high angular resolutioo diffusion mr for the determination ... · high angular resolution diffusion...
TRANSCRIPT
High Angular Resolutioo Diffusion MR for the Determination of Fibre Structure
Elisabeth von dem Hagen
A thesis submitted in conformity with the requirements for the Degree of Master of Science
Graduate Department of Medical Biophysics University of Toronto
O Copyright Elisabeth von dem Hagen 2001
A uisitions and Acquisitions et B8iographic SeivlccK services bibliographiques 395 W W l g b n Street 395, rus Wellington CHtawaON K I A W OnawaôN K t A W canada Canada
The author has granted a non- exclusive licence dowing the National Library of Canada to reproduce, loan, distribute ois seli copies of this thesis in microform, paper or electronic formats.
The author retains ownership of the copyright in this thesis. Neither the thesis nor substantid extracts fiom it may be printed or otherwise reproduced without the author's permission .
L'autew a accordé une licence non exclusive permettant à la BiblioWque nationale du Canada de reproduire, prêter, distribuer ou vendre des copies de cette thèse sous la fonne de microfiche/film, de reproduction sur papier ou sur format électronique.
L'auteur conserve la propriété du droit d'auteur qui protège cette thèse. Ni la thèse ni des extraits substantiels de celle-ci ne doivent être imprimés ou autrement reproduits sans son autorisation.
High Angular Resolution Diffusion MR for the Determination of Fibre Structure
Elisabeth von dem Hagen Master of Science, 2001
Department of Medical Biophy sics University of Toronto
Abstract
Diffision MR has recently been developed as a technique for tracking neuronal pathways
in the brain. Current methods, however, are unable to resolve fibre orientation in voxels
with multiple fibre tract directions. In this thesis, a physical mode1 of restricted difision
analogous to that in nerve fibres is presented. Difision measmments at high angular
resolution in sarnples with different fibre orientations are compared with theoretical
calculations for restricted diffision in a cylindrical geornetry. Orientational diffusion
measurements are shown to reflect fibre geometry and theoretical predictions to within
10%. Theoretical calculations of signal decay for a single fibre cm be used to predict the
angular dependence of diffision coefficients for any distribution of fibre orientations, and
simulations of fibre crossings at high gradient strengths refiect the presence of the
crossed fibres. Theoretical predictions rnay therefore allow experimental measurements
of similar in vivo fibre geometry to be resolved.
Acknowledgements
1 would like to express my gratitude and thanks to everyone who has helped me
throughout the past two years. At Sunnybrook, 1 want to thank Greg Stanisz for being an
indispensable source of 'diffbion' and SMIS information. Thanks also to Carneron
Chiarot for providing me with the EM images of the plastic tubing. 1 am especially
grateful to al1 those who took the time to read through this thesis, and provided me with
helpful comments and suggestions. 1 would like tu thank the membea of my supervisory
cornmittee, Simon Graham and Peter Burns, for their help and support. i would especially
like to thank my supervisor Mark Henkelman for his guidance and for teaching me to
think like a scientist. The past two years would not have been the same if it Iiadn't been
for al1 those who dragged me out and made me laugh: Mark, Michelle, Warren, Gal,
Nick, Nom, Bart, my roommates and many others. Thanks to Josette, my training
partner, for several triathlons, a marathon, and countless conversations. 1 also had the
pleasure of being a part of the U of T Nordic ski team for which 1 am very grateful.
Finally, and most importantly, 1 would like to thank my fmily for their continued
support.
iii
Contents . . Abstract ............................................................................................. 11 . . Acknowledgements ............................................................................ 111
Contents ............................................................................................ iv .................................................................................... List of Figures v
........................................................................... List of Abbreviations vi
Chapter 1 ........................................................................................................................ 1 Introduction
1 . 1 Motivation ............................................................................................................... 1 1.2 Outline ..................................................................................................................... 2 1.3 Diffision in MR ...................................................................................................... 3
.......................................................................... 1.3.1 Measurement o f Difision 4 1.3.2 Restriction ....................................................................................................... 7 1.3.3 Anisotropy .................................................................................................. 10
.................................................................................................. 1.3.4 Applications 1 1 1.4 Thesis Staternent ................................................................................................... 17
Chapter 2 .................................................................................................................... 18 High Angular Resolution Diffusion MR for the Determination of Fibre Structure
Submitted as a paper to Magnetic Resonance in Medicine Sept 2001 2.1 Introduction ........................................................................................................ 18 2.2 Experimental Methods ......................................................................................... 20 2.3 Theoretical Methods ........................................................................................... 24 2.4 Results and Discussion ......................................................................................... 26 2.5 Conclusions ........................................................................................................... 36
Chapter 3 .................................... .... ......................................................................... 38 Future Work
........................................................................................................... 3.1 Introduction 38 3.2 Orientational Diffision Measuements in White Matter ...................................... 39
3.2.1 A Modei of White Matter .......................................................................... 40 ........................................................................... 3.2.2 Determining fibre structure 45
3.3 Other MR techniques for the rneasurement of difision ...................................... 47 ............................................................................. 3.3.1 Weighting of Eigenvalues 47
......................................................................... 3.3.2 Diffision Spectrum Imaging 48 ........................................................................... 3.3.3 Multiple Diffision Tenson 49
3.4 Conclusions ........................................................................................................... 50
List of Figures
Fig . 1 : Stejskal-Tanner pulse sequence ................................................. 4
............................................................. Fig . 2: Diffusion decay curve 7
Fig . 3: Diffision tensor and ellipse ...................................................... 1 5
Fig . 4: Electron micrograph ............................................................... 21
Fig . 5: Fibre sample orientations .......................................................... 22
Fig . 6: PFG MSE pulse sequence ........................................................ 23
Fig . 7: Coiled fibre approximation ...................................................... 26
Fig . 8: Decay curves for X. Y. Z directions .......................................... 27
Fig . 9: Experimental and theoretical ADC contours ............................... 29
................................. Fig . 10: Theoretical decay curves to hi& b values 31
Fig . 1 1: Simulated ADC contours for crossed fibres .............................. 33
Fig . 12: In vivo nerve fibre mode1 ...................................................... 44
Fig . 13: Crossing and kissing fibres .................................................... 46
List of Abbreviations
ADC
DT-MRI
F m
MR
MRI
PFG MSE
RF
SNR
Apparent Diffision Coefficient
Diffusion-Tensor Magnetic Resonance Irnaging
Full Width at Half Maximum
Magnetic Resonance
Magnetic Resonance Imaging
Pulsed Field Gradient Multi Spin Echo
Radiofiequency
Signal-to-noise Ratio
Chaptea 1
Introduction
1.1 Motivation
Neuronal connections mediate complex behavioural and cognitive processes by
facilitating signalling between different areas of the brain. The neural networks involved
in these higher order cerebral functions are poorly understood, due in part to the difficulty
in elucidating the elaborate structures underlying integrated nervous system activity.
Determining nerve fibre tract connectivity wouid enable the establishment of a neural
network map to link regions of the brain. The pathways or fibre tracts associated with
specific functions could be used to correlate functional regions, which may provide
insight into human behavioural processes. In addition, the detection of changes in the
organisation and structure of nerve axons or fibres rnay be early indicators of pathologie
degradation of the nervous system.
Until recently, establishing fibre connectivity in humans has been bas-d on invasive
tracer studies in animals and subsequent human-primate conelations (1 -3). These
methods provide limited information on complcx human cognitive functions or disorders
associated with higher level cerebral processes, since these cannot always be inferred
fiom animal studies.
The sensitivity of magnetic resonance (MR) to molecular diffusion, specifically of water,
has been the focus of much research since it was established as an imaging signal contrast
method in the early 1980s with numerous clinical applications (4-6). In brain white
matter, water preferentially difises along nerve fibres. This preferred direction of
molecular motion can be detected by MRI and used to establish fibre tract orientation and
connectivity non-invasively (7-10). This technique, however, is still at the pre-clinical
stage since several difficulties must be resolved before accurate and continuous tracking
is possible.
Multiple fibre bundle orientations within a single voxel are dificult to resolve with
current techniques. While several new methods have been proposed to address this
problem (1 1 -14), in some cases their usefulness in establishing fibre orientation has yet to
be determined. This thesis will adàress one such technique, high angutar resolution
diffision measurements, in an in vitro mode1 malogous to nerve fibres, in order to
determine its adequacy in i n f ' n g fibre orientation.
The background for both the principles behind diffision MRI and its applications are
discussed in the rest ofchapter 1. At the end of the chapter, the specific aims of the thesis
2
are revealed. Chapter 2 describes the development and application of the experimental
and theoretical models to examine diffusion in single voxels with different fibre
orientations. Finally, in Chapter 3 the implications of the thesis for possible fuRue work
are discussed.
1.3 Diffusion in MR
In MR, the measured signal originates from protons of hydrogen nuclei. When placed in a
magnetic field, the magnetic moment vectors of the protons, or spins, tend to align in the
direction of the magnetic field, and resonate at the Larmor frequency au, which is
proportional to the magnetic tield Bo:
0 0 = yB0 [il
where y is the gymmagnetic ratio equal to 42.58 M H f l . By applying a radiofrequency
(RF) pulse tuned to the Larmor frequency a,, the spins can be excited out of equilibriurn
into the transverse plane, i.e. orthogonal to the main magnetic field Bo. The rotating
magnetisation vectors of the spins induce an electromotive force in a receiver coil, which
constitutes the MR signal.
Water molecules, which account for most of the hydrogen in the body and are therefore
the basis of the MR signal, are not stationaiy and exhibit random Brownian motion. As a
result, spins may move during the time of the experiment. In the presence of magnetic
field inhomogeneities, spins will move and precess at different fiequencies depending on
the gradient strength at their location. This loss of phase or coherence of spins results in a
loss in detected signal. Although undesirable in most MR applications, it is this change in
signal due to molecular diffision which forms the basis for MR diffision studies.
1 Al Measurement of Diffusion
Although the measurement of diffision using spin echoes was first described by Hahn
(1 5) in 1950, it was Stejskal and Tanner's seminal experiments in 1965 which enabled
very precise measurements of difision through the use of pulsed tield gradients (16).
Their expenmental pulse sequence is depicted in Fig. 1 and is based on the simple spin
echo sequence. In spin echo, a RF pulse tips the magnetisation by 90' (n/2 radians), such
that al1 the magnetisation lies in the transverse plane. The application of a second n RF
pulse reverses the loss of phase experienced by spins precessing at different frequencies.
By applying time-dependent gradients, or magnetic fields of a certain duration that
change strength linearly with position, the spin echo sequence can be used to mesure
diffision by deliberately inducing phase shifts in the spin population as follows.
Fig. 1. StejskaCTanner Pulse Sequence with gradient strength G, gradient duration 6. and diffusion time A. The applied RF pulses are denoted by the
their tip angles in radians.
When a gradient G is applied along a certain axis, e.g. the z-axis, the spins along that axis
are given a phase shift that depends on their position:
4, = y 1: Gzi dt = y06zi [2]
where zi is the spin position, G is the gradient strength, 6 its duration, and y is the
gyromagnetic ratio. Following the n RF pulse, the phase shift due to the first gradient
pulse 4l will become -41. AAer a tirne A fiom the onset of the first gradient pulse, the
application of a second gradient pulse will impart an additional phase shift to the spins:
+2 = y / ~ A + S Oz2 dt = yGGzt [3]
where z2 is the spin position during the second gradient pulse, and G, 6, its strength and
duration. The net phase shifi following the application of the second gradient pulse
becomes:
A+ = $2 - 41 = yGw1-z~) [4 1
if the spins have remained stationary, zl=q and the net phase shift is zero. However, if
the spins have moved, the phase shift will be non-zero and will cause a loss in measured
signal. In MR, the detected signal is proportional to the magnetisation. The attenuated
signal due to diffusion, SB,, is then simply the surn of the individual magnetisations of
the moving spins (Zexp (iyG6(zl - zr))). This surn can be evaluated if the probability
distribution P(z11z2, A) of a spin beginning at 21 and moving to z2 in a time A is known.
The attenuated signal equation becomes:
Siso = p(z) P(zi 1 22, A) exp (iyG6(z1 - 22)) dzi dz2 F I
where p(z) is the spin density, and P(zI Jz2, A) is the spin displacement probability .
For free diffusion in one dimension, the probability that a spin will move fiom position zi
to 22 in a time A is Gauuian (17):
P(zi 1 22, A) = I I(~IIDA)'~ exp (-(zi - z ~ ) ~ / ( ~ D A ) ) [61
where D is the diffision coefficient. Integmtion of the signal equation by substituting Eq.
[6] into Eq. [5] yields:
SIS, = exp ( - ( y ~ ~ ) 2 AD) [7]
or altematively:
ln (S/S,) = - (?~6)~ AD 181
which relates difisional signal loss to the difision coefficient. Equation [8] assumes
that 6 is short with respect to A. However, this assumption is rarely mie in MR difision
experiments, where gradient strength limitations result in the application of longer
gradient durations to produce observable diffusion effects. In these cases, Eq. [8]
becomes, to a first approximation in 6, (1 7):
ln (SB,) = - ( y ~ ~ ) 2 (A-613)D
This equation can be rewritten as:
In (SB,) = - b~ El01
where b = ( @ ~ ) ~ (A-W3). In order to determine experimentally the diffision coefficient,
the signal attenuation is measured at different b values. The b value is changed by
increasing either the gradient strength, duration or diffision time. The logarithm of the
measured signal attenuation is plotted as a function of b, and the dope of this line is the
diffision coeficient (see Fig. 2 for example).
Fig. 2. Diffusion decay curve or 'b-plot' where b=(y~6)2 (6513). The dope of the line is the diffusion coefficient D.
For fieely diffising spins, the measurement of the difision coefticient is stniightforward
since the slope is constant, Le. the relationship between the signal attenuation and the
diffision coefficient is monoexponential. In tissues, however, the translational motion of
spins may be restricted due to cellular structures, which affect the signal decay curve in a
number of ways.
1.3.2 Restriction
Diffision is restricted when boundaries obstnict the movement of molecules. In tissues,
the presence of cellular structures, such as cell membranes, create barriers to the diffising
spins. The measurement of restriction is dependent on experimental parameters. If the
experimental difision tirne is short, spins may not have time to reach barriers and will
appear to be diffising fieely. In this case, the displacement probability remains Gaussian.
As the diffision time is increased, a greater number of spins will encounter restrictive
boundaries, and their displacement distribution deviates fiom the case of unbounded
diffusion. This deviation is due to the relationship between diffision time and the
distance spins will have moved in that time. Einstein showed that the root mean square
displacement for fiee diffision in one dimension is related to the diffusion coefficient and
diffision thne A by (1 8):
= DA)'^ [ l u
In a restrictive environment, the root mean square displacement or diffision distance will
no longer Vary linearly with difision tirne. Instead, when the difision distance
approaches the length of the restrictive cornpartment, the distance spins have travelled
will no longer increase in direct proportion to the diffision time. As a result, the effects
of restriction in MR experiments become apparent when the diffision distances are
comparable to the restrictive length.
These effects are dependent on the type of restriction and the shape of the restrictive
volumes. For simple geometnes, such as spins diffising between planes, in spheres or in
cylindrical geometry, the deviation of the spin displacement distribution fiom the
Gaussian fiee diffision distribution has been calculated (1 9-2 1 ), and the exact behaviour
of the signal decay curve is known.
In tissues, however, restricted diffision is much more cornplex. Spins may be diffbsing in
many different geometries due to the presence of water in multiple tissue compartments.
In addition, barriers may be permeable, ailowing exchange between compartments. The
presence of macromolecules may also affect diffising spins by acting as additional
obstacles to their fkee movement. As a result, the presence of restricted diffusion in
tissues is evidenced experimentally by qualitative changes in the signal decay curve.
One of the primary charact~ristics of restricted diffusion is the detection of non-linearity
in the semilog plot of signal attenuation vs b. Although the origins of this upward
cwature - implying slower diffision - remain somewhat controversial, several
explanations have been put forward. Since water is present in multiple tissue
compartments and the voxels in diffision experiments may incorporate several
compartments, the non-monoexponential behaviour of the decay curve has been ascribed
to the presence of multiple diffision coefficients representing diffision in the different
compartments (22-27). Sirnilar changes in the signal decay curve, however. are seen in
expenments with a single restricted cornpartment (28). In such cases, the upward
curvature has been ascribed to complete dephasing of spins with high mobility near the
centre of the restrictive compartqent since higher gradient strengths cause greater
dephasing, such that the detected signal is originating solely fiom spins close to the
restrictive boundary. Thus, diffusion will appear more restricted at higher b and hence
less signal loss will occur, which is depicted by the upward curvature of the signal decay
c w e .
An upward curvature of the signal decay curve alone, however, is not suficient to
determine whether diffision is restricted Le. confined to one or more compartments. It
has been shown that diffusion in certain systems, such as water diffising in the presence
of glass beads, will also demonstrate non-monoexponential signal decay (29). In this
case, spins are said to be obsûucted rather than restricted. When spin movement is
limited by boundaries however, diffision decay curves will aiso exhibit dependence on
diffusion time, by shifting upwards as the latter is increased. As mentioned earlier, this
relationship results fiom a greater number of spins encountering restrictive barriers as the
experimental difision time is increased, which causes a decrease in the measured
difision coefficient. Since the signal equation, as solved in Eq. [9], constrains spins to
have a Gaussian displacement distribution, the presence of restriction will manifest itself
as a decrease in the measured diffision coefficient, i.e. a difference in signai attenuation
for the same b value.
Although the dependence on diffision time and the upward curvature of the decay curve
are signs of restricted diffusion, the b values, or gradient strengths, used in current
clinical applications are rarely high enough for the detection of non-monoexponential
signal decay. The presence and degree of restriction are therefore measured by changes in
tfie initial slope of the diffision decay curve. At these low b values, the relationship
between signal attenuation and diffision coefficient is monoexponential, and the
measured slope on a semilog plot is called the apparent diffision coefficient or ADC.
1.3.3 Aniso'tropy
Due to the presence of restrictive barriers, diffision may not be the sarne in al1 directions.
This dependence of diffision on direction, or diffision anisotropy, can also occur in the
absence of restriction (30,3 l), but in tissues it is ascribed to spins encountering differing
levels of restriction as a function of direction. In such cases, a single scalar diffision
coefficient is insufficient to describe diffision within a voxel. Basser et al. described the
use of the diffusion tensor to replace the diffision coefficient in anisotropic systems (32).
The signal attenuation equation then becomes:
S/S,=~X~(- b g T ~ g )
where D is the diffusion tensor
D=
g is the diffision encoding unit vector, and b = ( y ~ 6 ) 2 ( ~ - ~ 3 ) . The diffusion tensor
diagonal components, Du, D,, and D,, reflect correlations between molecular
displacements dong x, y, and z, whereas the off-diagonal components, Dxy, Dxz, and Dy,,
reflect conelations in difision in orthogonal directions. Since D is symmetric, six
noncolinear diffision measurements are required to estimate the diffision tensor, in
addition to a measurement in the absence of diffision encoding gradients to determine S,.
The diffision tensor is used to determine the degree of anisotropy present in a voxel, and
to detemine rotationally invariant diffision quantities. These measures allow cornparison
of diffisivities in voxels so that relative changes due to pathology may be detected
despite anisotropy (33). Howevcr, its most recent application is in fibre tractography, as
will be discussed below.
1.3.4 Applications
The availability of MR as a tool to measure molecular displacements has been the basis
for extensive research into possible applications for the study of normal and diseased
tissues. One of the primary clinical applications of diffusion MR is in the detection of
changes in water diffision during cerebral ischemia or stroke. Moseley et al.
demonstrated a dramatic reduction in ADC in the region of the inf'arct (5). This regional
ADC decrease, and the resulting change in signal contrast in diffusion-weighted MR
images, is apparent long before changes in conventional Tz-weighted images. As such, it
provides a means of earlier detection of stroke. Changes in ADC have also been obsewed
in a number of other pathologic processes which affect the brain, such as Alzheimer's
disease (34), multiple sclerosis (6), and brain tumours (35). In emphysematous lungs,
measurement of the increased diffision coefficient using hyperpolarized gas techniques
reflects the destruction of lung tissue structure (36).
Changes in tissue structure, and the subsequent change in the measured diffision
coefficient within a voxel, are often indicators of pathology, such as in the detection of
stroke and emphysema. However, the detection of restricted or anisutropic diffision cm
also be used to examine healthy tissue structure. As mentioned above, the diffision
tensor is used to characterize anisotropy in highly structured tissue. To date, the
dependence of diffision on direction has been reported in a nurnber of tissues with a high
degree of structural organisation, such as kidney (37), skeletal muscle (38), cardiac
muscle (39) and most notably, brain white matter (40,41).
Diffusion in White Matter
The dependence of difision with direction in white matter has been attnbuted to the
presence of nerve fibres. Whereas grey matter consists largely of neurons and suppon
cells, white matter is made up prYnarily of bundles of nerve fibres. Axons, or nerve
fibres, are the projection amis of the neuron ce11 bodies found in grey matter. They range
between 1 and 2 0 p in diameter, but may be over a metre in length. In white matter,
their role is to provide communication between different regions of the brain. The mode
of signalling dong the fibre is through electric impulses, and nerve fibres are highly
stnictl;isd xi* ri surrounding layer of insulation known as the myelin sheath.
MR difhsion measurements in white matter have found that diffision along the fibres is
much greater than difision across the fibres. In studies of healthy hurnan volunteers, the
diffision coefficient measured with fibres paralle1 to the gradient is on the order of
1 .OX 1 0%m2/s (4 1). and has been reported as high as 1 . 5 ~ 1 0"cm2/s parallel to the corpus
cailosum (33). Across the fibres, however, the diffision coefficient is much lower,
ranging fiom 0.3 to 0.6~10'~crn~/s (33,41). In addition, diffision measurements at higher
b values (> 1 500 s/mm2) demonstrate non-monoexponential signal decay (22,23 &),
which may suggest that diffision in neural tissue is restricted as well as anisotropic.
The observation that diffision is greatest along the length of nerve fibres has provided a
non-invasive method for establishing fibre directionality. The direction of highest
difision can be used to infer fibre tract orientation within successive voxels and thus
infer the anatomic path of neuronal connectivity.
Fibre Tracking
To date, fibre tracking has been performed using diffision tensor MM (DT-MW). As
mentioned eariier, the diffision tensor for a voxel incorporates more information than a
single scalar diffusion coefficient. ADC measurements along each of the three orthogonai
axes alone are insufficient to properly detemine anisotropy and fibre orientation, since
the gradient or laboratory x, y and z axes rarely coincide with the local fibre axes. The
offdiagond elements of the diffision tensor therefore provide additional information on
fibre structure and correlate displacements along x, y and z when the mobilities in these
directions differ (43).
In order to establish fibre tract orientation within a voxel, six non-collinear diffision
measurements are made to determine the elements of the diffision tensor. Since D is
positive definite and symmetric, its three eigenvectors and associated eigenvalues are
orthogonal:
D E = E A 1141
where A is the diagonal matrix of eigenvalues Li, k2, k3, and E contains the
corresponding orthonormal eigenvectors, arranged in columns. The eigenvalues and
eigenvectors represent the effective diffusivities and local fibre CO-ordinates i i yactively.
They therefore eliminate the need for the laboratory or gradient axes to coincide with the
voxel's fibre axes. The eigenvalues and eigenvectors are used to describe the diffision
ellipsoid, the surface of which represents the mean displacement of spins within the
diffusion time. The eccentricity of the ellipse describes the degree of anisotropy present
within a voxel and provides a pictorial representation of the information contained in the
tensor. Figure 3 depicts the relationship between difîüsing environment (upper row),
difision ellipsoids (middle row), and diffision tensor (bonom row).
Fig. 3. Relationship between diffusing environment (upper row), diffusion ellipsoids (middle row), and diffusion tensor (bottom row). In isotropic diffusion (a), ellipse is spherical and diffusion represented by one diffusion coefficient D. In anisotropic diffusion with coincident fibre and laboratory axes (b) ellipse is elongated and the tensor's diagonal elements are the eigenvalues of the ellipse. In (c) the fibre and laboratory axes do not coincide and the nine tensor elements depend on the relative orientation of these axes.
In Fig. 3% diffision is free and isotropic, represented by the spherical ellipsoid and only
one diffision coefficient. Fig. 3b, on the other hand, depicts a case of restricted diffision
with coincident laboratory and fibre axes. There are three tensor elements representing
diffision along the three principal axes. Findly, in Fig. 3c, the fibre frame has been
rotated with respect to the laboratory frame. Al1 nine elements of the tensor are required
to correlate diffision along the different axes. It is apparent fiom these images that the
eigenvalue with the greatest difisivity represents diffision along the direction of the
fibre and its eigenvector provides the fibre tract orientation. Based on this observation, a
number of tracking algorithm have been developed to perfom continuous tracking fiom
one voxel to the next (7-10).
Although some of the larger fibre bundles in the brain have been traced using DT-MN,
many dificulties must be resolved before fibre backing can become a useful clinical tu01
(10). DT-MN studies are plagued with low signal-to-noise-ratio (SNR), due in part to
fast imaging techniques. In addition, validation of tracking is problematic since the
current gold standard for fibre tractography is histology, which is inappropkite for in
vivo studies.
Currently, one of the largest challenges for fibre tracking is resol-{ing fibres lacking
directional coherence within a single voxel. Although some fibre bundles are relatively
large and highly directional, consisting of 800,000 or more individuai nerve fibres / mm2,
many fibre bundles are smaller, have collateral branches, or are in close proximity to
other fibre tracts. As a result, multiple fibre tracts may be present in an imaging voxel,
each with independent orientations. An example of this situation occurs where fibre
bundles cross. Due to the lack of a predominant direction of diffision, the diffision
tensot and its corresponding ellipsoid are unable to resolve fibre tract orientations in
these voxels. In such cases, six independent difision measurements are insufficient to
characterize the complex molecular motion.
An alternative method to resolving the presence of multiple fibre tracts was introduced by
Tuch et al. (12) and Frank (14). They described the use of high angular resolution
diffusion measurements to provide additional information on difision in regions of low
fibre directionai coherence. This technique consists of applying diffusion-encoding
gradients at a high number of angles to adequately represent diffision in al1 possible
directions, instead of limiting these measurements to the six axes required for the
difision tensor. The resulting ADC measurernents as a fùnction of gradient angle have a
higher degree of angular variability than the diffision ellipsoid. Whereas the ellipsoid
represents the spin displacements in the 'fibre' m e , the measured ADC contour
represents the projection of the spin displacements ont0 the gradient or laboratory axes.
In the case of two crossed fibres, the ADC contour has a cloverleaf shape, in contrast to
the spherical diffusion surface defined by the tensor formalism for the same fibre
structure. Although these diffusion coefficient contours reflect a higher angular
variability, the degree to which they reflect fibre bundle orientation has yet to bc
deterrnined.
1.4 Thesis Statement
The goal of this thesis was to determine the role of high angular resolution diffision
measurements in the determination of fibre tract orientation. An in vitro mode1 of
anisotropic diffision analogous to that produced by nerve fibres was developed, so that
experiments could be performed in a controlled sample with known fibre orientations.
Experimental results for different fibre orientations were compared with theoretical
calculations for diffision in a cylinder to detennine the accuracy and feasibility of
difision meesumnents at multiple angles in determining complex fibre structure.
Chapter 2
High Angular Resolution Diffusion MR for the Determination of Fibre Structure
This Chapter was submitted as a paper to Magnetic Resonance in Medicine in September 200 1,
2.1 Introduction
The signal contrast in MRI is due to tissue molecular content and structure, which affect
such MR parameters as Ti, T2, magnetisation transfer and difision. In diffision imaging
studies, the contrast is based on spin dephasing and the rate of diffision can be measured
by the application of diffusion-encoding gradients (1 6). The presence of ordered
biolopicd membranes, however, can provide obstacles to the difising spins, restricting
their movement in certain directions.
The degree of restriction encountered by spins is thus a reflection of cellular structure.
Echo attenuation curves due to diffision have been denved for spins diffising between
planes (1 9,2O), in spheres (19,20), and in cylindncal geometry (20,2 1). Measurements of
the differences in these echo attenuation plots, however, require extremely strong
diffision gradients or high b values. In clinical imaging, b values are generally
sufficiently small ( < lx1 0' s/cm2) that the diffision decay curve can be approximated as
monoexponential, and estimates of the diffision coefficient can be made by measuring
the slope, or apparent diffision coefficient (ADC). The level of restriction encountered
by spins, or the presence of cellular structures, is then determined by relative changes in
the ADC with diffision gradient orientation.
The dependence of diffision on direction, or diffisional anisotropy, due to restrictive
barriers has been reported in a variety of tissues exhibiting a high degree of structural
organisation. Cleveland et al. first observed anisotropy of difision in skeletal muscle
(38), Reese et al. in cardiac muscle (39), and, more recently, diffision anisotropy has
been linked to the highly structureci nerve fibre tracts in brain white matter (40,41).
In white matter, nerve fibres have a high degree of restriction across their width and
relatively free diffision dong their length. Thus, the measured diffision coefficients
reflect fibre tract orientation (7- 10). By following the direction of highest difision fiom
one voxel to the next, fibre bundles can be tracked and neuraI connections established.
Fibre tracking algorithrns are king actively developed and have successfully traced
many of the larger known fibre bundles in the brain (7-1 0). Preliminary studies have
already shown that establishing in vivo neural connectivity may provide a means of early
disease detection for pathologie processes afTecting the integrity of neuronal connections
(44,45), as well as providing a greater understanding of fùnctional relationships between
different regions of the brain.
There are, however, outstanding issues which must be resolved before fibre tracking can
successfully be implemented as a clinical tool. Current techniques based on diffision
tensors are inappropriate for assessing multiple fibre tract orientations within a voxel,
which may be attributed to the fact that diffision coefficients dong six noncolinear axes
are insufficient to describe the complex underlying fibre structure (1 0). Wedeen et al.
recently proposed imaging diffision spectra (1 1) as an alternative, in which the
displacement probabilities of spins are used to detemine fibre orientation. High angular
resolution difision measurements have also been explored to characterize diverse fibre
structure within a single voxe1(12,14), however, the ability to resolve fibre orientation
using this technique bas not been addressed. Although many new techniques are being
explored, to date there has been no report of evaiuating diffision acquisition schemes
using a model of restricted diffision similar to nerve fibres.
In this paper, we present a physical model andogous to diffusion in nerve fibres. By
performing single voxel diffision experiments, we examine the effects of changing fibre
orientation and multiple fibre orientations on diffision measurements at high angular
resolution. The experimental results are compared with a known theoretical model for
diffusion in a restricted cylindrical geomeuy, and these theoretical calculations are used
to assess the use of difision rneasurements at multiple angles and high b values in
determining fibre bundle orientation.
2.2 Experimental Methods
Hollow plastic fibres (PTFE ultramicrobore tubing P-06417-70, Cole-Parmer Instrument
Company) were used to create an experimental model of white matter fibre bundles. Each
sample was prepared with a length of 2.5 m of tubing. The tubing, or fibres, had an inner
diameter of approximately 50 Pm (roughly twice the size of larger scale nerve fibres in
the brain) and an outer diarneter of 325 p, as show in the eiectron micrograph in
Figure 4.
Fig. 4. Electron micrograph of plastic tubing in cross-section with outer diameter 325 Fm and inner diameter 50 Fm.
Water was inserted into the fibre by placing the tip inside a 22 '/1 gauge needle and gluing
the fibre to the needle. Water was injected manually into the iength of fibre and was
sealed by melting both ends after removing the needle. The fibres were then arranged to
mate three different fibre bundle orientations: aligned, coiled, and oriented at random
(Fig.5).
Fig. 5. Orientation of fibre samples A) aligned fibres 8) wound fibres C) randomly oriented fibres. x, y, and z represent the CO-ordinate axes, G the gradient axis, 0 the angle between x and G, and Ba the direction of the main magnetic field.
The samples were 1 cm in height, 0.5 cm in width, and were placed inside borosilicate
g las tubes and oriented such that the long axis of the sample was perpendicular to the Bo
direction (Fig. Sa) inside the superconducting 1.5T magnet (Naiorac, Martinez, CA)
controlled by a programmable console (SMIS, Surrey, England). Diffision measurements
were acquired using a pulsed field gradient multi spin echo (PFG MSE) pulse sequence
(Fig. 6). This sequence uses a series of x refocusing pulses between the difhsion gradient
pulses to diminish the effects of susceptibility and diffision caused by susceptibility
gradients (46) (any additional gradients may cause unwanted diffision).
Fig. 6. Pulsed Fiekl Gradient Muhi Spin Echo Sequence (PFG MSE) with 85 x pulses between gradients and TR=3s.
The diffusion-encoding gradient used was a three mis gradient (XL Resonance, London,
Ontario) with a 5 cm inner diameter and novel multilayer coi1 design to achieve high
gradient strengths (47). Eddy currents in this gradient were < 0.5% afler 100 ps, and the
gradient was unifonn over 4 cm. Diffision-encoding gradients were rotated in both the
x-y and x-z planes in increments of 15". At each orientation, the gradient strength was
varied fiom O to 1500 mT/m in 10 unifonn steps. The gradient pulse duration was 300 ps
with 65 n pulses between the diffision-encoding gradients for a total diffision time of
130 ms. This diffision time corresponds to a 23 pn fiee diffision length for water at
20°C with Dc2.02 x IO-' cm2/s (48). Following the last gradient pulse, 10 echoes were
obtained to measure signal strength. Data were averaged 80 times (TR=3s) to improve
signal-to-noise-ratio.
The echo attenuation for each angular gradient orientation was plotted against b to obtain
the initial dope (ADC), as described by the Stejskal-Tanner formula (16). ADC values
were plotted as r function of gradient angle in polar co-ordinates to display ADC
contours for the three sample orientations.
2.3 Theoretical Methods
The signai attenuation SIS, as a function of gradient strength in restricted cylindncal
geometry has been derived by S6deman et al. (21) by solving Fick's second law of
diffision with appropriate boundary conditions. The relationship is given by:
where ak., are given by the roots of the Bessel equation J',(a)=O (with the convention
that aio=O), q=ygW2x and Kn, are integer constants, where Knm=l if n=m=O, Knm=2 if
n#O and m=O, or m t O and n=O, K,,,=4 if n, mtO. The angle between the gradient
direction and the symmetry axis of the cylinder is given by 0. Actual expenmental values
were used for difhision time (A= 130 ms), and gradient strength (g=O to 1500 m T h ) and
duration (6=300 ps), with a difision coefficient of water ~ = 2 . 0 2 ~ 1 0 ~ ~ c r n ~ / s . The length
of the cylinder, L, and its radius, R, were set equal to 1 cm and 50 pm respectively, in
order to represent the previously described sarnple fibre geometry.
Theoretical results for echo attenuation as a fùnction of gradient strength were computed
using Matlab (Mathworks Inc., Natick, MA) according to Eq. [IS]. Summation was
perfonned to k, m=10 and n=1000, since higher order terms did not contribute
significantly to signal attenuation. Calculations were repeated for vatious gradient angles
and the ADC obtained f ~ r cach gradient orientation.
Calculations of diffision decay curves at high b values were obtained by convolving the
theoretical results for a single fibre with the fibre orientation probability distribution:
S(0') = S(û) P(û'-û)dû 161
where 0 represents the direction of diffision in three dimensions, S(0) is the signal decay
in any direction for a single cylinder, P(9) the probability distribution of fibre
orientations, and S(8') the resulting signal decay for the given fibre orientation
distribution. By using Eq. [16], signal decay curves for any set of known fibre
orientations can be obtained. For the case of the wound fibres in the plane of the winding
(y-z plane) and the randomiy onented fibres, the fibre orientation probability densities
were assumed to be equivalent to a star-shaped distribution of fibres (see Fig. 7). This
simplification was deemed appropriate since the mean fiee path of the spins is small
during the time of the experiment. The estimated diffision coefficient at different b
values for simulated fibre orientations was calcuiated by taking the dope of a straight line
joining b=O and the chosen b value on the diffusion decay cuves.
Fig. 7. Coiled Fibre Approximation
2.4 Results and Discussion
In Figure 8, the signal attenuation as a function of gradient strength is plotted for the X
@=O0), Y (e=90°) and Z directions in the aligned fibre bundle, where 0 is the gradient
rotation angle in the x-y plane. Diffision dong the fibres resembles fkee diffision of
water, ADC= 1.90k0.03~ 1 O-' cm2/s, whereas across the fibres, the ADC is decreased by
almost one half, ADC= 1.05k0.03~ 10" cm2/s. This is an anisotropy factor of 1.8 : 1,
which is comparable to that reported experimentaily in brain white rnatter (1.7-2.3 : 1 for
gradients parallel to and perpendicular to nerve fibres) (4 1). The lines represent the
theoretical calcuiations with no fiee parameters for diffision at 8=0° and 0=90° in a
cylinder, and agree well with experimental measurements. In ail cases, experimental and
theoretical calculations agree to within 10%.
Gradients along
I Theoretical calculations - 8=û0
- 1 1 0=90° 1
Fig. 8. Experimental and theoretical diffusion decay curves for gradients along the X, Y, Z directions in the aligned fibre bundle
A slight upward deviation with increasing b value is apparent in the experimental results
for the aligned fibres with the gradient oriented along X. This deviation h m theoreticai
calculations cm be explained by the small signal strength, which is no longer
distinguishable fiom the standard deviation of the noise. We know from theoretical
predictions that an upward curvature, or bi-exponentiality of the signal decay curve, as
has previously been observed in restricted diffûsion (22,23,42), would appear at much
higher b values than those obtained in our experiments.
Although not readily apparent in Fig. 8, there was a tendency for the first data point, the
measmd signal in the absence of diffusion encoding gradients @=O), to be slightly lower
than the signal attenuation following the application of the h t gradient increment. It is
possible that this discrepancy was caused by the presence of a slight gradient dong die
sample, which would cause a loss of signal in the absence of the diffusion-encoding
gradients. Calculations of the ADC, however, were unaffected whether or not this low
first data point was taken into account. As a result, al1 ADC calculations were perforrned
with the aberrant fint data point for consistency.
The ADC is plotted in Figure 9 as a function of angle between the diffusion gradient and
the axis of the fibres. Difision gradients were rotated in both the x-y and x-z planes for
al1 fibre samples, but only the x-y data are shown here. At 8=0° and 1 80°, diffision is
essentially the fiee diffision of water at 20°C for the aligned fibre bundle (Fig. 9a) and
was determined experimentally to be 1.90*0.03x 1 O-' cm2/s and 2.1 0k0.03~ 1 o - ~ cm2/s
respectively. However, as the gradient angle increases to 90' and 270°, the degree of
restriction is increased, which is reflected in the decreased ADC values,
ADC= 1 .OS&O.O~X 10" cm2/s and l.13k0.03~ 105 cm2/s respectively. The experimentally
obtained peanut-Iike ADC contour for the a l i p d fibre bundle is compared with
theoretical caîculations for a single cylinder (solid line, Fig. 9a). The calculation has the
same shape, with a somewhat higher degree of restriction across the fibres. The peanut-
like shape of the ADC contour for a single fibre can be expleinecl by the nature of the
ADC measurement. Whereas the difision ellipse represents the actual spin
displacements, the ADC contour is the projection of these spin displacements onto the
diffising gradient axis.
When the fibres were wound in a coil-like structure, the highly restricted dimension, the
diameter of the fibre, has now been rotated by 90° and lies dong the x-axis (Fig. 9b). The
ADCs measured across the diameter of the fibres dong X at 0=0° and 1 80°,
ADC= 1.06k0.03~ 1 O" cm2/s and 1.1 1 k0.03~ 10" cm2/s, are consistent w ith ADCs across
Fig. 9. Experimental results (dots) and theoretical calculations (solid line) of AOC vs gradient angle 9 for samples shown in Figure 5. A) aligned fibres 6) wound fibres C) randomly oriented fibres. All AOC contours are of the same scale with concentric circles 0 . 5 ~ 1 ~~crn* /s apart. Errors on AOCs t 0.03x105cm2is.
the fibres in the case of the aligned fibre bundle. As the gradient is rotated into the y-z
plane, the ADC increases but remains significantly less than fiee diffusion. The reduction
in the ADC in the y-z plane occurs due to the equal representation of al1 fibre directions
and is equal to the average of the free difision dong the fibres and the restricted
diffision across the fibres. Theoretical calculations for the wound fibres were obtained by
convolving the individually derived cylindrical diffision ADC contour with a circular
distribution of possible fibre orientations, as described in Eq. [16], and agree well with
experimental measurements.
Figure 9c depicts the ADC contour for a sample of randomly oriented fibres. The ADCs
are similar regardless of direction since spins will experience on average the same degree
of restriction in dl orientations. Whereas, previously in the wound sample, the
probability of fibre orientations was equally distributed in the y-z plane only, in the
randomly oriented sample, al1 fibre orientations are represented equally in three
dimensions. This is reflected in the theoretical ADC contour, which is spherical, and
results frorn an equal probability of fibre orientations in al1 directions. Some discrepancy
between the experimental results and theoretical calculations in the case of the randomly
oriented fibres can be attributed to the difficulty in assuring a truly random orientation for
scrambled plastic tubing (there is a slight tendency for fibres to track dong the wall of the
g las tube).
To explore the behaviour of the signal decay curves as a îùnction of gradient angle at
high b values, Fig. 10 is a plot of the theoretically calculated diffision decay for a single
fibre or cylinder, but calculated to a much higher b value (bmm=7.5~10S s/cm2). The
highest decay curve represents signal loss due to diffusion with the gradient onented
along the transverse axis of the fibre (8=90°). Successive decay curves are plotted in S0
increments until the gradient is oriented along the long axis of the fibre (8=0°).
Fig. 10. Theoretical diffusion decay curves to high b value (7.5x10~s/crn~) for a single fibre with gradient angles from O0 to 90° in 5" increments. The dashed line
indicates the estirnated Ob at b=4.2x10~s/cm~ for 8=90°.
The upward curvature at high b values in these calculations cannot be explained by the
presence of water in multiple compartments, since the theoretical calculations are based
on diffision inside an impermeable cylinder. It may, however, be the result of complete
spin dephasing near the centre of the cylinder at high gradient strengths so that the
measured signal attenuation is pnmarily due to spins close to the cylinder walls. As a
result, diffision will appear more restricted, manifested by an upward cwature in the
decay curve. This is further supported by the fact that the cwature is higher as the
gradient is aligned with the symmeûy axis of the cylinder (8=0°). Since the cylinder is a
closed structure and has lids at its top and bottom, unlike nerve fibres within a voxel, the
signal will appear to be the result of even more restricted diffusion at higher gradient
strengths, since it will originate primarily with spins close to the lids, as well as the
swrounding walls (28).
Fig. 1 la is a simulation of the change in shape of the estimated difision coefficient
contour for a single fibre as the b value is increased. The b values were chosen at
b=0.5xl0~ dcrn2, 4.2~10' s/cm2, 13.4~10' s/cm2. The dashed line on the signal decay
curve in Fig. 10 represents the estimation method for the difhsion coefficieiii. While this
method is a gross approximation, since the diffusion is clearly non-monoexponential, it is
the method rnost ofken reported in the literature for clinical difision measurements. At
low b values, this technique is equivalent to measuring the ADC, but as the b value
increases, it is apparent that the estimated diffision coefficient Db will be biased
depending on the b value. The decrease in the estimated diffusion coefficient at high b
might be misinterpreted as a higher level of restriction or a change in the structural
geometry within the voxel, but is simply reflective of the upward curvature of the signal
decay curve. As a result, as show in Fig. 1 1, the peanut-like shape becomes more
constrained dong the transverse axis of the fibre with increasing b and, at very high b,
becomes constrained dong its long axis as well.
This is fiuther illustrated in the case of the wound fibres in the plane of the winding and
for the randomly oriented fibres (Fig. 1 1 b). As b value is increased, the estimated
diffision coefficient Db for the randomly oriented fibres decreases due to the bend in the
decay cuves, and the contour becomes progressively smaller. However, it is not possible
to resolve the orientational distribution of the fibres.
Fig. 11. Estimated diffusion coefficient Ob vs gradient angle at b=0.5xi ~ ~ s l c r n ~ (outermost contour), 4 .2~1 ~ ~ s / c r n ~ , 13.4~1 ~ ~ s l c m * for A) single fibre 6) wound fibres in plane of winding and randomly oriented fibres C) two perpendicular fibres oriented along O0 and 90° D three fibres oriented along 0°, 45O, and 90°. 4 Concentric circles are 0.5x10~cm 1s apart.
In order to address the ability of high angular resolution diffision measurements coupled
with high b values to resolve fibre structure, Fig. 1 lc is a simulation of two identical
fibres oriented perpendicular to each other. n ie fibres are oriented along 0=0° and 0=90°.
At low b value (b=0.5xl0~slcm~), which is the approximate regirne for clinical ADC
measurements, the estimated difision coefficient is equal in al1 directions and provides
no information on fibre structure. As the b value is increased, however, the shape of the
resultant difision coefficient contour suggcsts the presence of two fibres, but the lobes,
which becorne more pronounced as b increases, are oriented at 4S0 to the fibre directions.
In the direction of the fibres, the estimated diffision coefficient decreases at high b
values due to the logarithmic relationship between signal decay and the diffision
coefficient. In the presence of multiple cornp&ments as seen in fibre crossings with
equivalent fibres, the estirnated diffision coeficient will be governed by the most
restricted cornpartment. In the case of two perpendicular fibres, the lowest diffision
coefficient appears dong ïhe gradient angle where the long axis of the first fibre is
parallel to the short axis of the second fibre. In these directions, the spins moving across
the highly restricted transverse axis of the second fibre dominate the signal decay,
because the rapidly decaying signal fiom the fieely diffising spins moving dong the long *
axis of the other fibre contribute negligibly. As a result, the measured diffision
coefficient contour is counter-intuitive and appears to have more 'restricted' diffision
dong the long axis of the fibres.
As the number of crossing fibres increases, the shape of the diffision coefficient contour
becomes more complex. In Figure I Id, the estimated diffision coefficient Db at three b
values is plotted as a function of gradient angle for three fibres oriented at 0°, 4S0 and 90"
respectively. As seen previously, the most restricted dimension of the three fibres appears
to have the highest estimated Db. The mon complex shape makes the detection of fibre
directionality much more difficult.
Current clinical b values are best represented by the lowest b value used in Our
simulations (b=0.5xl oSs/cm2). Our calculated diffision coefficient contours at these b
values do not demonstrate a high degree of variability with respect to the difision-
encoding gradient orientation. The upward cwvature in signal decay, however, has been
observed at much lower b values in white matter (22-27). Whereas our simulations do not
reflect non-monoexponentiality until b < 3 . 0 ~ 1 ~~s/cm*, in vivo studies of white matter
have shown this behaviour occurs at much lower gradient strengths b < 1.5xl0~s/cm~. We
would therefore expect difision coefficient contours measured in vivo to display much
greater variation with fibre orientation at lower b values than were observed in our
simulations at similar b values. The advent of stronger and faster clinical gradients will
also see the use of higher gradient strengths for diffision measurements, making the
detection of diffision coefficient contours with greater angular variability clinically
feasible.
Deteminhg the contours done, without a priori knowledge of fibre structure, is
insufficient to resolve multiple fibre orientations within a voxel. We have shown however
that knowledge of signal decay as a function of gradient angle for a single cylinder allows
us to detennine diffusion coefficient contours for any arrangement of fibres. A
comparable mode1 of diffision for white matter could be established by incorporating
multiple water compartments, permeable membranes, and possible exchange between
compartments. Similar models have already been suggested in the literanire (49-52).
Knowledge of the signal decay in such an environment would enable the determination of
diffusion coefficient contours for any probability of fibre orientations. These contours
could then be fit to experimental diffusion measurements at high angular resolution to
resolve complex fibre structure.
2.5 Conclusions
We have described a physical mode1 of restricted difision which demonstrates similar
anisotropy to that observed in nerve fibres. Our orientational diffision measurements
reflect the fibre geomeûy of the sample, and theoretical calculations for diffusion in
cylindrical geometry are in good agreement with the experimental results. Although the
magnitude of the measured ADCs are much larger than those reported in white matter,
the model's di ffisional anisotropy and structural similarity to nerve fibres may have
potential applications in the evaluation and cornparison of difision tensor imaging
sequences, as well as other methodologies for the study of restricted diffision.
Simulations of signal decay at high b values show diffision coefficient estimates which
assume monoexponentiality of the diffision decay curve to be very dependent on the b
value of the expenment. While the shape of the subsequent diaision coefficient contours
may provide information on the orientations of fibre bundles present within the voxel, the
apparent directionality of the contour alone does not indicate fibre orientation.
Our results suggest that high angular resolution diffision measurements provide
information on the presence of restriction and reveal dinusional anisotropy in fibre
bundles, which rnay not be detected using other analysis methodologies such as the
diffision ellipse. Further work is needed to fully characterize changes to the diffision
coefficient contour as the nurnber and orientation of fibres within a voxel change before
difision measurements at high angular resolution can become a clinically feasible
method for resolving fibre structure.
Chapter 3
Future Work
3.1 Introduction
In the previous chapter, experimental measurements and theoretical calculations were
used to examine the capability of difision measurements at multiple angles to resolve
complex fibre structure within a voxel. The problem of fibre crossings and branchings
must be addressed before fibre tractography and its applications to the detection of
diseases afl'ecting white matter can be successiùlly used in a clinical setting. A better
understanding of diffision in regions with low fibre directional coherence is essential for
continuous fibre tracking and understanding the organisation of complex neural networks
in the brain. The previous chapter discussed orientational diffision measurements at
multiple angles as a proposed solution to this problem. The experimental and theoretical
results were in good agreement, and m e r theoretical calculations demonstrated high
angular variability of the predicted diffision coefficient for simulated fibre orientations.
38
The experimentd model presented, however, is a simplification of the restricted and
anisotropic diffision in nerve fibre bundles, and may not accurately represent difision in
brain white matter. The theoretical calculations were based on diffusion in a simple
cylindtical, restricted geomeuy, which agreed well with our experimental model of
cylindrical fibres, but does not account for many of the more complex structural and
organisational characteristics of white matter.
In this chapter, possible implications for translating this study h m an in vitro model to
in vivo applications are discussed. The limitations of the current model in describing
diffision in highly structured tissues like brain white matter are described, as well as
suggested improvements to the model to better incorporate white matter tissue
characteristics. Finally, the feasibility of resolving difision and fibre orientation l using
diffision measurements at high angular resolution with an appropriate in vivo model are
discussed.
Orientational diffision measurements, however, are not the only proposed solution for
the stdy of molecular motion in reg ions of varied nerve fibre bundl e structure. Several
other techniques have also k e n suggested as alternatives to the diffision tensor
formalism. These methods will be discussed in the latter half of this chapter.
3.2 Orientational Diffusion Measurements in White Matter
In order to translate this study to an in vivo model, several additionai factors must be
incorporated into the model. These factors will be discussed below in detail. Previously
reported models of white matter and their limitations for this study will also be described.
3.2.1 A Model of White Matter
As mentioned in Chapter 1, brain white matter consists primarily of bundles of nerve
fibres, which facilitate communication between different regions of the brain, as well as
providing signalling to and fiom the rest of the body. The structure of individual nerve
fibres is cylindrical and their inner diameter ranges between 1-20 Fm. The experimental
and theoretical models presented in this thesis have the same approximate geornetry, and
demonstrate a similar degree of anisotropy to that reported in white matter. However, the
models assume diffision occurs in a single cornpartment, and is solely due to
contributions fiom spins inside the fibres. This is a simplification, since water is present
in al1 tissue compartments and the upward curvature of the signal decay curve has often
been attributed to different spin compartments (22-27).
Nevertheless, since the theoretical mode1 displays many of the characteristics associated
with diffision in white matter such as anisotropy and non-mononexponential signal
decay, the first step in translating the technique presented in this thesis to an in vivo
setting is performing a set of difision measurernents at high angular resolution in a white
matter tissue sample. These results can then be compared with theoretical calculations for
diffision in a cylinder to determine their validity in approximating white matter. Certain
discrepancies, which are intrinsically apparent, include the appearance of a rnuch delayed
upward curvanire in signal decay for the theoretical calculations. In addition, the top and
bottom wdls or lids of the theoretical cylinder create boundaries to the spins, which are
not present in white matter fibre geometry.
In order to address these issues, the theoretical model for diffusion in a cylinder could be
expanded to better represent white matter. Although difision in white matter is
undoubtedly anisotropic and restricted, controversy remains as to the origins of this
anisotropy and restriction. The axonal membrane may provide barriers to both
intraaxonal and extracellular diffision of water. In addition, the intnnsic structure within
the nerve fibre itself, consisting of neurofilarnents and microtubules associated with
axonal transport, may be contributing factors to the observed diffusion anisotropy. Some
studies are attempting to explain the behaviour of the signal decay curve as the result of
many water compartments with di fferent structurai characteristics (22-27).
Compartrnentalizing the signal, however, is difficult due to the presence of exchange
between the various compartments.
It is apparent from these previous snüiies that many factors must be incorporated into a
model of diffision in white matter. Several theoretical models of diffision in tissues have
been previously described and combine many of these elements (49-52). Szafer et al. (5 1)
presented a two-pool model for diffision in tissues with ceIl membrane permeability
between intria- and extracellular compartments. They showed that, considering the
typically low biological ceIl membrane pemeability, intra- and extracellular
compartments c m be regarded as independent and only weakly linked through membrane
pemeability. Their model, however, does not address the geometry specific to white
matter. Stanisz et al. (52) descnbed an analytical model of restricted diffision in bovine
optic nerve, incorporating both axons and spherical glial cells. They found that omitting a
third difising cornpartment consisting of sphekai glial support cells resulted in a poor
fit to their expenmental measurements of bovine optic nerve tissue sample. However,
their model did not take into account the possibility of multiple axonal orientations within
a voxel and the distribution of these restricted dimensions.
As demonstrated by the above models, an appropriate description of diffision in white
matter would have to consist of several components. Based on previous work, at least two
compartments, intraaxonai and extracellular, should be modelled, and contributions fiom
both must be accounted for in diffision signal decay. Axonal water signal contributions
can be represented by the theoretical model of cylindrically restricted diffusion used in
Chapter 2. Extracellular water in tissue models has been show to be dependent on the
degree of tortuosity encountered by the spins (51). In white matter, this tortuosity would
primarily depend on the distribution of nerve fibres, but rnay also be influenced by other
cells, such as glial cells, which provide obstacles to diffising spins. The degree of
tortuosity due to axons is dependent on direction in a similar manner to the anisotropy of
diffision of intraaxonal water.
In addition to the presence of two diffising water compartments, another factor which
must be taken into account is the presence of exchange between these compartments.
This is determined by the permeability of the axonal membrane. Permeability and
exchange rates are related to each 0 t h via the surface-to-volume ratio. The degree to
which the permeability will af5ect signal contributions, however, is very dependent on
experimental pammetea. For short diffision times, few water molecules will have time
to diffuse across the membrane, whereas longer diffision times may see complete mixing
of the compartments.
Another factor to be considered when examining diaision in multiple compartments is
the relative volume fractions of these compartments. Studies in the literature place the
extracellular volume fraction at roughly 0.2 and the intracellular volume fraction at 0.8.
However, these may not be easily distinguishable, depending on experimental
parameters.
Using these approximations to diffusion in white matter, the measured signal will be a
combination of water in two compartments:
sm(e, b) =X si@, b, p) +f, Se@, b, p) 1171
where Si, Se represent the signal fiom spins difising intraaxonally and extracellularly
respectively,J and5 the relative spin Fractions, and p the pemieability. Si may be
detemined fiom Eq. [15] for signal attenuation in a cylinder, with the 'lids' of the
cylinder moved to infinity, and with appropriate axon dimensions. In addition, the
intraaxonal signal will be afFected by exchange, spins leaving the axonal cornpartment
and spins moving into the axon from the extracellular space. Diffusion in the extracellular
space can be described by:
De@) = Df / ~ ( 0 ) ~ [18]
where h is the tortuosity, which is dependent on the gradient angle 0. The signal in the
extracellular space will also be govemed by exchange in a similar manner to the
intraaxonal space. The angular dependence of both the axond and extracellular signal
contributions will be govemed by the orientational distribution of fibre bundles, P(8).
The measuced signal Sm will therefore be a hction of P(B), Di, De, p,fi,fr, and b. Fig. 12
depicts the tissue mode1 with its pararneten.
Fig. 12. In vivo nerve fibre rnodet with intraaxonal diffusion coefficient Di, extracellular diffusion coefficient De, and permeability P.
This model of diffision in white matter fibre bundles is still grossly simplified, but may
incorporate some aspects of tissue diffision, which are neglected when diffision is
approximated by a single compartment inside a closed, impermeable cylinder, as
presented in this thesis. The addition of more parameters, however, increases the
complexity of the model and thus experimental parameters must be modified accordingly.
The presence of exchange assumes knowledge of the scale of this exchange, whether it
occurs rapidly (complete mixing of the compartments), slowly (no mixing of the
compartments) or at some time M e between those extremes. Several studies have been
aimed at associating diffision fractions and exchange regimes to T2 components in an
effort to isolate compartments according to their transverse relaxation parameter
(26,37,42,53). It is therefore important to be aware of the experimental parameters such
as diffision tirne, echo t h e , and gradient strength, to properly assess whether signal
contributions are originating predominantly from one compartment, two individual
compartments, or complete mixing of both spins groups (54). Assuming the presence of
only two compartments may be somewhat oversimplified (52), however the adequacy of
two compartments should be determined before unnecessarily complicating the model.
3.2.2 Determining fibre structure
Once the signal decay behaviour due to diffision in white rnatter has been modelled, and
is defined for any diffusion-encoding gradient orientation with any distribution of fibre
bundles, difision coefficient contours can be obtained for any fibre orientation. If the
single cylinder model used in Chapter 2 is sufficient to adequately describe in vivo white
matter diffision, the signal decay for any orientation of fibre bundles can be obtained
using Eq. [16]. If a more complex model of white rnatter is required, the theoretical signal
decay can be detemined according to Eq. [17], taking care to account for experimental
parameters which affect the permeabilities, exchange rates and relative volume fractions
of the diffising compartments.
Experimental measurements can then be performed in regions of known fibre
arrangement in the brain to assess the capability of the theoreticai model in replicating
signal decay due to diffusion. With an adequate model, the orientation of fibres can be
determined by finding a theoretical fit to the experimentally obtained diffision
coefficient contours, using the fibre orientation distribution as a fiee parameter.
There are limitations, however, to this technique. As demonstrated in the experiments and
simulations presented in this thesis, a completely random orientatiop of fibres is
indistinguishable fiom isotropie diffision, aside fiom an overall decrease in the measured
diffision coefficient. Using a relative decrease in the measured difision coefficient to
infer fibre structure rnay be difficult due to inter-subject vanability and the additional
dependence of the measured diffusion coefficient on b value. There is, therefore, a limit
to the number of fibre bundle crossings which can be resolved using this method. This
limit has not been determined. However for the detection of simple fibre crossings, or a
limited number of multi-directional bundles, diffusion measurements at high angular
resolution cm reveal a great deal of detail in the variability of the diffision coefficient
contours at high b values. Combined with a priori knowledge of diffision contours for
any orientation of fibres as determined by theoretical calculations, complex fibre bundle
distributions might be resolved.
Another limiting factor will be the difficulty in distinguishing crossing fibres from
kissing fibres (see Fig. 13).
Fig. 13. Crossing and kissing fibres
The ability for diffision measurements at multiple angles to resolve these similar fibre
structures is unclear since the diffision contours will likely appear very similar. Some a
priori anatomical information about fibre tract directions may be required since the
models of white matter suggested here do not take curvatwe of the fibres into account.
3.3 Other MR techniques for the measurement of diffusion
Although this thesis suggests that diffision measurements at multiple angles may be a
promising solution to the problem of diffision in multiple tibre bundles within a voxel,
several other techniques have been deveioped in an attempt to address the complexity of
difision in these regions. Three of these techniques and their possible limitations are
discussed below. Although, diffision spectrurn imaging shows the greatest promise in
resolving fibre structure, both the weighting of eigenvalues and the use of multiple
tenson are also discussed below since they provide insight into the presence of several
fiber tract orientations.
3.3.1 Weighting of Eigenvalues
Current fibre tractography makes use of the direction and magnitude of the largest
eigenvalue/eigenvector combination of the diffision ellipse to determine fibre orientation
within a voxel, and to track the fibre bundle in successive voxels. Wiegeil et al. (13)
suggested using the information in the second and third eigenvalues and vectors to further
characterize molecular motion in regions of higher fibre structure complexity. This
technique, however, does not provide a direct solution to the resolution of fibre
orientation. Instead, it serves simply as an indicator of the degree of anisotropy and
suggests the presence of multiple fibre bundles orientations without the capability to
resoive them.
3.3.2 Diffusion Spectrum Imaging
Difhion Spectnun Imaging is a terni coined by Tuch et al. (12) to describe a novel
technique for the examination of diffision in complex structurai regions. It is based on q-
space imaging, which was first described by Callaghan (1 7), and uses spin displacement
distributions to observe molecular difision. This technique stems fiom the origin of the
signal equation due to diffision. The signal attenuation Eq. 151, which was presented in
Chapter 1, can be rewritten as:
where P(zi( 22, A) is the displacement probability distribution, p(z) is the spin density, and
q=yg6/2n. Q-space is therefore simply an alternative representation of the same data. 1 t is
apparent fiom Eq. [19] that the signal attenuation is simply the Fourier transform of the
displacement probability distribution, where q is the reciprocal space vector:
By taking the inverse Fourier transform of the measured signal attenuation, the spin
displacement distribution can be obtained. In regions of complex fibre structure, such as
crossing fibres, the displacement distribution functions deviate strongly fkom a Gaussian
probability distribution and demonstrate highly irregular shapes, which reflect the
convoluted molecdar motion in these voxels. Whereas conventional MR difision
measwments are based on solving the integral in Eq. [5] assuming a Gaussian
probability distribution of spin displacements, q-space representation of dimision
examines the shapes of these displacement distributions, and therefore does not constrain
the movement of spins to a Gaussian probability distribution. Q-space diffision
measurements are often used to determine the restricted dimension of the diffusing
cornpartment since the full-width-at-ha1 f-maximum (F WHM) of a Gaussian represents
the mean displacement of the spins during the time of the experiment (55-59).
In the case of diffision in white matter, the displacement distributions obtained using
difision spectnun imaging have been show to deviate strongly from the 'free' Gaussian
distribution. The highly lobulated displacement distributions have been used to
characterize diffision in regions of complex fibre structure, and show great promise in
resolving fibre orientations within single voxels.
3.3.3 Multiple Diffusion Tensors
Another method that attempts to describe diffision in regions of complex tibre structure
involves the application of more than one diffision tcnsor (60,61). The rationale for this
technique stems fiom the observation that diffision signal decay in highly structured,
restrictive tissues, such as white matter, is non-rnonoexponential, and is often described
as bi-exponential. A bi-exponential fit to signal decay curves results in two diffision
coefficients to represent diffision along a certain axis within the voxel of interest.
Most often, the measwment of the diffision coefficient is the initial slope or ADC of the
decay curve, and bi-exponentiality is neglected. The presence of this bi-exponentiality,
however, has prompted some researchers into iworporating multiple diffision
coefficients into the description of tissue diffision. Specificaily, severai attempts have
been made to link the two decaying components along an axis with two contributing
water compartments (22-27). The application of two diffision tenson is attempting to
describe diffision decay behaviour based on spins in two compartments within a voxel.
The signal equation becomes:
SE, + (1 [21]
where Di and D2 represent the two diffision tenson, and f represents the weighting
fraction of the two diffising components.
This work was first applied to observe cardiac muscle fibre orientation, and preliminary
results used the anisotropy of the two tensors to compartmentalize the signal (62).
Although this technique seems promising in distinguishing difising water
compartments, it is dificult to separate compartments completely due to the movement
of water across membranes. As such, the use of multip!~ tensors will require fiuther
development and experimental verification before it may be applied to distinguishing
contributions from spins in noncolinear fibre bundles.
3.4 Conclusions
In vivo fibre tractography has widespread potential clinical applications. When combined
with fùnctional imaging for exarnple, it may provide insight into the complex neural
networks associated with higher order cerebral processes by linking functional regions of
the brain. Other highîy structured tissues, such as the myocardium, can also benefit fiom
tractography methods to detect changes in heart muscle structure. Difision MRI is
idealIy suited for these tasks due to its sensitivity to molecular motion in the micron
range. As a result, diffision MRI is an excellent probe of tissue structure and is able to
establish fibre connectivity. The inability to resotve a distribution of fibre tracts within a
voxel, however, is a severe limitation in current tractography using diffusion tensor MRI,
particularly in the brain which is known to have complex branching patterns.
In this thesis, orientational diffusion measurements were used to study fibre orientation.
Diffision measurements at high angular resolution in plastic fibre samples with different
fibre anangements were shown to be reflective of fibre structure and in good agreement
with theoreticai predictions for diffision in a restricted cylindrical geometry. Simulations
of diffision coeficient contours for fibre crossing patterns showed that the direction of
greatest diffision in these measurements does not correspond to fibre orientation.
However, the variability of the measured diffision coefficient with gradient angle does
provide detail about the presence of multiple fibres. Moreover, with the theoretical
calculation for diffision in a single fibre, the difision contours for any distribution of
fibre orientations may be obtained.
Knowledge of the contours for any arrangement of fibre bundles within a voxel would
allow the resolution of fibre orientation by finding a theoretical fit to diffision
measurements at multiple angles. The theoretical model presented in this paper must be
compared with in vivo fibre bundle measurements to determine its adequacy in
representing diffision in tissue. Although some modifications of the model to incorporate
certain white matter diffision characteristics may be required, once these have been
determined, high angular resolution diffusion rneasurements show great promise in
resolving the crossing of fibre tracts within a single voxel.
In vivo fibre tractography with diffision MR is a technique still in its infancy. As such,
many issues remain to be addressed before its establishment as a mutine clinical tool.
This thesis has approached only one of these issues. The plethora of applications for
tractography, however, from the study of white matter disease to the understanding of
basic neuronal connectivity, will ensure the rapid development of this technique as a
powerfùl research and diagnostic tool.
References
Ding SL, Elberger AJ. A modification of biotynilated dextran amine histochemistry
for labeling the developing marnmalian brain. J Neurosci Methods 1995;57:67-75.
Card JP. Explorhg brain circuitry with n e ~ o p i c viruses: new horizons in
neuroanatomy . Anat Rec l998;2S3: 1 76-1 85.
Petrides M, Pandya D. Dorsolateral prefiontal cortex: comparative cytoarchitectonic
anaîysis in the human and the macaque brain and corticocortical connections. Eur J
Neurosci 1999; 1 1 : 10 1 1 - 1036,
Le Bihan D, Delannoy J, Levin RL. Temperature mapping with MR imaging of
molecular diffision: application to hyperthermia. Radiology 1989; 1 7 1 :853-85 7.
Moseley ME, Cohen Y, Mintrorovitch J, Chileuitt L, Shimizu H, Kucharczyk J,
Wendland MF, Weinstein PR. Early detection of regional cerebral ischemia in cats:
cornparison of diffision- and T2-weighted MRI and spectroscopy. Magn Res Med
1 990; 14:3 30-346.
Larsson HBW, Thomsen C, Frederiksen J. In vivo magnetic resonance diffusion
measurements in the brain of patients with multiple sclerosis. Magn Res Im 1992; 10-
7-12.
Mori S, Crain BJ, Chacko VP, van Zijl PC. Three-dimensiond hgcking of axonal
projections in the brain by magnetic resonance imaging. Am Neurol 1999;45:265-
269.
Jones DK, Simmons A, Williams SCR, Horsfield MA. Non-invasive assessrnent of
axonal fiber connectivity in the human brdin via diffision tensor MN. Magn Res
Med 1 999;42:3 7-4 1 .
9. Conturo TE, Lori NF, Cul1 TS, Akbudak E, Snyder AZ, Shimony JS, McKinstry RC,
Burton H, Raichle ME. Tracking neuronal fiber pathways in the living human brain.
Proc Natl Acad Sci USA l999;96: 1 0422- t 0427.
10. Basser PJ, Pajevic S, Pierpaoli C, Duda J, Aidroubi A. In vivo fiber tractography
using DT-MM data. Magn Res Med 2000;44:625-632.
1 1. Wedeen VJ, Reese TG, Tuch DS, Weigel MR, Dou J-G, Weisskoff RM, Chessler D.
Mapping fiber orientation spectra in cerebrd white matter with fourier-transfonn
diffision MRI. In: Proceedings of the 8' Annual Meeting of ISMRM, Denver, 2000.
p 82.
12. Tuch DS, Weisskoff RM, Belliveau JW, Wedeen VJ. High angular resolution
diffision imaging of the human brain. In: Proceedings of the 8" Annuai Meeting of
ISMRM, Philadelphia, 1999. p 3 2 1.
13. Wiegell MR, Larsson HBW, Wedeen VJ. Fiber crossing in human brain depicted
with diffision tensor MR imaging. Radiology 2000;2 17(3):897-903.
14. Frank LR. Anisotropy in high angular resolution diffusion-weighted MN. Magn Res
Med 2001 ;45:935-939.
15. Hahn EL. Spin-echoes. Phys Rev l950;80:580-594.
16. Stejskal EO, Tanner JE. Spin diffision measurements: spin echoes in the presence of
a time-dependent field gradient. .i Chem Phys 1965;42(1):288-292.
1 7. Callaghan PT. Principles of nuclear magnetic resonance microscopy. Oxford,
England: Oxford University Press; 199 1.
18. Einstein A. Investigations on the theory of the Brownian movement. New York:
Dover; 1926.
19. Balinov B, Jtinsson B, Linse P, S6derma.n O. The NMR selfdiffision method applied
to restricted d i h i o n . Simulation of echo attenuation fiom molecules in spheres and
between planes. J Magn Reson A 1993; 104: 17-25.
20. Callaghan PT. Pulsed-gradient spin-echo NMR for planar, cylindrical, and spherical
pores under conditions of wall relaxation. J Magn Reson A 1995;113:53-59.
2 1. S6derman O, J6nsson B. Reseicted diffision in cylindrical geometry. J Magn Reson
A l995;ll7:94-97.
22. Niendorf T, Dijkhuizen RM, Noms DG, van Lookeren-Campagne M, Nicolay K.
Biexponential diffusion attenuation in various States of brain tissue: implications for
diffusion-weighted imaging. Magn Res Med 1996;36:847-857.
23. Assaf Y, Cohen Y. Non-mono-exponential attenuation of water and n-Acetyl
Aspartate signais due to diffision in brain tissue. J Magn Reson 1998; 13 1 :69-85.
24. Kraemer F, Darquie A, Clark CA, Le Bihan D. Separation of two diffision
compartrnents in the human brain. In: Proceedings of the 7' Annual Meeting of the
ISMRM, Philadelphia, 1999; p 1808.
25. Mulkem RV, Gudbjartsson H, Westin C-F, Zengingonul HP, Gartner W, Guttman
CRG, Robertson RL, Kyriakos W, Schwartz R, Holtzman D, Jolesz FA, Maier SE.
Multi-component apparent diffision coefficients in human brain. NMR Biomed
1999; l î :5 1-62,
26. Peled S, Cory DG, Raymond SA, Kirschner DA, Jolesz FA. Water difhsion, T2, and
compartmentation in fiog sciatic nerve. Magn Res Med 1 999;42:9 1 1-9 18.
27. Clark CA, Le Bihan D. Water d i h i o n compartmentation and anisotropy at high b
values in the human brain. Magn Res Med 2000;44:852-859.
28. Hurlimann MD, Helmer KG, de Swiet M, Sen PN, Sotak CH. Spin echoes in a
constant gradient and in the presence of simple restriction. J Magn Res A
1995; 1 1 3:260-264.
29. Latour LL, Mitra PP, Kleinberg RL, Sotak CH. Time-dependent difision coefficient
of fluids in porous media as u probe of surface-to-volume ratio. J Magn Res A
1993; 10 1 :342-346.
30. Moseley ME. Anisotropic solvent translational diffusion in solutions of poly (g-
benyl-L-glutamate). J Chem Phys I983;87: 18-20.
3 1. Callaghan PT, Sodennan O. Examination of the larnellar phase of aerosol OWwater
using pulsed field gradient nuclear magnetic resonance. J Chem Phys 1983;87: 1737.
32. Basser PJ, Mattiello J, Le Bihan D. MR difision tensor spectroscopy and imaging.
Biophys J 1994;66:259-267.
33. Pierpaoli C, Basser PJ. Toward a quantitative assessrnent of difhsion anisotropy.
Magn Res Med 1 996;36:893-906.
34. Kantarci K, Jack Jr CR, Xu YC, Campeau NG, O'Brien PC, Smith GE, Ivnik RJ,
Boeve BF, Kokmen E, Tangalos EG, Petersen RC. Mild cognitive impairment and
alzheimer disease: regional diffisivity of water. Radiology 200 1 ; 2 19: 10 1 - 107.
35. Krabbe K, Gideon P, Wang P, Hansen U, Thomsen C, Madsen F. MR diffusion
imaging of human intracraniai tumors. Nemrad 1 997;39:483-489.
36. Chen XJ, Hedlung LW, Moller HE, Chawla MS, Maronpot RR, Johnson GA.
Detection of emphysema in rat Iungs using magnetic resonance measurements of ' ~ e
d i h i o n . Proc Nat1 Acad Sci USA 2000;97(2 1 ): 1 1478-1 148 1.
37. Henkelman RM, Stanisz GJ, Kim JK, Bronskill MJ. Anisotropy of NMR properties of
tissues. Magn Res Med l994;32:592-60 1.
38. Cleveland GG, Chang DC, Hazlewood CF, Rorschach HE. Nuclear magnetic
resonance measurement of skeletal muscle. Anisotropy of the diffusion coefficient of
intracellular water. Biophys J 1976; 16: 1043- 1053.
39. Reese TG, Weisskoff RM, Smith RN, Rosen BR, Dinsmore RE, Wedeen VJ. lrnaging
myocardial fiber architecture in vivo with magnetic resonance. Magn Res Med
1 995;34:786-79 1.
40. Moseley ME, Kucharcyk J, Asgari HS, Norman D. Anisotropy of difhsion-
weighted MM. Magn Res Med 1 99 1 ; l9:32 1-326.
41. Hajnal IV, Doran M, Hall AS, Collins AG, Oatridge A, Pemock JM, Young IR,
Bydder GM. MR imaging of anisotropically restricted diffusion of water in the
nervous system: technical, anatomic and pathologie considerations. J Comput Assist
Tomogr 1991;15(1):1-18.
42. Stanisz GJ, Henkelman RM. Diffisional anisotropy of T2 components in bovine optic
nerve. Magn Res Med l998;40:405-4 10.
43. Mon S, Barker PB. Diffision magnetic resonance imaging: its principle and
applications. Anat Rec 1 999;257: 102- 109.
44. Pierpaoli C, Barnen AS, Pajevic S, Virta A, Basser PJ. Validation of DT-MFü
tractography in descending motor pathways of human subjects. In: Proceedings of the
9" Annual Meeting of ISMRM, Glasgow, 2001. p 501.
45. Wiegell MR, Reese T, Tuch DS, Sorensen AG, Wedeen VJ. Diffbsion spectnun
imaging of fiber white matter degeneration. In: Proceedings of the 9th Annual
Meeting of ISMRM, Glasgow, 2001. p 504.
46. Smtyr GE, Henkelman RM, Bronskill MJ. Variation in measured transverse
relaxation in tissue resulting fiorn spin focking with the CPMG sequence. S Magn
Reson 1988;79:28-44.
47. Chronik BA, Alejski A, Rutt BK. A 2000mTlm multilayer gradient coi1 for mouse
imaging. In: Proceeàîngs of the 8h Annual Meeting of ISMRM, Philadelphia, 1999. p
469.
48. Tofis PS, Lloyd D, Clark CA, Barker GJ, Parker GJM, McConville P, Baidock C,
Pope JM. Test liquids for quantitative MRI measurements of self-diffusion coefficient
in vivo. Magn Res Med 2000;43:368-374.
49. Karger J, Pfeifer H, Heink W. Principles and applications of self-diffusion
measurements by NMR. Adv Magn Reson 1988; 12: 1 -89.
50. Latour LL, Svoboda K, Mitra PP, Sotak CH. Time-dependent diffision of water in a
biological model system. Proc Nat1 Acad Sci USA 1 994;9 1 : 122% 1233.
5 1. Szafer A, Zhong J, Gore JC. Theoretical model for water diffision in tissues. Magn
Res Med 1995;33:697-712.
52. Stanisz GJ, Szafer A, Wright GA, Henkelman RM. An analytical mode1 of restricted
diffision in bovine optic nerve. Magn Res Med 1997;3 7: 103- 1 1 1.
53. Does MD, Snyder R. Multiexponential T2 relaxation in degenerating peripheral
nerve. Magn Res Med 1 996;35:207-2 1 3.
54. Norris DG. The effects of microscopie tissue parameters on the diffision weighted
magnetic resonance imaging experiment. NMR Biomed 200 1 ; 14:77-93.
55. Cory DG, G m w a y AN. Measurement of translational displacement probabilities by
NMR: an indicator of compartmentation. Magn Res Med 1990; 14:435-444.
56. Callaghan PT, Macgowan D, Packer KJ, Zelaya FD. High resolution q-space imaging
in porous structures. J Magn Res 1990;90: 1 77- 182.
57. Callaghan PT, Coy A. PGSE NMR and molecular translational motion in porous
systems. In: Nuclear magnetic resonance probes of molecular dynamics. Amsterdam:
Kluwer Acadernic Publishers; 1994; p 489-523.
58. King MD, Houseman J, Roussel SA, van Bruggen N, Williams SR, Gadian DG. Q-
space imaging of the brain. Magn Res Med l994;32:707-7 13.
59. Assaf Y, Cohen Y. Structural idonnation in neuronal tissue as revealed by q-space
diffision NMR spectroscopy of metabolites in bovine optic nerve. NMR Biomed
1999; 1 S:335-344.
60. Inglis BA, Bossart EL, Buckley DL, Wirth 111 ED, Mareci TH. Visualization of
neural tissue water compartments using biexponential diffision tensor MRI. Magn
Res Med 2001;45:580-587.
61. Alexander AL, Hasan KM, L a m M, Tsuruda JS, Parker DL. Analysis of partial
volume effects in diffusion-tensor MRI. Magn Res Med 200 1 ;45:770-780.
62. Hsu EW, Buckley DL, Bui JD, Blackband SJ, Forder JR. Two-component diffision
tensor M N of isolated perfùsed hearts. Magn Res Med 200 1 ;45 : 1039- 1045.