ADVANCED DIFFUSION MRI TECHNIQUES: METHODOLOGICAL

DEVELOPMENT AND CLINICAL APPLICATION

by

Adam Scott Bernstein

W
Copyright © Adam Scott Bernstein 2019
IE
A Dissertation Submitted to the Faculty of the
EV
DEPARTMENT OF BIOMEDICAL ENGINEERING

In Partial Fulfillment of the Requirements

PR

For the Degree of

DOCTOR OF PHILOSOPHY

In the Graduate College

THE UNIVERSITY OF ARIZONA

2019




ProQuest Number: 13865295




All rights reserved

INFORMATION TO ALL USERS
The quality of this reproduction is dependent upon the quality of the copy submitted.

In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.



W

IE


EV
ProQuest 13865295

Published by ProQuest LLC (2019 ). Copyright of the Dissertation is held by the Author.


All rights reserved.
PR

This work is protected against unauthorized copying under Title 17, United States Code
Microform Edition © ProQuest LLC.


ProQuest LLC.
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106 - 1346
 2

THE UNIVERSITY OF ARIZONA

GRADUATE COLLEGE

As members of the Dissertation Committee, we certify that we have read the dissertation
prepared by Adam Scott Bernstein, titled Advanced Diffusion MRI Techniques:
Methodological Development and Clinical Application and recommend that it be
accepted as fulfilling the dissertation require­ment for the Degree of Doctor of
Philosophy.
Date: 8 February 2019
'
Theodore P Trouard

W
Date: 8 February 2019
Peter J Bassei
IE
Date: 8 February 2019
EV
Nan-kuei Chen

Date: 8 February 2019

Maria M Altbach
PR

Date: 8 February 2019

�----
Final approval and acceptance of this dissertation is contingent upon the candidate's sub­
mission of the final copies of the dissertation to the Graduate College.
I hereby certify that I have read this dissertation prepared under my direction and recom­
mend that it be accepted as fulfilling the dissertation requirement.

,�
#� \ •• -

Disserta irect
\ 1, -- '-._(
2
Theodore P'rrouard
Date: 8 February 2019
�
B Date: 8 February 2019
ation Directo : Peter
� J Basser
 3

STATEMENT BY AUTHOR

This dissertation has been submitted in partial fulfillment of the requirements for an ad-
vanced degree at the University of Arizona and is deposited in the University Library to

W
be made available to borrowers under rules of the Library.

Brief quotations from this dissertation are allowable without special permission, provided
that accurate acknowledgment of the source is made. Requests for permission for ex-
IE
tended quotation from or reproduction of this manuscript in whole or in part may be
granted by the copyright holder.
EV
PR

SIGNED: Adam Scott Bernstein

 4

ACKNOWLEDGEMENTS

I would first and foremost like to thank my advisors, Drs. Ted Trouard and Peter Basser.
Dr. Trouard welcomed me into his lab as an undergraduate, and inspired me to remain
curious. Dr. Basser recruited me to spend part of my PhD training with him at the
National Institutes of Health. Dr. Basser took great interest in my personal, professional,
and academic growth, and has been a great source of support and encouragement for me.
I am also grateful to have worked under the mentorship of Dr. Alexandru Avram. Dr.
Avram’s technical proficiency and knowledge of MRI, along with his abounding knowl-
edge of UFC fighting and obscure foreign film made for very insightful and enlightening

W
discussions.
Drs. Nan-kuei Chen and Ying-Hui Chou could not have come to Arizona at a better
time, and have been the embodiment of collaboration and inclusion. I am very grateful to
have worked (and to continue to work) with both Drs. Chen and Chou on both technical
IE
development and clinical applications of advanced imaging techniques.
I would like to thank the clinical collaborators involved in many of my projects, in-
cluding Dr. Duke Duncan, Dr. Craig Weinkauf, Dr. Gloria Guzman, Dr. Wei Zhou, and
EV
Dr. John Butman, for all of their clinical insights, and for providing the inspiration behind
many of the projects I was involved in.
I would like to thank Chidi Ugonna for his extensive technical support and enthusiasm.
I am grateful that a true computer expert was able to step in and clean up the many IT
issues that I created.
PR

Lastly, I would like to thank all of my friends and family for their ongoing support all
through this process. Their have been many obstacles, and unforeseen roadblocks, and
they were always there to comfort and support me. A special thanks to Hannah for her
patience, support and love.
 5

DEDICATION

To Mom, Dad, and Hannah

W
IE
EV
PR
 6

TABLE OF CONTENTS

LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

CHAPTER 1 Introduction to Magnetic Resonance Imaging . . . . . . . . . . . . 19

W
1.1 The Origin of the Magnetic Resonance Signal . . . . . . . . . . . . . . . 20
1.2 The Bloch Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.3 The Hahn Spin-Echo Experiment . . . . . . . . . . . . . . . . . . . . . . 25
IE
1.4 Diffusion Nuclear Magnetic Resonance . . . . . .
1.5 Magnetic Resonance Imaging . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
29
38
1.6 Diffusion Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
EV
CHAPTER 2 Developments in Diffusion Data Processing and Analysis . . . . . 65
2.1 The Diffusion MRI Protocol . . . . . . . . . . . . . . . . . . . . . . . . 65
2.2 Diffusion MRI Artifacts and Their Correction . . . . . . . . . . . . . . . 67
2.3 Development of an Artifact Correction Pipeline . . . . . . . . . . . . . . 76
PR

CHAPTER 3 Carotid endarcterectomy leads to structural and network changes in

carotid vascular disease using diffusion MRI . . . . . . . . . . . . . . . . . . . 79
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

CHAPTER 4 Diffusion MRI in Cerebral Palsy . . . . . . . . . . . . . . . . . . . 96

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

CHAPTER 5 Conclusions and Future Directions . . . . . . . . . . . . . . . . . . 111

5.1 Summary and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.2 Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
 7

TABLE OF CONTENTS – Continued

APPENDIX A Bootstrap analysis of diffusion MRI parameters derived from multi-

band diffusion MRI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
A.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
A.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
A.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
A.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
A.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

APPENDIX B A population template of MAP MRI-derived diffusion propagators

and microstructural parameters in healthy subjects . . . . . . . . . . . . . . . . 157
B.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

W
B.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
B.3 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
B.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
IE
B.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
B.6 Appendix B.A The MAP MRI basis functions . . . . . . . . . . . . . . . 174
B.7 Appendix B.B Jensen-Shannon divergence for MAP propagators . . . . . 175
EV
APPENDIX C A diffusion-matched principal component analysis (DM-PCA) based
two-channel denoising procedure for high-resolution diffusion-weighted MRI . 178
C.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
C.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
PR

C.3 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

C.4 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
C.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
C.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
C.7 Ackowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196

APPENDIX D Python Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199

D.1 Selecting Diffusion Directions . . . . . . . . . . . . . . . . . . . . . . . 199
D.2 Preprocessing Pipeline . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
D.3 Diffusion MRI Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
 8

LIST OF FIGURES

1.1 Simplified representation of a hydrogen nucleus as a spinning ball of pos-

itive charge. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.2 Graphical representation of the net magnetization, M ~ , being excited by a
circularly polarized RF excitation pulse. At rest, M ~ is parallel to B0 (A).
Immediately after the application of a small magnetic field, B ~ 1 initially

W
along the x-axis, M ~ begins to nutate about the x-axis (B). If B ~ 1 is ap-
plied at exactly the Larmor frequency of the spins making up M ~,M ~ will
~
continue to nutate about the rotating B1 field until it is turned off. . . . . . 23
1.3 Graphical representation of the net magnetization, M
IE ~ , being excited and
nutating down into the x-y plane, represented in the rotating frame of
reference. In the rotating frame of reference, B~ 1 remains aligned with the
0
x axis, and M~ will lay along the y axis. . . . . . . . . . . . . . . . . . .
0
24
EV
1.4 Graphical representation of signal generation and reception in a typical
MRI experiment. In panel A, the net magnetization, M ~ , precesses in the
x-y plane after excitation, and generates a voltage (C) in a receiver coil (B). 24
1.5 Dephasing of spins due to magnetic field inhomogeneities shown in the
rotating frame of reference. The net transverse magnetization is depicted
PR

as a red arrow. The net magnetization, M ~ is depicted as a red arrow.

Immediately following excitation, all of the spins are coherent and M ~ xy is
at its maximal value (A). In the presence of a heterogenous external field,
the spins accumulate differentially and M ~ xy shrinks (B). . . . . . . . . . 26
1.6 Spin Echo pulse sequence diagram. Radio frequency (RF) pulses are
shown along the top row, and the free induction decay (FID) is shown
on the bottom. The FID decays as a function of T2∗ immediately follow-
ing the 90° excitation pulse. At a time TE following the 90° excitation
pulse, the spins with now reversed phase following the 180° pulse come
together again and form an echo. . . . . . . . . . . . . . . . . . . . . . . 27
1.7 Evolution of the phase of several representative spins (dashed blue lines)
during a 90-180 spin echo experiment. After time TE/2, spins have ac-
crued a phase and M ~ xy has decayed (A). Immediately following the ap-
plication of the 180° pulse, the phase of the spins is reversed, and they
begin to converge (B). At time TE, all of the phase accrued due to field
inhomogeneity is reversed and an echo is formed (C). . . . . . . . . . . . 27
 9

1.8 Carr-Purcell-Meiboom-Gill spin-echo experiment. Multiple 180° pulse

are applied with a 90° phase shift relative to the excitation RF pulse, and
multiple spin echoes are formed. The amplitude of the echoes is shown
to decay as a function of T2 . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.9 Creating a linearly varying magnetic field using a magnetic gradient coil. 32
1.10 Distribution of displacements from a monte-carlo diffusion simulation.
The shape of the distribution is nearly gaussian. . . . . . . . . . . . . . . 34
1.11 Expected signal modulation factor due to diffusion (left), and the expected
root mean square phase evolution (right) for a constant background gra-
dient multi spin echo experiment. . . . . . . . . . . . . . . . . . . . . . . 35
1.12 Comparison of the signal decay for multiple echo formations (top) and

W
a single echo formation (bottom) over the same total experimental time.
The signal from the single echo experiment is significantly decreased due
to diffusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IE 35
1.13 Pulse sequence diagram of the PGSE experiment. . . . . . . . . . . . . . 37
1.14 The expected signal modulation due to diffusion (top), expected root
mean square of the phase evolution (middle), and the pulsed diffusion
EV
encoding wave forms (bottom) for a standard PGSE NMR experiment. . . 38
1.15 Idealized gradient waveforms (left), and the k-space trajectory (right) for
a standard EPI readout. . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
1.16 A full diffusion weighted imaging pulse sequence. The first section de-
picts a slice-selective excitation. Next, the excited magnetization is dif-
PR

fusion weighted using the PGSE scheme. Finally, the signal is spatially
encoded, and an image acquired using the EPI readout. . . . . . . . . . . 45
1.17 A simple in vivo example of fitting the diffusion coefficient, D, across an
entire image with several b-values collected. . . . . . . . . . . . . . . . . 46
1.18 Graphical representation of a diffusion tensor, with primary axes λ1 , λ2 ,
and λ3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
1.19 Diffusion tensor shape as a function of increasing FA (left-right) and in-
creasing diffusivity (top-bottom). . . . . . . . . . . . . . . . . . . . . . . 49
1.20 DTI derived parameter maps. The directionally encoded color map (DEC)
(E) is red when water diffuses primarily along the x-axis, green when
water diffuses primarly along the y-axis, and blue primarily along the z-
axis, as indicated by the legend on the bottomr right. . . . . . . . . . . . 50
1.21 MAP basis functions in 2-dimensions in diffusion space. Red indicates
positive values, and blue indicates negative values. . . . . . . . . . . . . 54
1.22 Example parameter maps derived from MAP MRI. . . . . . . . . . . . . 56
1.23 Diffusion orientation distribution functions for each of the MAP-MRI ba-
sis functions up to order 4. . . . . . . . . . . . . . . . . . . . . . . . . . 57
 10

1.24 Fiber orientation distribution functions fit for every voxel in a brain. . . . 59
1.25 Spherical harmonic basis up to order four. The sphere is modulated by
the spherical harmonic amplitude for every θ and φ. Positive values are
shown in red, and negative values in blue. . . . . . . . . . . . . . . . . . 59
1.26 Apparent fiber density calculated for every lobe of the fODFs computed
previously. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

2.1 The effects of eddy currents are shown for a single line of data (red line in
image A) in B. As can be seen the edge of the brain appears to shift by 1-2
voxels for each diffusion image collected. After eddy current correction,
the edge of the brain is stable throughout the entire set of diffusion images

W
(C), providing the required alignment for further processing. . . . . . . . 71
2.2 The effects of EPI distortion are shown in A and B above, where the phase
encoding direction has been reversed between the two images. In A, the
IE
PE direction is anterior to posterior, and leads to a compression of the
brain near the frontal sinuses. When the PE direction is reversed (B), the
brain is streched near the frontal sinuses. In C, the distortions have been
corrected by finding the midway point of images A and B. In panels D
EV
and E, the same brains from A and B are overlayed on an undistorted T1 -
weighted image. The white arrows point out regions where the brain is
obviously distorted relative to the anatomical reference, and in panel F,
the distortions are shown to be largely removed, leading to much better
PR

alignment of the EPI images with the structural image. . . . . . . . . . . 73

2.3 Local principal component analysis greatly increases the SNR of a
dataset, and the effects are particularly noticeable at high b-values. The
scales in the above images are not equal, in order improve visibility of the
underlying signal at high b-values. . . . . . . . . . . . . . . . . . . . . . 74

3.1 Steps involved in creating a structural connectome. A T1 -weighted image

(top left) is first parcellated into well-defined anatomical regions using
FreeSurfer (top right). Diffusion images (bottom left) are processed using
CSD in order to generate a whole brain tractogram (bottom right). The
number of fibers that begin in one defined region and end in another are
used to weight the edges between each node in the connectome (middle).
This connectome is then converted into a connectivity matrix for network
analysis (right). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.2 Results of the Montreal cognitive assessment prior to surgery and on 4-6
month follow-up. A paired t-test demonstrates a significant increase in
cognitive function following endarcterectomy. . . . . . . . . . . . . . . . 87
 11

3.3 Fixel-based analysis average pair-wise differences of pre and post-surgery

AFD. The top row contains five evenly spaced axial slices of apparent
fiber density differences overlayed on the study template. The bottom row
contians five evenly spaced coronal slices over the part of the brain where
differences were the greatest. Regions that are red-yellow are regions
where the AFD increased on average (p < 0.1), and the blue regions are
regions where the AFD decreased on average following surgery (p < 0.1). 88
3.4 Voxel-wise average paired difference maps of several DTI and MAP-MRI
parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.5 Diffusion parameter derived from an ROI analysis using the ICBM DTI
Atlas. The red box and whisker plots summarize the distribution of mean

W
diffusion parameters within each of the ROIs defined in the ICBM DTI
atlas when measured prior to carotid endarcterectomy. The blue box and
whisker plots show the same information for the parameters measured 4-6
IE
months following the procedure. . . . . . . . . . . . . . . . . . . . . . . 94
3.6 Connectome edges that demonstrated statistically significant increases in
connectivity prior to multiple comparisons correction are highlighted in
EV
the images above. The left putamen (top row, red arrow) showed signif-
icant increases in structural connectivity with several regions. The right
superior temporal gyrus (bottoms row, red arrow) also demonstrated in-
creased connectivity with several regions nearby. . . . . . . . . . . . . . 95
PR

4.1 Crossover study design where group one receives the standard of care
therapy for 36 weeks followed by 12 weeks of intense therapy. Group
two receives 12 weeks of intense therapy first, followed by the standard
of care therapy. Imaging is performed at zero and 36 weeks, before group
one has undergone intense therapy. . . . . . . . . . . . . . . . . . . . . . 99
4.2 Schematic of the generation of along-tract DTI parameter profiles. First,
ROIs are drawn on the cerebral peduncle (top of step 1) and the internal
capsule (bottom of step 2). Then deterministic CSD-based tractography is
performed to identify the CSTs (2). Then maps of FA and RD are sampled
at 100 evenly spaced points along the length of the generated tractograms
(3). Finally the FA and RD are averaged across the width of the CST at
every sampled point to generate an along track profile (4). . . . . . . . . . 101
4.3 Line plots showing the change in GMFM scores at baseline and 36 week
followup. Subjects 3 and 5 are in group one, and subjects 1, 2, and 4 are
in group two. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
 12

4.4 Diagnostic T2 FLAIR images (top) and anatomical T1 images (bottom)

for each of the five subjects enrolled in the study. Displayed slices for
each subject were selected based on the location of white matter hyperin-
tensities as well as to highlight major anatomical changes if applicable. . 102
4.5 Corticospinal tracts defined for each subject using deterministic tractog-
raphy on the pre (left) and post (right) therapy dMRI data. . . . . . . . . 107
4.6 Scatter plots of GMFM vs. the average FA (left) and RD(right) of the
entire affected (blue) and unaffected (red) CST. The Pearson correlation
coefficient, R, is calculated for the average parameter on each side of the
brain. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.7 Along track FA measurements for all 5 enrolled subjects. Along track FA

W
prior to treatment (solid line) is shown in blue (dashed line is standard
deviation), and the along track FA following 36 weeks of treatment is
shown in red. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IE 109
4.8 Along track RD measurements for all 5 enrolled subjects. Along track RD
prior to treatment (solid line) is shown in blue (dashed line is standard
deviation), and the along track RD following 36 weeks of treatment is
EV
shown in red. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

A.1 Visualization of the average parameter maps derived from DTI and MAP-
MRI for three different collection schemes. The top row are the maps de-
rived from the minimum TR (i.e. 3.7s) 3x MB sequence, followed by the
PR

TR=10.7s 3x MB sequence, and finally the parameter maps derived from

the standard single band sequence with minimum TR (10.7s) for full brain
coverage. The first and second column show fractional anisotropy (FA)
and mean diffusivity (MD, mm/s2 ) derived from DTI analysis. The third
and fourth column show propagator anisotropy (PA) and return to origin
probability maps (RTOP1/3 , mm−1 ) derived from MAP-MRI analysis. . . 149
A.2 Scatter plots of MB derived parameter values plotted against SB derived
parameter values on a voxel by voxel basis for the entire brain. Voxels
are categorized as either gray matter (red), white matter (blue), or CSF
(green). The black, dashed line has a slope of 1.0 and is plotted for refer-
ence. The top row shows parameters derived from the MB sequence using
a minimal TR (3.7s) plotted against the same parameters derived from the
SB sequence. The bottom row shows parameters derived from the MB
sequence using a matched TR (10.7s) plotted against the same parame-
ters derived from the SB sequence. The range for both the x- and y-axis
are 0-1 for FA; 0-0.003 mm2 /s for MD; 0-1 for PA, and 0-100 mm− 1 for
RTOP1/3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
 13

A.3 Maps of the coefficient of variation of the parameter maps derived from
both DTI and MAP for all 3 collection schemes. The top row are the
maps derived from the minimum TR SMS sequence, followed by the 10
second TR SMS sequence, and finally the parameter maps derived from
the standard single band sequence with minimum TR for full brain coverage.152
A.4 Difference maps between the CVs of parameters derived from MB and SB
acquisitions. The top row shows the CV of the short TR MB (TR=3.7s)
derived parameter values subtracted from the CV of the SB derived pa-
rameter values. The bottom row shows the CV of the TR-matched MB
(TR=10.7s) derived parameter values subtracted from the CV of the SB
derived parameter values. All CV maps are plotted on the same scale,

W
where red and yellow indicate that the MB CV is lower than the SB CV,
and blue indicates that the MB CV is larger than the SB CV. . . . . . . . 153
IE
B.1 A flowchart for spatial standardization of MAP MRI data and for con-
structing a population template of MAP MRI 3D diffusion propagators. . 160
B.2 Microstructural MAP MRI parameters computed from the template of 3D
diffusion propagators measured in a healthy population: DTI-parameters:
EV
mean diffusivity (MD), fractional anisotropy (FA), and direction encoded
color (DEC) map; along with MAP MRI microstructural parameters: the
Return-to-origin probability (RTOP), Propagator Anisotropy (PA), and
Non-Gaussianity (NG). . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
PR

B.3 Visualization of orientation distribution functions (ODFs) derived from

the population template of 3D diffusion propagators. . . . . . . . . . . . 168
B.4 Voxel-wise absolute angular difference measure between individual sub-
ject and template MAP MRI data. A. Multiple slices across the entire
brain in a representative subject. B. A representative slice from 5 subject. 169
B.5 Correlation plots showing the agreement of several MAP-MRI derived
parameters (RTOP left, PA middle, NG right) quantified using ROI-
analyses in subject and template space. In subject space, propagators
are measured using a unique set of MAP basis functions computed from
the subjects DTI data. In template space, propagators for all subjects
are measured using the same set of MAP basis functions determined by
the DTI population template. In each plot, the x-axis corresponds to the
ROI-average value of a MAP-MRI parameter in subject space, while the
y-axis corresponds to the ROI-average value of a MAP-MRI parameter in
template space. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
 14

B.6 Voxel-wise absolute angular difference measure between MAP MRI

propagators acquired longitudinally in the same subject undergoing vas-
cular occlusion surgery at three different clinical time points: Visit 1 -
baseline. Visit 2 - 1 month after surgery. Visit 3 - 6 months after surgery . 171

C.1 The schematic diagram of the new complex-domain DM-PCA based two-
channel denoising procedure. . . . . . . . . . . . . . . . . . . . . . . . . 183
C.2 Comparison of DWI denoising through filtering signals across nearest
neighboring voxels and diffusion-matched voxels: The red dot in (a)
shows a target voxel (displayed on top of mean DWI map), whose sig-
nals in 6-direction DWI scans are to be denoised. In many existing de-

W
noising methods, signals of nearest neighboring voxels in a patch (see b)
are the input of a filtering procedure. In contrast, we identify a group
of voxels that demonstrate very similar signal variation patterns along
IE
the diffusion dimension but are not necessarily neighboring (see c) for
subsequent filtering procedures. Panels d, e and f show an input image,
nearest-neighboring PCA produced image, and DM-PCA produced im-
age, respectively. Residual maps obtained with nearest-neighboring PCA
EV
and DM-PCA methods are shown in panels g and h, respectively. . . . . . 184
C.3 A simulation study for comparing magnitude-domain DM-PCA and two-
channel complex-domain DM-PCA in terms of the accuracy in ADC
fitting: (a) Noise-free DWI data corresponding to b = 0, 200, 400 ...
PR

2200 (s/mm2). (b) DWI data affected by Rician noise. (c) DWI data
denoised by magnitude-domain DM-PCA. (d) DWI data denoised by a
two-channel complex-domain DM-PCA procedure. (e) Signal intensities
of noisy DWI data (solid curve in orange) and the ground truth (dashed
curve in blue). (f) Signal intensities of magnitude-domain DM-PCA pro-
duced data (solid curve in orange) and the ground truth (dashed curve in
blue). (g) Signal intensities of complex-domain DM-PCA produced data
(solid curve in orange) and the ground truth (dashed curve in blue). (h)
Errors in ADC fitting for data with different SNR levels. . . . . . . . . . 188
 15

C.4 A simulation study for comparing magnitude-domain DM-PCA and two-

channel complex-domain DM-PCA in terms of the accuracy in fitting
ADC values from parallel DWI data: (a) Images reconstructed with the
2xSENSE algorithm from noise-free under-sampled k-space data cor-
responding to b = 0, 200, 400 ... 2200 (s/mm2). (b) Images recon-
structed with the 2xSENSE algorithm from noisy under-sampled k-space.
(c) SENSE-produced data denoised by magnitude-domain DM-PCA. (d)
SENSE-produced DWI data denoised by a two-channel complex-domain
DM-PCA procedure. (e) Signal intensities of noisy parallel DWI data
(solid curve in orange) and the ground truth (dashed curve in blue). (f)
Signal intensities of magnitude-domain DM-PCA produced data (solid

W
curve in orange) and the ground truth (dashed curve in blue). (g) Signal
intensities of complex-domain DM-PCA produced data (solid curve in
orange) and the ground truth (dashed curve in blue). (h) Errors in ADC
IE
fitting for data with different SNR levels. . . . . . . . . . . . . . . . . . . 190
C.5 FA maps obtained from high-resolution DWI images (0.85 mm3 voxel
size), before and after DM-PCA based denoising, corresponding to dif-
EV
ferent SNR levels in input data. . . . . . . . . . . . . . . . . . . . . . . . 194
C.6 (a) and (b) show coronal-plane mean DWI and FA maps, respectively, de-
rived from high-resolution data after DM-PCA based denoising, with ar-
rows indicating the left hippocampus. The corresponding zoom-in images
shown in (c) and (d), respectively. The FA map derived from images with-
PR

out DM-PCA denoising is shown in (e). The coarse hippocampal struc-

tures revealed by the mean DWI map are highlighted in (f). Anatomic
structures that can be identified from color-coded FA map are shown in
(g) and (h). Region 1 in (g) corresponds to the dentate gyrus; Region
2 shows fibers that connect hippocampus and entorhinal cortex to other
brain areas; Region 3 contains hippocampal CA1, CA2, and CA3. . . . . 197
C.7 Application of DM-PCA denoising to human brain DWI data at conven-
tional resolution (1.8 mm3 voxel size): Panels a and b compare one of
the DWI images before and after DM-PCA denoising, respectively, for
4 of the participants. Panels c and d show the corresponding FA maps
obtained from data before and after DM-PCA denoising, respectively. . . 198
 16

LIST OF TABLES

3.1 Summary of subjects enrolled in imaging and neurocognitive testing be-

fore and after carotid endarcterectomy. . . . . . . . . . . . . . . . . . . . 81

B.1 The average standard deviation of PA (top row), NG (middle row) and
RTOP (bottom row) across all 50 WM ROIs and all 5 subjects in subject
space (left) and template space (middle). The difference is shown in the

W
last column. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168

IE
EV
PR
 17

ABSTRACT

Diffusion magnetic resonance imaging (dMRI) has become a mainstay for better under-
standing the structure of the normal human brain, as well as for understanding the changes
that occur in disease. Using the motion of water as a probe for the microscopic environ-
ment of tissues, dMRI provides insight into the structure and organization of the brain.

W
Diffusion tensor imaging (DTI), a dMRI provided the neuroimaging community with a
means of further characterizing the microstructure of the brain. Further, the macroscopic
structure of the brain can be characterized by means of tractography using DTI. While
IE
DTI provides a wealth of valuable information, measures of anisotropy and diffusivity
derived from DTI are often obscured by partial volume effects and complex neuronal
EV
fiber geometries. DTI further assumes that diffusion is gaussian, which is not the case in
the presence of barriers to diffusion, such as cell membranes and the extracellular matrix,
resulting in small errors and a diffusion time-dependence to the measures of anisotropy
PR

and diffusivity. More advanced frameworks for diffusion imaging, including mean appar-
ent propagator- (MAP) MRI, and constrained spherical deconvolution (CSD) are capable
of both resolving complex crossing fiber architectures, as well as better characterizing the
true, nongaussian nature of diffusion in tissue. In the following work, DTI, MAP-MRI,
and CSD are applied to cerebral palsy (CP) and carotid artery disease (CAD) in order to
better characterize the effects of intervention in children with CP and adults with CAD.
In the children with CP, we find that there are general structural changes in the corti-
cospinal tracts following 36 weeks of physical therapy, as well as a correlation between
the gross motor function measure (GMFM) and both FA and RD that may be trending
towards significance. In subjects with CAD, we find that 6 months following carotid en-
darcterectomy, a surgical procedure to remove the accumulated atherosclerotic placque
from the lining of the carotid artery, there are significant increases in the apparent fiber
density and structural connectivity largely in the ipsilateral hemisphere of the occluded
 18

artery. These works are preliminary, and require the inclusion of more subjects before
any major conlusions are drawn, but the results are compelling nonetheless.

W
IE
EV
PR
 19

CHAPTER 1

Introduction to Magnetic Resonance Imaging

Magnetic resonance imaging (MRI) has become a mainstay diagnostic tool in radiol-
ogy. The nuclear magnetic resonance (NMR) phenomenon was first observed in 1938 by

W
Rabi et al. [1]. In 1946, Felix Bloch and Edward Purcell independently published works
demonstrating the measurement of the NMR phenomenon [2, 3], and Bloch’s 1946 arti-
cle introduced the Bloch equations that describe the NMR process. With the discovery of
IE
spin-echoes by Erwin Hahn in 1950 [4], the deleterious effects of diffusion on the NMR
signal were noted, and thus the field of diffusion NMR was born. The ability to measure
EV
the random diffusive motion of water proved to be a valuable tool for characterizing the
local microstructure that water is diffusing in. After contributions from Hahn Hahn [4],
Carr and Purcell Carr and Purcell [5], and others, the pulsed gradient spin-echo diffu-
PR

sion experiment was introduced by Stejskal and Tanner in 1965 [6]. In the mid 1980s,
Le Bihan and others proposed a method for imaging diffusion in humans as a means to
better understand the microscopic environment of tissue [7]. In the late 80s and early
90s, researchers began to suggest that anisotropic diffusion in the brain could be used
to better understand the general alignment of tissues. And finally, in 1994, Peter Basser
[8, 9] introduced the diffusion tensor as a means of mathematically characterizing the 3D
microstructure of tissue. While diffusion tensor imaging (DTI) remains the most popular
diffusion imaging technique in clinical research [10], new, more sophisticated techniques
such as q-ball imaging [11], constrained spherical deconvolution [12, 13], diffusion spec-
trum imaging [14], and mean apparent propagator MRI [15, 16] can provide additional
microstructural information. In this chapter, the physics of diffusion and MRI will be
introduced as a primer for the clinical applications of diffusion imaging presented in the
following chapters.
 20

1.1 The Origin of the Magnetic Resonance Signal

All atomic and subatomic particles possess a property called spin. While the term spin
is somewhat misleading, as these particles are not actually spinning, a simple, classi-
cal physics representation provides a very straightforward description of the NMR phe-
nomenon. If a proton is represented as a spinning ball of charge (figure 1.1)[17, 18], then
one could expect that a tiny magnetic moment would be generated along its axis of rota-
tion. If this tiny magnetic moment is then placed in a much larger external magnetic field,

W
it tends to align with the external magnetic field and begins precessing about the axis of
that field. The rate of precession is given by:
IE
ω0 = γB0 (1.1)

where ω0 is the precession frequency in radians/s, known as the Larmor frequency, γ

EV
is a constant called the gyromagnetic ratio, which is 2.68 x 108 rad/s/T for hydrogen
protons, and B0 is the strength of the external magnetic field in Teslas (T). The precession
frequency is thus dependent on the strength of the external magnetic field and is in the
PR

radiofrequency (RF) range.

Figure 1.1: Simplified representation of a hydrogen nucleus as a spinning ball of positive

charge.
 21

Placing many protons in this external field can generate a combined magnetic moment
large enough to produce a detectable signal. The magnitude of this measurable magnetic
moment, M ~ , is given by [19]:
2 2
|M~ |= ρ0 γ h̄ B0 (1.2)
4kB T
where ρ0 is the proton density, h̄ is Planck’s constant, kB is Boltzmann’s constant and T
is the temperature in Kelvin. From equation 1.2, the magnitude of the magnetic moment,
and therefore the measured signal, is proportional to B0 . Surprisingly, thermal energy,

W
as measured by kB T , also acts to randomize spins in a static magnetic field, so a large
B0 is needed to align them. This explains why clinical MRI scanners use such high field
strengths, ranging from 1T - 7T, several tens of thousands of times stronger than the earths
IE
magnetic field.
The above description provides a simplified overview of a much more complicated
EV
quantum mechanical phenomenon that will not be discussed in this work for brevity.
While spin is a quantum mechanical property of atomic and subatomic particles, and the
interaction of individual particles with an external magnetic field is complex, the com-
PR

bined effect of a large ensemble of particles (e.g. 1018 hydrogen protons in 1 mm3 of
average tissue) results in a macroscopic magnetization that can be accurately and com-
pletely described using classical physics, first elucidated by Felix Bloch and described by
the famous Bloch equations.

1.2 The Bloch Equations

In 1946, Felix Bloch introduced the equations of motion of nuclear magnetization, now
typically referred to as the Bloch equations [2]. These equations, shown below, are the
~ , with
basis of all MR techniques, and describe the interaction of the net magnetization, M
 22

~
an external magnetic field, B.
      
Mx − T12 γBz −γBy Mx 0
d   
M  = −γBz − 1
   
γBx  My  +  0  (1.3)
   
dt  y   T2
M0
Mz γBy −γBx − T11 Mz T1

The system of differential equations in 1.3 describe the evolution of the magnetization
~ in the presence of any external magnetic field as a result of exciting the system.
vector M
~ as spins
The relaxation time constants T1 and T2 are used to describe the change in M

W
interact with their surroundings and release thermal energy (T1 or spin-lattice relaxation)
and as spins interact with one another, leading to an irreversible dephasing, or loss of co-
IE
herence among the excited spins (T2 or spin-spin relaxation). These relaxation processes
are fundamental to understanding and designing MR experiments, and play an important
role in the signal to noise ratio of diffusion NMR data, and are discussed in great detail in
EV
many other sources [19, 2, 18, 17].

1.2.1 Excitation and Measurement of the NMR Signal

PR

Figure 1.2A illustrates the initial state of an NMR experiment, where a sample is placed
in an external magnetic field, B0 , oriented along the z-axis. At rest, the net magnetization,
M~ , is aligned with B0 . If an additional excitation field, B~1 , is introduced to the system,
initially aligned with the x-axis (Bx and By in equation 1.3 are made nonzero), a torque is
applied to M ~ , such that M
~ will nutate about the x-axis towards the x - y plane as shown
in figure 1.2B. If B~1 is applied at the resonant frequency of the system, ω0 , such that it is
circularly polarized, M ~ will continue to nutate, following a spiral trajectory as illustrated
in figure 1.2C.
~ with the net external magnetic field can be greatly
The analysis of the interaction of M
simplified by ignoring the effects of the static B0 field. This can be achieved by selecting
a frame of reference that rotates at the Larmor frequency, often called the rotating frame
of reference, whose coordinate axes are denoted as x0 , y 0 , and z 0 . In the rotating frame of
 23

Figure 1.2: Graphical representation of the net magnetization, M ~ , being excited by a

W
~
circularly polarized RF excitation pulse. At rest, M is parallel to B0 (A). Immediately
after the application of a small magnetic field, B ~ 1 initially along the x-axis, M ~ begins to
~ 1 is applied at exactly the Larmor frequency of the spins
nutate about the x-axis (B). If B IE
~ ~
making up M , M will continue to nutate about the rotating B ~ 1 field until it is turned off.

~ experiences an effective magnetic field, Bef f , given by:

reference, M
EV
ω0
Bef f = B0 − (1.4)
γ
Expressed in the rotating frame of reference, the only nonzero magnetic field component
is B~1 , assuming that B0 is perfectly uniform, and thus M
~ will nutate around the x0 axis
PR

only. The same excitation process is shown in figure 1.3, with the rotating frame of
reference. Here, the analysis of small deviations from the Larmor frequency are much
more apparent, and the math is simplified greatly. The duration and amplitude of the B~ 1
RF pulse determines the angle through which M0 is tipped, also called the flip angle (α):
Z τ
α=γ B1 (t)dt (1.5)
0

where τ is the duration of the RF pulse. For the majority of diffusion MRI experiments,
B~1 (t) is designed such that α = 90° .

1.2.2 NMR Signal Measurement

~ now precessing about B0 in the x-y plane, there is a time-varying magnetic field,
With M
which can induce a voltage in a receiver coil (pick-up coil) according to Faraday’s law of