo time (TE) of 0 msec; Ti and T2 are the longitudinal and transverse relaxation times, respectively; ‘y is the gyromagnetic ratio; #{244} and Ci are the du-
90
180
[1
H---
5
Echo
0-
I
TE
90
Echo
180
Sift.
::::i omueoo
TE
b.
Figure 1. The Stejskal-Tanner pulsed-gradient spin-echo sequence for measuring diffusion (a) and the pulse sequence used to acquire the diffusion-weighted image (b). G is the amplitude of diffusion-sensitizing gradient pulses that can, in principle, be applied to any direction and in our study was varied along x, y, and z axes of a superconducting magnet. Parameters #{246} and i refer to the duration
and
separation
of
the
matched
diffu-
sion-sensitizing gradient pair, respectively. The diffusion-sensitizing gradient pair G was placed on each of the section (Slice) and frequencyand phase-encoding axes in separate experiments to assess diffusional anisotropy. Rf = radio frequency.
Abbreviations: RF = radio TE = echo
CSF frequency, time.
cerebrospinal TR repetition
fluid, time,
439
ration and the amplitude of the diffusion-sensitizing gradient pulses, respectively; is the interval between the leading edges of the diffusion-sensitizing gradient pulses; and D is the water apparent self-diffusion coefficient. The first two terms in this equation represent Ti and T2 melaxation processes, with the third representing the diffusion-dependent term. Division of the two images obtained with use of the same MR imaging parameters with and without diffusion-sensitizing gradient pulses allows determination of the apparent diffusion coefficient according to the following equation (1,2,7): ln S(TE,
Gi)/S(TE, =
-‘y22Gi2(
0) -
#{244}/3)D -bD,
where the gradient attenuation factor (3,4) b is defined as ‘y2#{244}2Gi2( t5/3). This equation suggests a major advantage of measuring diffusion with the use of MR imaging methods for in vivo applications in that the direction of the diffusion-sensitizing gradient pulses can be controlled and apparent diffusion along the respective direction can be measured. By variation of the time separation (is) between the diffusion-sensitizing gradient pulses, the mean path length can be estimated with the Emstein equation and the measured diffusion coefficient (9). One potentially important in vivo application of diffusion imaging is the detection of microscopic motion in the various intracranial fluid cornpantments. As the strength and duration of the diffusion-sensitizing gradient pulses are increased, the apparent diffusion of the more freely diffusing protons, such as those found in cemebrospmnal fluid (CSF) and pulsatile blood, is manifested as signal loss on the observed magnitude-calculated, diffusion-weighted image, leaving only slower-diffusing proton motions to contribute to image intensities. This has been of use in the observation of gray matter and white matter structures devoid of hypenintense CSF and blood signals. In another application, heavily diffusion-weighted MR imaging has been found to be very efficient in the early detection of cerebral ischemia (8). The ability to measure the diffusion tensor (the magnitude as a function of direction) allows assessment of the anisotropy of diffusion. The presence of nonrandom barriers will impart anisotnopic proton diffusion, resulting in a slower apparent diffusion coefficient for protons along a -
440
#{149} Radiology
more restricted direction compared with a direction in which the spins can translate freely. Stejskal-Tannem MR diffusion methods have been invaluable in determining the anisotropic (differing diffusional rates along one or more of the three magnet axes) diffusion behavior of small molecules in ordemed or oriented matter (9-il) and of tissues in vitro (12-14). Strong anisotropy in the motion of water molecules has also been demonstrated in bovine tendons, presumably because of the regular parallel molecular amrangement of the collagen molecule surfaces (15). Anisotropy of diffusion, although not measured, has been suggested by Thomsen et al (7) as a possible explanation for the large regional differences in in vivo apparent diffusion coefficients of human white matter. In the present study, we examined the relationship between the apparent diffusion behavior of water protons in normal gray and white matter of cat brain and spinal cord with the diffusion-gradient strength and, principally, the diffusion-gradient direction.
Figure 2. Demonstration of heavy diffusion-weighted anisotropy is shown from images of fresh sugar cane (1,800/80, 3-mm section, 80-mm field of view, two signals averaged) with a b value of 1,413 sec/mm2. Image A was acquired with diffusion gradient applied along z axis (out of plane on A). The hypointensity
with
AND
Nonfasting mongrel studied to date) weighing sedated injection
METHODS cats
(30 have been 2.5-3.5 kg were
with a 1-2 mg/kg of acepromazine,
with l%-2% isoflurane, with a Drager respirator maintain normal Pao2
intramuscular anesthetized and ventilated (Telford, Pa) to and Paco2 levels.
Special care was taken to ensure that the cats were firmly restrained and positioned in the head coil and probe. Pancuronium bromide (Pavulon; Organon, West Orange, NJ) (0.5 mL/kg) was administered for
intravenously muscle
random
paralysis motion.
all cats ing the
every to reduce The
core
applied
stalk
bundles.
was maintained at 37#{176}Cby placanimal on a thermally regulated
for
comparison
and
and
cyclohexane,
magnets, equip smaller
although
inserts for
are
it is not
lange-bore
clinical
magnets
gradient
inserts.
Diffusion-
sion-weighted echo images mm sections, field of view)
diastole singleand
and and axial
to
pulses
(half-
(#{244}) of 20 msec, a difseparation time (z) of 40 in sec/
gra-
This
ing for
and were
ces), resulting 16 minutes. In separate of the
applied
pulses at these sections were calculated to be respectively (3,4). similar for all axial,
coronal
images.
averaged
phase-encoding
with
for
steps
(128
in a total
Four each
on eight
of
X 256
128
matni-
collection
experiments,
the
time
of
direction
diffusion-sensitizing
gradi-
ent pulses was varied between the x, y, and z axes of the superconducting magnet, with the z axis taken as the B0 field direction and the x and y axes taken as the horizontal and vertical magnet direc-
end multidiffu-
(Stejskal-Tanner) spin(TRITE 1,100-1,800/80, 1-mm section gaps, 80-mm were acquired with diffu-
scale. Diffusion phanthe sugar cane in water, pure corn oil,
value takes into account the shape of the diffusion-sensitizgradient pulses. The residual b values the section-selection and frequency-
signals
weighted MR imaging was performed with a low-pass bird-cage proton imaging coil with an 8.5-cm inner diameter (19). Candiac-gated (to end systole) and nongated section coronal, sagittal,
is shown at a lower
half-sine
small-bore
difficult
in
from applied
msec, and applied diffusion-sensitizing gradients of 0-5.6 G/cm, resulting dient b values (3,4) of up to 1,413 mm2.
commercial-
relatively
orientation
gradient
durations
with
sagittal,
only
(honi-
respectively.
shielded
gradient
vascu-
x axis
is displayed
relative intensity tubes placed above image are of pure
encoding gradient and fields of view 14 and 32 sec/mm2, These values were
available
fibrous
the
fiber
(X4) torn each
water blanket within the magnet. A CS! 2.0-1 unit (GE Medical Systems, Fremont, Calif) equipped with selfshielded gradient coils (Acustar; GE Medical Systems) (±20 G/cm, 15-cm free bore) was used (16-18). These strong, selfly
along
The
fusion-gradient
of
the
sugar cane can thus be determined knowledge of the direction of the gradient. Image D (b 0 sec/mm2)
sine)
further
temperature
gradient
sion-sensitizing
45 minutes
in
zontal on image, arrow) shows markedly slower diffusion (hyperintensity) in sugar cane bundles when fiber direction is perpendicular to applied gradient direction. The subtraction image C (B minus A) mdicates the presence of large anisotropy only in
MATERIALS
observed
lar bundles indicates fast diffusion when fiber direction is parallel to the direction of applied gradient. Corresponding image B
3-
tions, respectively. diffusion-sensitizing were used, resulting
Although very strong gradient pulses in b values much
August
1990
Figure 3. Coronal diffusion-weighted spin-echo images of a cat showing the directional dependence of increasing diffusion-sensitizing gradient strengths. MR parameters are similar to those given for Figure 2 (1,800/80, four signals averaged, 3-mm section, 80-mm field of view). Images A-C were obtained at b ‘ 0, 628, and 1,413 sec/mm2 along the x axis (arrow), while images D-F were obtained similarly along the z axis. Note that these images are not windowed, showing image intensities down to noise levels. Phantoms are of corn oil, water, and cycbohexane, respectively. All images were cardiac gated. Anisotropy, which can be seen at a b value of 628 sec/mm2, is readily apparent at a b value of 1,413 sec/mm2. Because of dynamic range, images C and F are displayed at X2 intensities.
ed images of fresh sugar cane were obtained with the parameters listed above and are shown in Figure 2. With the diffusion-sensitizing gradient applied parallel to the orientation of the vascular bundles (the fibrous sieve tubes or columns) of the sugar cane (the z axis of the magnet; Fig 2a), the apparent water diffusion mate was observed to be larger than the rate observed with the gradient applied perpendicular to the z axis (x and y axes of the magnet; Fig 2b) by at least a factor of five. This is seen as water-proton hypointensity (Fig 2a) and hypenintensity (Fig 2b), respectively, from the fibrous bundles. Subtraction (Fig 2c) of the direction-dependent diffusion-weighted images illustrates the presence of anisotropic diffusion, emphasizing the contnibution of directional information to the image contrast. Regions of isotropic (no anisotropy) diffusion subtract out (such as for the 5-mm outer diameter tubes of water, corn oil, or cycbohexane positioned above the sugar cane and for the sugar cane parenchyma or pith; Fig 2c). Figure 3 shows the effect of both strength (b 0, 628 and 1,413 sec/ mm2) and direction (x and z axes) of the diffusion-sensitizing gradient pulses on image contrast in cardiacgated coronal images from a representative cat. At b values of 1,413 sec/ mm2 (5.6 C/cm), fast diffusing protons in CSF and ocular fluid, which are normally hypenintense on T2weighted images, were not visible on the diffusion-weighted images. At these high b values, gray and white matter tissue contrast was found to be very dependent on the direction of the diffusion-sensitizing gradient pulses because of the directional dependence of white matter and the lack of corresponding dependence of gray matter regions. From regional signal intensities measured at b 0, 628, and 1,413 sec/mm2, apparent diffusion coefficients for gray and white matter regions have been estimated from a number of the cats studied. In cortical and subcontical gray matter and in basal ganglia, the mean apparent coefficient were observed to be 0.67 X i0cm2/sec ± 0.10 X i0, 0.75 X i0cm2/sec ± 0.10 X i0, 0.69 X i0cm2/sec ± 0.09 X i0 averaged for the x, y, and z directions, respectively. Within a given region of cortical white matter (corona radiata), however, the apparent diffusion coefficients varied considerably and were calculated to be as low as 0.45 X i0 cm2/sec ± 0.06 X 1O (in the x direction) and as high as 1.0 X iO ‘
higher
than
those
quency-encoding gradient pulses quence, it can terms pulses
changing gradients The in
this
in be
between and the
an important age acquisitions
from
the
the
and section-selection the MR imaging argued that the
(1). were
frequency(swapping
To address repeated and axes).
secross
Ti, T2, and heavy
ten
and
diffusion
chloride.
shown
images
with
weighting.
in a diffusion-weighted
atively region of the
faster apparent diffusion within or voxel will result in attenuation voxel signal and, consequently,
image,
nela
which
the
signal
(on
intensity
the diffusion effect, requires images of varying b values sualization of anisotnopic on diffusion-weighted
up
sec/mm2 tion-selection gradients).
to
1,413
sec/mm2
inin
solely
two (3,4,7).
or more The vi-
and
to
is easier which are
14-30
(the
residual b values from secand frequency-encoding These images with low b val-
ues are images
referred where
Volume
176
b
to in our study 0 sec/mm2).
#{149} Number
2
as “b
direction of tubes
along of wa-
solvents
cycbo-
isotropic
estimated
b
from
and methyhene eddy
current
simple
one-
line shapes of water aca variable delay following gradient pulses 20 msec).
of 5.6
x-,
G/cm
RESULTS
devoid of CSF and blood signals (which are very hyperintense on T2-weighted and pure diffusion images). Except where noted, all images were acquired without cardiac gating. In our study, images were obtained at b values
gradient in a series
gradient
z-axes duration,
were
the gradient in eight incre-
Residual spectral after
25#{176}C, the measured diffusion coefficients of methylene chloride, acetone, cyclohexane, benzene, and water were all within 5% of values in the published literature (9 and reference cited therein), regardless of the gradient direction chosen. No changes in the position, intensity, or line width of the water line shapes could be detected at delays longer than 3 msec after a gradient pulse of 5.6 C/cm along any axis. From this assessment of the gradient recovery time, the minimum delay after the diffusion-sensitizing gradient pulses in the spin-echo diffusion pulse sequence was set at 6 msec. To test the utility of MR imaging in the assessment of anisotropic waten diffusion, axial diffusion-weightAt
is due
detail images,
images
acetone,
pulse quined
of
of the same parameters the cat model. Images
by varying sec/mm2
pure,
reliability
an
hypointensity from that voxel. Conversely, slower apparent diffusion will produce a smaller signal attenuation, resulting in regional relative hypenintensity in those fast-diffusion voxels. The image,
use with
benzene, were
observed
of a “pure-diffusion” incoherent motion)
the
effects
y-, and (pulse
and
measurements,
with used
ments) and the x, y, and z axes
this, imwith inter-
images
accuracy
acquired (0-1,413
hexane,
magnitude
the
diffusion
were factor
play
Thus,
calculation travoxel
check
obtained as those
phase-encoding
diffusion-weighted are
To
fre-
the diffusion-sensitizing other gradients may robe
study
inherent
0”
Radiology
. 441
5.
4.
Figures
4, 5. (4) Coronal nongated, diffusion-weighted, spin-echo images of a cat at the level of the inferior colliculi. MR parameters are similar to those of Figure 2 (1,800/80, four signals averaged, 3-mm section, 80-mm field of view). Gradient strengths are the same as those used for Figure 2 (b . 0 sec/mm2 in A and 1,413 sec/mm2 in B-D). The direction of the applied diffusion-sensitizing gradient was varied among the x axis (left-to-right direction in B, white arrow), y axis (vertical direction in C, arrow), and z axis (out-of-plane direction in D). When white matter tracts are oriented parallel to the direction of applied diffusion gradient, fast directional (relatively unhindered or unrestricted) diffusion of water protons is indicated by regions of hypointensity. Thus, in B, transhemispheric fibers, such as the corpus callosum (open arrow) and posterior commissure, have low signal intensity. Water diffusion perpendicular to applied gradient direction is slower (more hindered by the structure and orientation of white-matter bundles), which accounts for the high signal intensity of the corona radiata in B (solid black arrow), and corpus callosum and posterior commissure in D. No clear relationship between white matter signal intensity and gradient direction can be discerned in C, because the plane of applied gradient is oblique to the orientation of most white matter tracts at this level of the neuraxis. (5) Subtraction of the axial y axis (A) and x axis (B) direction-dependent diffusion-weighted images produces images that show little signal from gray matter regions (C [A minus B], D [B minus A]). Subtraction images clearly demonstrate the presence and orientation of white matter. Absolute image intensities of subtraction images, produced from diffusion-weighted images, do contain Ti and T2 information but are useful in visualization of directional effects.
cm2/sec ± 0.1 X i0(z direction). The corpus calbosum was also observed to be an ordered structure from the pronounced dependence of the apparent diffusion coefficient on the z and x axes (0.35 X iO cm2/sec ± 0.07 X i0and 1.3 X 10 cm2/sec ± 0.12 X i0-, respectively). The profound orientational dependence of white matter signal intensity can be observed on coronal diffusion-weighted images from a different cat acquired along all three magnet axes together with the comesponding spin-echo b = 0 image (Fig 4). This dependence of white matter
ages
signal
isotropy
intensity
can
also
be
observed
the corresponding nongated, diffusion-weighted images at one axial level of the cat shown in Figure 4 and is even more pronounced on the respective subtraction images (Fig 5). To assess the contribution of possible artifacts, additional experiments (Fig 6) were performed. To assess the effect of the cardiac cycle on image intensity, gated (to end diastole [Fig 6a] and to end systole [Fig 6b]), nongated (Fig 6c), and postmortem (Fig 6d) imin
442
#{149} Radiology
1,413
dress
along the x direction with all other parameters (eg, TR) remaining constant. No apparent differences between cardiac-gated images and nongated images were observed. The
tion
were
obtained
at b values
of
sec/mm2
postmortem
diffusion-weighted
age displayed isotropy
the
observed
image;
the
only
matter
hypenintensity
on
pattern the
difference
of an-
was due
which cardiac-gated, weighted images quency-encoding
were sitizing (magnitude
in
to
Figure
gray
7, in
direction).
gradient
z axes
fre-
interchanged
These
sagittal junction)
along
in Figure images
cord
spinal
of
(near
the
shown
in
8. post-
cerviFigures
10 are in agreement with the observations made above. Fast-diffusing water protons in CSF and blood appear hypointense. Furthermore, gray matter in the spinal cord shows no
clear
directional
white
whereas be
im-
ages, in which the cross terms are different, produced similar contrast charactemistics, implying that the anisotropic contrast seen on the image is due primarily to the diffusion-sensitizing gradient pulses. To further ad-
the
dependence,
matter
predicted
entation
and
pulses
are shown
and
mortem
can
obtained with the diffusion-sengradient pulses held constant and
x and
9 and
postmor-
diffusionwith phaseaxes
sion-sensitizing
cothoracic
tern pathophysiobogic processes (8,20). The effect of the normal gradient pulses and the cross term between them and the diffusion-sensitizing gradient pulses on the observed anis shown
postmortem
Axial
premortem
contribution of moto cardiac-gated, diffuimages, pre- and images with the diffu-
relative
sion-weighted
im-
same
the
artifacts
in
applied
by
This
anisotropic
ubamby
well
visualized
to
effect when
diffusion-weighted
rectional
(D in Fig
subtracted
intensity
anatomic
the diffusion-sensitizing
relation
dient.
are
signal its
on-
direction
of gra-
is partic-
diimages
two
9).
DISCUSSION On
the
presented
(1,2),
basis
of principles originally Stejskal and Tanner orthogonal direction-depenby
August
1990
pattern of diffusional the axes.
6.
dent diffusion gradients have been used to probe anisotropic influences on molecular diffusion in a variety of model systems (9-11) and tissues in vitro (12,14). To accurately study ondening in matter, the residual “background” gradients (magnet homogeneity, section-selection gradients, frequency-encoding gradients), must be small or negligible in effective strength (the b value) compared with the strength of the applied diffusionsensitizing gradient pair in the pulsed-gradient experiment. In addition, the suppression of eddy currents must be sufficiently thorough to avoid image distortions and gradient mismatch. One test of this latter prerequisite is that isotropic diffusion must, by definition, subtract out from diffusion-weighted images. This has been clearly demonstrated here with the phantoms studied and images obtained. Figure 2, in particulam, is a good example of gradient behavion and of the large amount of ondening (the degree of water proton diffusional anisotropy) observed in the oriented fibers in most fibrous fruits and vegetables. Diffusional anisotropy of small molecules, such as water protons, can be induced by any oriented lattice that presents a nonmandom array of barriers that hinders diffusion or motion in one direction relative to another. Anisotropy, however, can also be caused by “restricted” or cornpletely hindered diffusion, such as that, for example, imposed by cellVolume
176
#{149} Number
2
membrane barriers in elongated cells. Because the diffusion coefficient of free, unhindered water at 25#{176}C-37#{176}C is about 2-4 X i0 cm2/ sec (3,4,7,9) and the spin diffusion time is approximately #{244}/3 or 30 msec, the root mean square path length of proton diffusion along any one axis can be calculated to be 12-16 trn. In biologic sytems, restricted megional (intracellular) proton diffusion will occur if the cellular diameten is less than 12-16 jzm. Restriction may cause the apparent diffusion coefficient of intracellular protons to be smaller than that for protons in the extracellular or interstitial compartments; consequently, the resulting diffusion-weighted image intensity will be less attenuated compared with that in an area in which free diffusion prevails. A quantitative assessrnent of the effect of restriction can be derived from MR diffusion measurements in which the spin diffusion time (i 6/3) is systematically varied (2,12,14,21). The same amgument applies to restricted or hindemed diffusion of water protons through ordered barriers, in which the mean free path length will be longer in one direction than another. One consistent observation from the in vivo diffusion-weighted irnages is that the apparent diffusion rates of water protons in the white matter of the brain and spinal cord are fundamentally different from -
-
anisotropy
is retainedon
swapping
of
that of the gray matter. Cray matter diffusion does not show diffusional directional dependence, whereas white matter apparent diffusion is very anisotropic. It can be argued that factors other than orientationdependent diffusional anisotropy in white matter are responsible for the observed contrast that is charactenistic when the direction of diffusionsensitizing gradient pulse is varied. Effects of perfusion or physiologic pulsations or vibrations can be ruled out as the major cause of this apparent anisotropic diffusion in white matter by the observations that the degrees of anisotropy are similar for gated, nongated, pmemortern, and irnmediate postmortem images (Figs 6, 8-10). Diffusion-sensitizing gradient mismatch and other gradient effects would have resulted in observable artifacts in the subtraction images as well as incomplete subtraction of the phantom signal, which diffuse isotropically. The similarity in the images obtained with switched axes demonstrates the relative negligible effect of the cross terms between the phaseand frequency-encoding gradient pulses and the diffusion-sensitizing gradient pulses (Fig 7). The in vivo directional diffusionweighted images (Figs 3-10) indicate that water proton diffusion anisotropy is best observed in the large-diarneter, fast-conducting motor and somatosensory nerve fibers that form Radiology
#{149} 443
9.
8.
Figures 8, 9. (8) Premortem (A and C) and postmortem (B and D) diffusion-weighted images at a b value of 1,413 sec/mm2. The cardiac-gated premortem pattern of anisotropy (with the direction of the diffusion-sensitizing gradient along the x axis in A [arrow] and along the z axis in C) is retained post mortem. No windowing of the images was performed. The postmortem hypenintensity is thought to be due to pathophysiologic changes (8,20). (9) Axial diffusion-weighted spin-echo images of immediate postmortem cat spine near the cervicothoracic junction (1,800/80, four signals averaged, 3-mm section, 60-mm field of view; original magnification, X2) for b 0 sec/mm2 (A) and 1,413 sec/mm2 (B, C). The direction of the applied diffusion gradient was varied between the z axis (along the long axis of spinal cord in B) and the y axis (vertical direction in C, arrow). Fast-diffusing protons of CSF in the subarachnoid space (black arrow in A), which are hypermntense with T2 weighting, are hypointense on all diffusion-weighted images. The H-shaped central gray area has a slightly lower signal intensity than does CSF on the T2-weighted image (A). Same area has a variable appearance on the diffusion-weighted images because the myelinated nerve fibers, which enter spinal cord via dorsal roots and exit via ventral roots, also run through the central gray area. Heavily myelinated large-diameter axons, including the proprioceptive fibers in the dorsal columns, appear hypointense when the diffusion-gradient direction is parallel to the long axis of the cord (in the z direction in B). White matter appears bright when the gradient direction is in the transverse plane (C). Subtraction of the two diffusion-weighted images (D) shows the effect of diffusional anisotropy, reflecting the predominantly cephalocaudal orientation of white matter in the spinal cord.
the posterior limb of the internal capsule. In addition to the heavily myelinated propnioceptive axons, optic and olfactory nerves, corpus calbosum, and anterior and posterior cornmissures appear to be associated with significant anisotropic effects. This ordering or hindering of water diffusion or motion probably occurs along the axis of the individual neurofibnils as well as along the axis of the myelm sheath itself. Thus, it seems possible to determine a priori the onientation
of
the
white
matter
tracts
from
knowledge of the direction dependence of the apparent diffusion. Perhaps the clearest correlation between anisotropy observed on MR images and the underlying anatomy can be seen in the altered signal intensities in spinal cord white matter as the dinection of the diffusion gradient was systematically varied. Apparent diffusion
was
faster
(yielding
signal
hy-
pointensity) along the axis of the spinal cord when a z-direction gradient was employed, whereas a diffusion gradient applied perpendicular to the axis of the cord (x-direction of 444
#{149} Radiology
gradient)
detected
slowed
diffusion
(high signal intensity) in white matter. Heavily diffusion-weighted MR imaging is a new and unique tool for the noninvasive assessment of white matter orientation. Measurements of the directional-apparent diffusion coefficients and the translational path lengths may be used to derive information concerning barmier sizes and directional diffusion, from which the barrier or lattice dimensions can be quantified. By oblique application of the diffusion gradients, off-axis onentations can also be studied. Clinical applications of gradient direction-selective diffusion MR imaging
are
immediately
larger
be achieved
gradient
clinically
of
MR
imaging
technique
the
strengths
through
can
the
local
orientation
of
through ton
coils.
Utilization
apparent
matter
improve assessment
disorders,
infamcts,
of water prohas the potenour underof demy-
diffusion
elination
of this to determine
white
measurement
tial to greatly standing and
white
matter
involving white and neonatal brain and development. U
neoplasms
matter
tracts,
spinal
cord
References 1.
Stejskal
EO,
Tanner
measurements: ence 2.
obvious.
While the threshold and sensitivity of the anisotropic effect is inherently rebated to the available gradient strength of the diffusion gradient pair in the sequence, irrespective of whether a Stejskal-Tanner spin-echo (1,2), steady-state free precession (5,22,23), or echoplanar (6) approach is used,
use
3.
Spin field
tropic,
restricted Phys
diffusion 1965;
pres-
gradient.
and
flow.
43:3597-3603.
D, Breton
E, Lallemand
ier P. Cabanis E, Laval-Jeantet aging of intravoxel incoherent application neurologic
the
1965; 42:288-292. Use of spin echo in pulsed gradient to study aniso-
Chem Bihan
diffusion in
of a time-dependent
Chem Phys Stejskal EQ. magnetic-field
Le
JE.
spin-echoes
to diffusion disorders.
D, GrenM. MR motions:
im-
and perfusion Radiology 1986;
in
161:401-407.
4.
Le Bihan D, Breton E, Lallemand M-I. Vignaud J. Laval-Jeantet
M.
tion voxel
in intraimaging.
diology
of diffusion incoherent 1988;
and perfusion motion MR
D,
Aubin
SeparaRa-
168:497-505.
August
1990
13.
14.
Hansen mobility
JR. in
Pulsed muscle
NMR study of water and brain tissues. Bio-
chem Biophys Acta 1971; 230:482-486. Cleveland GG, Chang DC, Hazelwood CF. Rorschach HE. Nuclear magnetic resonance meaurements of skeletal muscle anisotropy
of the
intracellular
diffusion
water.
coefficient
Biophys
of the
J 1976;
16:
1043-1053.
15.
Fullerton
GD,
entation and
Cameron
IL, Ord
of tendons its
effect
ology
in
on
1985;
T2
the
VA.
On-
magnetic
relaxation
field
times.
Radi-
155:433-435.
16.
Roemen PB, Edelstein WA, Hickey JS. Self-shielded gradient coils (abstn). In: Book of abstracts: Society of Magnetic Resonance in Medicine 1986. Berkeley, Calif: Society of Magnetic Resonance in Medicine, 1986; 1067-1068.
17.
Mansfield ic screening
P. Chapman of coils
B. Active static and
for
magnettime-de-
pendent magnetic field generation NMR imaging. J Phys E Sci Instrum 18.
19:540-545. Turner R, Bowley
of switched
a. Figure
Phys
b.
19.
10. Sagittal diffusion-weighted images (94-mm field of view) show white matter anisotropy in spinal cord and cerebellum. (a) Image was obtained with the diffusion gradient applied along the z axis (parallel to the spinal cord axis, arrow). (b) Image was obtained with the gradient applied along the y axis (horizontal direction, arrow). Differences in signal intensity between these images and the images in Figure 7 arise only from white matter contributions
to
signal
intensity
and
mirror
the
observations
in
Figure
E Sci
Hayes An
9.
Merboldt MH,
K-D, Hanicke
W,
“diffusion” suits
in
using
Magn 6.
Bruhn
R, Le
Echo-planar aging
M.
MRI
human
brain:
new
CE-FAST
Med
1989;
Bihan
D,
stracts:
Tesla
Society
Medicine Magnetic
and
of 9.
re-
sequence.
9:423-429. J, Pekar
Delannoy
diffusion
at 2.0
J, Gyngell
Frahm
Deimling
the
a modified
Reson
Turner
H,
and (abstr).
perfusion In:
of Magnetic
J.
of Ab-
Resonance Society 1989;
C, Henriksen
P.
of 11.
Thomsen
vivo measurement in the human
nance
Ring
In-
of water self-diffusion by magnetic reso-
brain
imaging.
0,
Acta
Radiol
1987;
spin-echo
sion.
Prog
28:353-
ischemia
Y, Mintorovitch J, et of regional cerebral
ME, Cohen detection in cats:
comparison
of diffusion-
176
#{149} Number
2
studies NMR
21.
14:330-346.
transform
of molecular 1987;
12.
whole
ME,
of acute
stroke:
weighted
and
imaging.
JE.
Transient
im-
63:
with
T2-
susceptibility-en-
AJNR
1990;
diffusion
by
to NMR field
1985;
correlation magnetic
MR
NMR
J. Mintorovitch MR imaging
Kucharczyk
partitioned
JF, et al. radio-
body
Reson
Diffusion-weighted
pulsed
19:1-
for
I, et al.
plication
diffu-
Schenk
homogeneous
at 1.5 T. J Magn
429. Tanner
19:876-879.
WA,
622-626. Moseley
tern
pulsed-gradi-
Spectroscopy
Lindbloms
G.
for
of dynamics
studies
Scand Tanner
Cooper
NMR
diffusion:
of membrane
1981; JE.
1976; RL,
sion
22.
a method
and
mesophase
lipids.
Acta
Le Bihan tion cession.
Chem 23.
B35:61-62. Measurements
in a sys-
permeable
barriers:
measurements
ap-
with
J Chem
gradient.
11:423-
Phys
a
1978;
of self-diffu-
systems by magneticspin echo methods. In: CG, eds. Magnetic resoscience. Chemical
D.
imaging
using
Magn
incoherent steady
Reson
Med
mo-
state
free
1988;
7:346.
pre-
Le Bihan D, Turner R, Macfall JR. Effects of intravoxel incoherent motions (IVIM) in steady-state
Res-
Intravoxel
aging: imaging.
free
application Magn
precession
(SSFP)
to molecular Reson
Med
im-
diffusion 1989;
10:324.
So-
16-29. Chang
CJ, Ancker-Johnson 1974;
Volume
1990;
spectroscopy.
screening
gradients.
69:1748-1754.
ciety,
Moseley al. Early
and
nance in colloid and interface Washington, DC: American
361.
8.
ent
Med
Fourier
sion in colloidal field-gradient, ing HA, Wade
139.
7.
Reson P.
structure
in
1989. Berkeley, Calif: Resonance in Medicine,
Magn Stilbs
MRI
45.
10.
im-
Book
T2-weighted
1986;
highly coil
hanced 5.
Instrum
efficient
aging
Passive
field
CE, Edelstein
frequency
20.
RM.
magnetic
in 1986;
in biophysical
DB.
Young
B. systems.
AC,
Restricted
Martin
diffu-
Biophys
14:161-177.
Radiology
#{149} 445
