University of Southern California Dissertations and Theses

Characterization of lenticulostriate arteries using high-resolution black blood MRI as an early imaging biomarker for vascular cognitive impairment and dementia

(USC Thesis Other)

Characterization of lenticulostriate arteries using high-resolution black blood MRI as an early imaging biomarker for vascular cognitive impairment and dementia

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Characterization of Lenticulostriate Arteries using High-Resolution Black Blood MRI as
an Early Imaging Biomarker for Vascular Cognitive Impairment and Dementia

By
Samantha Jenny Ma

A dissertation presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
in partial fulfillment of the requirements for the degree
DOCTOR OF PHILOSOPHY
(BIOMEDICAL ENGINEERING)

May 2020

© Copyright by
Samantha Jenny Ma
2020

iii
ABSTRACT OF THE DISSERTATION

Characterization of Lenticulostriate Arteries using High-Resolution Black Blood MRI as an
Early Imaging Biomarker for Vascular Cognitive Impairment and Dementia

By

Samantha Jenny Ma
Doctor of Philosophy in Biomedical Engineering
University of Southern California, 2019
Professor Danny J.J. Wang, Chair

Cerebral small vessel disease (SVD) is the most common vascular cause of dementia and
the cause of about one fifth of all strokes worldwide. Still, the underlying mechanisms of SVD
remain poorly understood, resulting in no specific guidelines for its staging, treatment, and
prevention. The knowledge gap in cerebral SVD is partly because cerebral small vessels, including
arterioles and capillaries, are largely inaccessible to existing, clinically available in vivo imaging
technologies. Prior development in magnetic resonance imaging (MRI) techniques has
demonstrated the feasibility of non-invasively visualizing cerebral small vessels, such as the
lenticulostriate arteries (LSAs), using Time of Flight (“bright blood”) MR angiography at ultra-
high field strengths. However, this flow-related enhancement or “bright blood” technique cannot
iv
capture the finer details at distal portions of the small vessels, and thus does not translate well to
lower field strength, which is more commonly used in the clinical setting. This dissertation aims
to bridge the gap between ultra-high field imaging and clinical implementation of novel imaging
technologies by systematically optimizing high-resolution 3D black blood MRI on both 3T and 7T
scanners to directly visualize the small lenticulostriate arteries and developing automatic tools to
characterize the morphological differences that occur with aging and/or vascular risk factors.
In this dissertation, Chapter 1 provides a general introduction to the research outlined in
the following chapters. It covers the clinical significance of cerebral small vessel disease in the
progression of vascular cognitive impairment and dementia, the current state of neurovascular
imaging, and the emergence of machine learning for segmentation problems. Chapter 2 describes
a study in which a high-resolution 3D T1-weighted turbo spin echo sequence with a variable flip
angle scheme (T1w TSE-VFA) was optimized to improve the contrast of the LSAs through “black
blood” imaging. A greater number of LSA branches can be detected compared to those by time-
of-flight MR angiography (TOF MRA) at 7T. The CNR of LSAs was comparable between 7T and
3T. T1w TSE-VFA showed significantly higher CNR than TOF MRA at the stem portion of the
LSAs branching off the medial middle cerebral artery. Chapter 3 presents the development of LSA
segmentation methods for quantification of morphological differences with age. This study not
only used manual segmentation for identifying age differences, but it also used the manual
segmentation labels as supervision in an application of deep learning in order to more accurately
and automatically perform the vessel segmentation task. The number of detected LSAs by both
TSE-VFA and TOF MRA was significantly reduced in aged subjects, while the mean vessel length
measured on 7T TSE-VFA showed significant difference between the two age groups. This study
also demonstrated the feasibility of high performance automatic small vessel segmentation from
v
optimized black blood MR images using deep learning. Lastly, Chapter 4 describes a study in
which the black blood MR images were evaluated for clinical utility in the staging of cerebral
small vessel disease. A four-point LSA delineation rating that qualitatively assesses LSA branch
number and tortuosity was developed and tested for performance in distinguishing patients with
higher vascular risk and subsequent reduced cognitive function. In the unique Los Angeles Latino
Eye Study cohort, patients with poorer LSA quality tended to exhibit reduced cognitive flexibility
and executive function, and they were more frequently diagnosed with very mild dementia (Global
Clinical Dementia Rating 0.5). Taken together, findings from these studies indicate that directly
looking at the small vessels can provide insight into the earlier vascular changes that could lead to
cognitive decline in vascular dementia. Rather than relying on imaging features that occur as a
later consequence of small vessel disease, there is value in observing the morphology and
understanding how it may indicate neurovascular degeneration in the presence of vascular risk
factors.

vi
DEDICATION
This dissertation is dedicated…
To my family.

vii
ACKNOWLEDGEMENTS
The research presented in this dissertation was supported by various funding sources
including grants from the National Institutes of Health (UH2-NS100614, S10-OD025312, K25-
AG056594 and P41-EB015922). My graduate training was supported in part by fellowships from
the National Institutes of Health (RFA DA-06-011) UCLA Semel Institute NeuroImaging Training
Program (NITP) as well as multiple travel grants from the USC Viterbi School of Engineering
Women in Science and Engineering (WiSE) program. I am grateful for all the participants and
volunteers who contributed their valuable time, patience, and effort in this research.
Chapter 2 is based on a portion of the work presented in Ma, S.J., Sharifi Sarabi, M., Yan,
L., Shao, X., Chen, Y., Yang, Q., Jann, K., Toga, A.W., Shi, Y., Wang, D.J.J. (2019).
Characterization of lenticulostriate arteries with high resolution black-blood T1-weighted turbo
spin echo with variable flip angles at 3 and 7 Tesla. NeuroImage 199:184-193. It is also based on
Ma, S.J., Yan, L., Jann, K., and Wang D.J.J. (2018). High Resolution Black-blood T1-weighted
Turbo Spin Echo with Variable Flip Angles for Visualization of Small Perforating Arteries at 3
and 7 Tesla. Paper presented at the International Society of Magnetic Resonance in Medicine
(ISMRM) Annual Meeting, Paris, France.
Chapter 3 is based on a portion of the work presented in Ma, S.J., Sharifi Sarabi, M., Yan,
L., Shao, X., Chen, Y., Yang, Q., Jann, K., Toga, A.W., Shi, Y., Wang, D.J.J. (2019).
Characterization of lenticulostriate arteries with high resolution black-blood T1-weighted turbo
spin echo with variable flip angles at 3 and 7 Tesla. NeuroImage 199:184-193. The deep learning
portion is based on previous machine learning experiences in Ma, S.J., Yu, S., Liebeskind, D.,
Yan, L., Scalzo, F., and Wang, D.J.J. (2017). Regional Detection of Hemorrhagic Transformation
using Kernel Spectral Regression and a Neural Network on Multi-modal MRI for Acute Ischemic
viii
Stroke, Paper orally presented at the Radiological Society of North America (RSNA) Annual
Meeting, Chicago, IL, as well as Wang, K., Shou, Q., Ma, S.J., Liebeskind, D., Qiao, X., Saver,
J., Salamon, N., Kim, H., Yu, Y., Xie, Y., Zaharchuk, G., Scalzo, F., and Wang D.J.J. (2020). Deep
Learning Detection of Penumbral Tissue on Arterial Spin Labeling in Stroke. Stroke 51:00-00.
The material presented in Chapter 3 was submitted to the 2020 ISMRM Annual Meeting in
Sydney, Australia: Ma, S.J., Sharifi Sarabi, M., Wang, K., Heidari Pahlavian, S., Tan, W., Lodge,
M., Yan, L., Shi, Y., and Wang, D.J.J. (2020). Deep Learning Segmentation of Lenticulostriate
Arteries on 3D Black Blood MRI.
Chapter 4 is based on abstracts presented at the 2018 ISMRM Annual Meeting and the
2018 and 2019 the Alzheimer’s Association International Conference (AAIC). Ma, S.J., Yan, L.,
Barisano, G., Cao, L., Casey, M., Toga, A., Law, M., Ringman, J., and Wang D.J.J. (2018). High
Resolution 3D T1-weighted Black Blood MRI of Human Lenticulostriate Arteries as Biomarker
for Small Vessel Diseases. Paper presented at the ISMRM Annual Meeting, Paris, France, was the
preliminary study regarding the clinical utility of black blood MRI. The development and
evaluation of the lenticulostriate artery delineation (LSAD) rating scale is based on Ma, S.J.,
Barisano, G., Shao, X., Yan, L., Cao, L., Casey, M., Ringman, J., and Wang D.J.J. (2018). High
Resolution 3D Black Blood MRI of Human Lenticulostriate Arteries as Imaging Biomarker for
Vascular Cognitive Impairment and Dementia. Alzheimer’s & Dementia: the Journal of the
Alzheimer’s Association, 14(7): P1641. Lastly, the evaluation of the interaction between LSAD
and middle cerebral artery perforator territory perfusion is based on Ma, S.J., Jann, K., Barisano,
G., Shao, X., Yan, L., Casey, M., D’Orazio, L., Ringman, J., and Wang D.J.J. (2019).
Characterization of Lenticulostriate Arteries using Arterial Spin Labeling and High-resolution 3D
ix
Black-Blood MRI as an Imaging Marker in Vascular Cognitive Impairment and Dementia.
Alzheimer’s & Dementia: the Journal of the Alzheimer’s Association, 15(7): P75-P76.
This work would not have been possible without the exceptional mentorship and
unwavering support from Dr. Danny J.J. Wang. You have shown me the possibilities for being an
outstanding scientist, and I admire your abilities as a visionary, collaborator, and leader in our
field. Thank you for always giving me the guidance and gentle push toward the direction I need to
succeed in my own scientific endeavors. I am also immensely grateful for the expert insight and
additional mentorship I have received from my committee members throughout this journey: Drs.
Arthur Toga, Yonggang Shi, Lirong Yan, and Krishna Nayak. Thank you for contributing your
valuable time and energy toward my scientific growth. In addition, I wish to thank Drs. Meng Law,
Fabien Scalzo, Hosung Kim, Brent Liu, Natasha Lepore, and Farshid Sepehrband for sharing their
expertise and supporting my work. A heartfelt thank you also goes to Chris Noll, Mischal Diasanta,
and William Yang for keeping me on track since the beginning of my Trojan career in BME.
To the Laboratory of Functional MRI Technology (LOFT) member past and present, I am
grateful to be a part of such an innovative and dynamic team of individuals. Thank you to Giuseppe
Barisano, Marlene Casey, Jin Jin, Kate Krasileva, Katherin Martin, Thomas Martin, Soroush
Heidari Pahlahvian, Qinyang Shou, Robert Smith, Yi Wang, Songlin Yu, Chenyang Zhao, and
Ziwei Zhao for sharing your expertise and knowledge in everything from working the
administrative system to clinical methods, theoretical physics, and random trivia. Thank you
especially to Xingfeng Shao, Kai Wang, Mona Sharifi Sarabi, Mayank Jog, Kay Jann, and Lirong
Yan for help with troubleshooting, Friday fish tacos from Sam’s, late night discussions during the
final push for ISMRM deadlines, and everything in between daily lab life. Your support and
x
valuable brainstorming over coffee/tea and meals have made graduate school life bearable, and it
has been an absolute joy to share every celebration of our small successes.
To my friends who are more like family, thank you for being in my life. I am grateful that
you were there to listen to me while I vented my frustrations, but also were ready to get boba with
me to celebrate all the little triumphs or meet me somewhere in the world while I was presenting
at a conference. Despite the geographical distance and difference in time zones, you were always
there for me through all the ups and downs, and I cannot express how much that means to me.
Most importantly, to my dearest family, thank you from the greatest depths of my heart for
your limitless love and support. I would not be who I am and where I am without your trust and
perpetual encouragement. You inspire me daily to make a difference and leave this world a better
place than how I found it. Thank you for everything.

xi
TABLE OF CONTENTS
ABSTRACT OF THE DISSERTATION ................................................................................... iii
DEDICATION .............................................................................................................................. vi
ACKNOWLEDGEMENTS ........................................................................................................ vii
LIST OF TABLES...................................................................................................................... xiii
LIST OF FIGURES.................................................................................................................... xiv
LIST OF ABBREVIATIONS ......................................................................................................xx
1. CHAPTER 1: General Introduction to the Dissertation ....................................................1
1.1. Cerebral Small Vessel Disease and Vascular Cognitive Impairment & Dementia ........ 1
1.2. Existing In Vivo Imaging Markers of SVD .................................................................... 2
1.3. Non-contrast In Vivo Neurovascular Imaging ................................................................ 3
1.4. Machine Learning and Deep Learning Applications for Segmentation Problems ......... 7
1.5. Overview of Studies ....................................................................................................... 9
2. CHAPTER 2: High Resolution 3D T1-weighted Turbo Spin Echo with Variable
Flip Angles for the Visualization of Lenticulostriate Arteries at 3 and 7 Tesla .............12
2.1. Abstract ........................................................................................................................ 12
2.2. Introduction .................................................................................................................. 13
2.3. Materials and Methods ................................................................................................. 14
2.3.1. Extended Phase Graph Simulation ............................................................... 14
2.3.2. Subjects ......................................................................................................... 18
2.3.3. MRI Experiments .......................................................................................... 19
2.3.4. Image Analysis .............................................................................................. 21
2.3.5. Statistical Analysis ........................................................................................ 22
2.4. Results .......................................................................................................................... 22
2.4.1. EPG Simulations ........................................................................................... 22
2.4.2. Optimization of T1w TSE-VFA ..................................................................... 24
2.4.3. Evaluation of T1w TSE-VFA ......................................................................... 28
2.5. Discussion .................................................................................................................... 29
2.5.1. Clinical Value of LSA Imaging ..................................................................... 29
2.5.2. T1w TSE VFA and 7T TOF MRA .................................................................. 30
2.5.3. Aging Effects on LSAs ................................................................................... 31
2.5.4. Limitations of the Study ................................................................................ 32
2.6. Conclusion .................................................................................................................... 34
3. CHAPTER 3: Vessel Segmentation and Morphology Metrics: A Deep Learning
Application ............................................................................................................................35
3.1. Abstract ........................................................................................................................ 35
3.2. Introduction .................................................................................................................. 37
3.3. Materials and Methods ................................................................................................. 38
xii
3.3.1. Subjects ......................................................................................................... 38
3.3.2. MRI Experiment ............................................................................................ 39
3.3.3. Image Analysis: Vessel Segmentation and Morphology Metrics .................. 39
3.3.4. Statistical Analysis ........................................................................................ 41
3.3.5. Deep Learning Model Development ............................................................. 42
3.4. Results .......................................................................................................................... 44
3.4.1. Evaluation of T1w TSE-VFA Manual Segmentation and Age ...................... 44
3.4.2. Deep Learning Segmentation of LSAs .......................................................... 47
3.5. Discussion .................................................................................................................... 49
3.5.1. Aging Effects on LSAs ................................................................................... 49
3.5.2. Manual Segmentation and Shape Quantification ......................................... 50
3.5.3. Automatic DL Segmentation ......................................................................... 51
3.5.4. Limitations .................................................................................................... 51
3.6. Conclusion .................................................................................................................... 52
4. CHAPTER 4: High Resolution 3D T1-weighted Black Blood MRI of Human
Lenticulostriate Arteries as an Imaging Biomarker for Vascular Cognitive
Impairment and Dementia ..................................................................................................53
4.1. Abstract ........................................................................................................................ 53
4.2. Introduction .................................................................................................................. 54
4.3. Materials and Methods ................................................................................................. 56
4.3.1. Subjects and Clinical Evaluation .................................................................. 56
4.3.2. MRI Protocol and Analysis ........................................................................... 57
4.3.3. LSA Delineation Rating Scale ....................................................................... 57
4.3.4. Statistical Analysis ........................................................................................ 58
4.4. Results .......................................................................................................................... 58
4.4.1. Evaluation of LSAD Rating ........................................................................... 58
4.4.2. LSAD Associations with Neurocognitive Assessment and Vascular Risk ..... 60
4.4.3. LSAD*MCAperf CBF Association with Cognition and Vascular Risk ......... 62
4.5. Discussion .................................................................................................................... 63
4.5.1. Clinical Utility of Black Blood MRI and the LSAD Rating Scale ................. 63
4.5.2. LSA Structure and Cerebrovascular Function ............................................. 64
4.5.3. Limitations .................................................................................................... 65
4.6. Conclusion .................................................................................................................... 66
5. CHAPTER 5: Conclusion and Ongoing Work ..................................................................67
5.1. Future Directions in Ultra-High Field .......................................................................... 67
5.1.1. Patient Safety ................................................................................................ 67
5.1.2. Parallel Transmission (pTx) ......................................................................... 68
5.2. Ongoing Development of the Deep Learning Model .................................................. 69
5.3. Clinical Translation ...................................................................................................... 71
References .....................................................................................................................................73
APPENDIX A ...............................................................................................................................85
xiii

LIST OF TABLES

Chapter 2
Table 2.1. Summary of imaging parameters for sequences...................................................... 21
Table 2.2 Summary of Sagittal vs. Coronal TSE-VFA Parameters ........................................ 27

Chapter 3
Table 3.1 Summary of Reeb graph metrics for 3T and 7T VFA-TSE manual
segmentations of young and aged subjects (mean ± SD) ........................................ 46

Chapter 4
Table 4.1 Criteria for 4-point LSA Delineation Rating Scale ................................................. 58

xiv

LIST OF FIGURES
Chapter 1
Figure 1.1 Parenchymal lesions observed with cerebral SVD. Adapted from: Dichgans, M,
Leys, D. Circulation Research. 2017;120:573-591 ................................................... 1
Figure 1.2 Standards for reporting and imaging of SVD: example findings, schematic
representations, and descriptive imaging characteristics on MRI. Adapted from:
Shi, Y and Wardlaw, JM, Stroke and Vascular Neurology, 2016 ............................. 2
Figure 1.3 Cadaver study of three lenticulostriate arteries branching off the middle cerebral
artery (1). Adapted from: Marinković, S., et al. Clinical Anatomy. 2001; 3: 190-
195 ............................................................................................................................. 3
Figure 1.4 Possible mechanisms occurring at the lenticulostriate arteries that cause a
lacunar infarct. Adapted from: Shi, Y and Wardlaw, JM, Stroke and Vascular
Neurology, 2016 ........................................................................................................ 4
Figure 1.5 TOF MRA at 7T allows visualization of fine perforating arteries (A, 10 mm thin
maximum intensity projection); 3D T1w TSE with isotropic 0.5 mm resolution
at 3T comparably delineates the lenticulostriate arteries in a healthy subject (B,
10 mm thin minimum intensity projection). .............................................................. 4
Figure 1.6 TOF MRA uses repetitive RF-pulses to saturate the background in the image
slice. Inflowing spins remain fully magnetized......................................................... 5
Figure 1.7 T1-weighted TSE-VFA at 3T (top) and 7T (bottom). ............................................... 5
Figure 1.8 Three views of 3D GRASE pseudo-continuous ASL CBF map at 3T with
isotropic 2.5mm resolution. ....................................................................................... 6
Figure 1.9 Axial slice of 0.3mm2 in-plane resolution minimum intensity projection
susceptibility weighted image (SWI). ....................................................................... 6
Figure 1.10 The 3D U-Net architecture where blue boxes are feature maps. Adapted from
Çiçek, et al., MICCAI, 2016. .................................................................................... 8
Figure 1.11 The HighRes3DNet architecture which uses dilated convolutions and residual
connections for dense segmentation. Adapted from Li, et al., IPMI, 2017 ............... 8
Chapter 2
xv
Figure 2.1 The variable flip angle schemes from the 3T Siemens Prisma (A, top) and the
7T Siemens Terra (A, bottom) T1w TSE_VFA sequence. Signal evolution
curves were created using extended phase graph simulations for 3T parameters
(B, top) and 7T parameters (B, bottom). Normalized point spread functions
along the phase encode direction were calculated by taking the Fourier
transform of the k-space ordered magnetization transfer function (MTF, in B)
to evaluate T2 blurring due to field strength (C, top) and echo train length (C,
bottom). ................................................................................................................... 16
Figure 2.2 The pulse sequence diagram (A) for T1-weighted TSE-VFA, demonstrating the
gradient pulses involved in the elliptical k-space coverage (B) with interleaved
linear re-ordering. For each echo train, the central echo in k-space is the second
echo. ........................................................................................................................ 17
Figure 2.3 Plots of the signal difference between arterial blood and either white matter
(WM) or deep gray matter (GM) as a function of TR for 3 different echo train
lengths (ETLs) and blood flow velocities (4.5 cm/s for aged (A, C), and 8.2 cm/s
for young (B, D)) respectively. The optimal TR is 1000 ms for T1w TSE-VFA
at 3T (top row) and 1200 ms at 7T (bottom row) for signal difference between
WM and blood. The optimal TR is 1200 ms for T1w TSE-VFA at 3T and 1300
ms at 7T for signal difference between deep GM and blood (dashed lines). .......... 24
Figure 2.4 Coronal 10 mm thin slice minimum intensity projections of pilot scans using
echo train length (ETL) of 40 or 60, and TR of 600 ms and 1000 ms at 3T (top
two rows), and TR of 1200 ms at 7T (bottom row), set to the same window level.
The pilot scan results confirm trends observed in the EPG simulation, in which
the best contrast is observed for ETL = 40 and TR = 1000 ms at 3T. Due to SAR
limitations for short TR at 7T, the theoretical optimal TR of 1200 ms was used,
and the contrast observed with ETL=44 is better than that of ETL=60. ................. 24
Figure 2.5 Thin 10mm minimum intensity projection images of a pilot subject without
(left) and with ECG triggering (right). Despite the improved delineation of
LSAs at distal regions, the prolonged scan time was a strong limitation. ............... 26
Figure 2.6 Thin 10mm minimum intensity projection images of a pilot subject scanned
with the +90°y Magnetization Restore Pulse (left) and without the the +90°y
Magnetization Restore Pulse (right). The magnetization restore pulse further
suppresses the signal in the distal portions of the LSAs (red arrows) while also
suppressing the CSF signal. However, the improved CNR of vessel and tissue
enables improved manual segmentation.................................................................. 26
Figure 2.7 Thin 10mm minimum intensity projection images of a pilot subject scanned
using a sagittal acquisition (left) and a coronal acquisition (right). While the
coronal acquisition offers more efficient coverage of LSAs and reduced imaging
xvi
time, LSAs branching off the medial portion of MCA were often missed (red
arrow). ..................................................................................................................... 26
Figure 2.8 Coronal 10 mm thin slice minimum intensity projections of both young and
aged subject TSE-VFA scans at 3.0 Tesla (top row) and 7.0 Tesla (middle row).
The bottom row shows coronal 10 mm thin slice maximum intensity projection
of 7T TOF MRA. TSE-VFA can seemingly resolve more LSAs than 7T TOF
MRA, especially for the LSAs located in the medial group along the middle
cerebral artery (white arrows). ................................................................................ 28
Figure 2.9 CNR measures between blood and WM background in 3T VFA-TSE, 7T VFA-
TSE, and 7T TOF images for different portions of the LSAs for age 19-35 years
and age > 60 years respectively. The image intensity values were obtained by
plotting the profiles across the regional planes (inset) and taking the mean of the
top quartile as the tissue background signal and the mean of the bottom quartile
of the signal plot profile as the vessel blood signal. Standard deviation of the
noise was acquired from a region of interest (dotted circle) in tissue with
relatively uniform contrast. * indicates significance (p < 0.05), ** indicates
significance (p ≤ 0.001). .......................................................................................... 29
Figure 2.10 (A) Co-registered 7T 10mm thin slice intensity projection images from a 40-
year-old female participant (left to right: 10mm minimum intensity projection
(minIP) of T1w TSE-VFA and 10mm maximum intensity projections (MIP) of
T2w TSE-VFA and TOF, respectively). (B) Raw images of 7T T1w TSE-VFA
without and with overlays from bright TOF (red) and T2w TSE-VFA (yellow)
signals, which indicate blood vessels and cerebrospinal fluid in perivascular
spaces, respectively. Although CSF also appears dark in T1w TSE-VFA thereby
enlarging the apparent thickness of the LSAs, the perivascular spaces filled with
CSF appear more prominently toward the middle rather than the stem of the
LSAs. There are also LSA branches/stems that can only be visualized in T1w
TSE-VFA but not in T2w TSE-VFA or TOF images. ............................................ 32
Figure 2.11 The effect of B1 variation on contrast between arterial blood and background
white matter (A). Contrast varies approximately linearly with the magnitude of
the B1 field, and this effect is larger at 7T (red) compared to 3T (blue). However,
the LSAs are located in the central “bright spot” of the B1+ field due to standing
wave shading artifacts (dielectric effects) at 7T, which is favorable for
enhancing the CNR of LSAs. This issue can also be addressed with the use of
pTx B1 shimming, which reduces the dielectric effects toward the base of the
brain (B). ................................................................................................................. 34
Chapter 3
xvii
Figure 3.1 Lenticulostriate artery morphological quantification workflow begins with
manual vessel segmentation using ITK-SNAP on the raw TSE-VFA image (a).
The vessel volumes are reconstructed, and a mesh surface is created in
preparation for shape analysis (b). Quantitative measures such as vessel length
and tortuosity are calculated from the Reeb graph (c). ........................................... 39
Figure 3.2 Flowchart of the pre-processing, data input, network training, and evaluation of
deep learning models. Black blood images were used for manual segmentation
in ITK-SNAP, which served as supervision. The images were cropped to the
subcortical region of interest, underwent non-local means filtering, and split into
hemispheres to increase the sample size. The models with 3D U-net and
Highres3DNet were trained with the training set and then evaluated on the test
set (n=14 hemispheres from 7 subjects) relative to OOF performance. ReLU =
rectified Linear Unit. ............................................................................................... 42
Figure 3.3 Three-dimensional renderings of left LSAs from the manual segmentations of
an aged subject (A,C,E) and a young subject (B,D,F). VFA-TSE at both 3T and
7T enabled the identification of more vessels in aged subjects, especially in the
medial region. 7T VFA-TSE enabled the identification of more branches in
general in the younger subjects (i.e. orange and turquoise branch vessels in D).
Despite longer segmentation of vessels using 7T TOF images, fewer vessels
could be identified. As summarized by the box plot of vessel count, significantly
more vessels were detected in young subjects than in aged subjects (**, p<0.01),
and more vessels were detected using VFA-TSE compared to 7T TOF MRA (*,
p<0.05). ................................................................................................................... 45
Figure 3.4 Additional examples of the 3D renderings of the manual vessel segmentation
for two younger subjects with larger number of vessels and branches detected.
Despite the high vessel density, the resolution was sufficient to distinguish
secondary branches from primary stems. ................................................................ 46
Figure 3.5 Comparison of mean vessel length and tortuosity of lenticulostriate arteries in
young (age 19-35 years) and aged (age > 60 years) groups for each modality. *
indicates significance p<0.05; ** indicates significance p<0.01. ........................... 47
Figure 3.6 3D projections of segmentation results using each method. The figure illustrates
an exemplary result for one subject. Labels are shown in the first column, and
segmentation by OOF, 3D U-Net, and HighRes3DNet in the top row of the other
three columns, respectively. The bottom row shows the error maps, where red
voxels indicate true positives, green voxels false positives, and blue voxels false
negatives. Overall, HighRes3DNet achieved the highest performance. (3D
interpolations may not translate real voxel-to-voxel differences.) .......................... 48
xviii
Figure 3.7 3D projections of segmentation results using each method. This figure illustrates
the tendency for manual segmentation to miss the distal portion of the LSAs
that deep learning and OOF can seemingly detect. ................................................. 48
Figure 3.8 Boxplots of the performance metrics for each segmentation method. Dice
similarity is measured in a range [0,1], while 95% Hausdorff distance and
average Hausdorff distance are measures in voxels. HighRes3DNet produced
significantly superior performance for segmentation results relative to the
manual segmentation label compared with OOF (p<0.01 for Dice and AHD)
and 3D U-Net (p<0.05 for 95HD). * indicates p<0.05, ** indicates p<0.01. ........ 49
Chapter 4
Figure 4.1 Examples of various subjects along the LSA Delineation Rating Scale with test-
retest reliability of LSA delineation. In the male subject with LSAD rating of 1
(far left), the vessels are hardly visible, and he had diabetes and a CDR score of
0.5. The healthy aged female subject with LSAD rating of 4 (far right) has more
than 6 relatively straight vessels on each side. ........................................................ 59
Figure 4.2 Added variable plots depicting the relation between CBF in MCAperf territory
(yellow ROI) and LSAD rating for vessel quality. Although bilateral MCAperf
CBF (top, adjusted for age, gender, and global CBF) has a significantly positive
relationship with LSAD rating (p=0.0448), the Pearson correlation was only
significant in the right hemisphere (p=0.039), where our predominantly right-
handed cohort may not have as good vascular reserve capacity in early cerebral
SVD. ........................................................................................................................ 59
Figure 4.3 Added variable plots depicting the significant relation between LSAD rating for
vessel quality and hemoglobin A1c, Flanker uncorrected and fully corrected
scores, PSMTb age-corrected score, PCPS uncorrected score, and pegboard
non-dominant score. In general, patients with higher vessel quality performed
better on cognitive tests and exhibited lower HbA1c.............................................. 61
Figure 4.4 Box plots demonstrating the distribution of LSAD rating history of
hyperlipidemia, history of diabetes, and global CDR. Patients with a history of
vascular risk factors and mild cognitive decline had significantly lower LSAD
ratings. ..................................................................................................................... 61
Figure 4.5 Added variable plots depicting the significant relations between
LSAD*MCAperf CBF (corrected for age, gender, and global CBF) and MoCA
z-score (p=0.005), Flanker (p=0.007), PSMTa (p=0.011), and PSMTb (p=0.001)
uncorrected standard scores, as well as Pegboard Non-dominant Score
(p=0.034). Higher scores indicate higher level of cognitive or motor ability,
xix
which is observed in the subjects with higher LSAD*MCAperf CBF or overall
better vessel quality and perfusion. ......................................................................... 62
Figure 4.6 Box plots of the distribution of LSAD*MCAperfCBF for Global CDR 0
(normal) vs. 0.5 (very mild dementia) (p=0.0051) and Normal vs. History of
Hyperlipidemia (p=0.011). In both cases, normal subjects had significantly
higher median LSAD*MCAperfCBF, or better vessel quality and perfusion,
than affected subjects. ............................................................................................. 63
Chapter 5
Figure 5.1 Destructive excitation field interference with a conventional circularly polarized
single transmission (1Tx) coil causes characteristic strong center brightening,
while parallel transmission (pTx) allows more spatial degrees of freedom for
shape-specific B1 shimming to achieve more uniform B1 field. ............................ 68
Figure 5.2 With pTx, it is possible to carefully control the homogeneity of RF excitation.
Each RF source produces a standing wave, which creates these shading artifacts
(dielectric effects) in the periphery. By simultaneously transmitting opposing
subfields, the net B1 field experienced by the tissue would be more
homogeneous. .......................................................................................................... 69
Figure 5.3 A globe-like structure on the LSA observed in a 64 yo hypertensive and diabetic
female patient. The globe structure could be: (1) With aging and brain
inflammation, the leakiness and increased BBB permeability allows for the
passage of metabolic waste products, including fibrin and other blood products
in the PVS, obstructing the CSF-ISF flow and PVSs, causing a “fibrin globe”
and subsequent dilation of the PVS. (2) True lenticulostriate microaneurysm,
related to hypertension and other vascular risk factors, visible on TOF. (3)
Pseudo aneurysm, blood products in the LSA vessel wall or
lipohyalinosis.Adapted from Barisano, G., Ma, S.J., et al, ISMRM, 2018. ............ 72
Figure 5.4 Multi-modal sub-millimeter resolution UHF MRI enabled the determination of
a fibrin clot due to BBB leakage in the perivascular space. Adapted from
Barisano, G., Ma, S.J., et al, ISMRM, 2018. ........................................................... 72

xx
LIST OF ABBREVIATIONS
AD . . . . . . . . Alzheimer’s Disease
ASL . . . . . . . . Arterial Spin Labeling
AVD . . . . . . . . Average Hausdorff Distance
CBF . . . . . . . . Cerebral Blood Flow
CDR . . . . . . . . Clinical Dementia Rating
CNR . . . . . . . . Contrast-to-Noise Ratio
CSF . . . . . . . . Cerebral Spinal Fluid
DL . . . . . . . . Deep Learning
ECG . . . . . . . . Electrocardiogram
EPG . . . . . . . . Extended Phase Graph
ETL . . . . . . . . Echo Train Length
FWHM . . . . . . . Full Width Half Maximum
GM . . . . . . . . Gray Matter
ICC . . . . . . . . Intraclass Correlation Coefficient
LALES . . . . . . . Los Angeles Latino Eye Study
LB . . . . . . . . Laplace Beltrami
LSA . . . . . . . . Lenticulostriate Artery
LSAD . . . . . . . . Lenticulostriate Artery Delineation
M1 . . . . . . . . First gradient moment
MCA . . . . . . . . Middle Cerebral Artery
minIP . . . . . . . . Minimum Intensity Projection
MIP . . . . . . . . Maximum Intensity Projection
ML . . . . . . . . Machine Learning
MRI . . . . . . . . Magnetic Resonance Imaging
MTF . . . . . . . . Magnetization Transfer Function
95HD . . . . . . . . 95% percentile Hausdorff distance
OOF . . . . . . . . Optimally Oriented Flux
xxi
PSF . . . . . . . . Point Spread Function
PVS . . . . . . . . Perivascular Space
RF . . . . . . . . Radiofrequency
SNR . . . . . . . . Signal-to-Noise Ratio
SPACE . . . . . . . Sampling Perfection with Application optimized Contrast using different
flip angle Evolution
SVD . . . . . . . . Small Vessel Disease
SVaD . . . . . . . . Subcortical Vascular Dementia
TOF MRA . . . . Time of Flight Magnetic Resonance Angiography
TSE . . . . . . . . Turbo Spin Echo
UHF . . . . . . . . Ultra-High Field
VCID . . . . . . . . Vascular Cognitive Impairment and Dementia
VFA . . . . . . . . Variable Flip Angle
WM . . . . . . . . White Matter
WML . . . . . . . . White Matter Lesion

1
1. CHAPTER 1: General Introduction to the Dissertation
1.1. Cerebral Small Vessel Disease and Vascular Cognitive Impairment & Dementia
According to the 2018 World Alzheimer Report, there will be a new case of dementia in
the world every three seconds, especially as people are now able to live longer. The burden of
cognitive impairment in society has become increasingly important with the current 50 million
people worldwide living with dementia estimated to more than triple to 152 million by 2050
(Patterson, 2018). Alzheimer’s disease (AD) is the most common cause of dementia, but vascular
contributions to cognitive impairment and dementia are becoming increasingly recognized
(Gorelick et al., 2011). Since common risk factors such as hypertension, obesity, and diabetes
occur in both AD and cerebrovascular disease, mixed AD/vascular dementia has become a
common cause of cognitive impairment in the aged (Iadecola, 2016). As such, the clinical
differentiation of AD from vascular cognitive impairment and dementia (VCID) is blurred
(Schneider, Arvanitakis, Bang, & Bennett, 2007), and it is unclear if there may be potential
preventative vascular interventions that may slow dementia progression.
Cerebral small vessel disease (SVD)
is a heterogeneous group of sporadic and
hereditary neurological conditions including
arteriolosclerosis (hypertensive SVD),
cerebral amyloid angiopathy (CAA), venous
collagenosis, and cerebral autosomal
dominant arteriopathy with subcortical
ischemic strokes and leukoencephalopathy
(CADASIL) (Pantoni, 2010). SVD is the
Figure 1.1 Parenchymal lesions observed with cerebral SVD.
Adapted from: Dichgans, M, Leys, D. Circulation Research.
2017;120:573-591
2
cause of about one fifth of all strokes worldwide (Norrving, 2008), and it is typically classified
based on subcortical lesions like lacunar infarcts, white matter lesions, hemorrhages, and
microbleeds – all consequences on the brain parenchyma (Figure 1.1). In terms of functional
effects, SVD is a slowly progressing disease that affects the frontal-subcortical networks, causing
gradually decreased executive and memory function with mild decline in processing speed
(Lawrence et al., 2015). Histological studies can elucidate some of the underlying pathology
involved with SVD by identifying features such as loss of smooth muscle cells, fibro-hyaline
deposits, lumen narrowing, elongated/dilated vessels (microaneurysms), and vessel wall
thickening in arteriolosclerosis (Furuta, Ishii, Nishihara, & Horie, 1991). However, while it is
widely agreed that SVD plays an important role in cerebrovascular disease leading to cognitive
decline and functional loss in the elderly, the mechanisms that link SVD with parenchymal damage
and neurological deficits are heterogeneous and not thoroughly understood.
1.2. Existing In Vivo Imaging Markers of SVD

Figure 1.2 Standards for reporting and imaging of SVD: example findings,
schematic representations, and descriptive imaging characteristics on MRI. Adapted
from: Shi, Y and Wardlaw, JM, Stroke and Vascular Neurology, 2016
3
The early degenerative processes that occur in cerebral SVD are poorly characterized due
to limitations in clinically available in vivo technology. As shown in Figure 1.2, the current clinical
diagnosis of SVD relies on conventional magnetic resonance imaging (MRI) findings including
lacunar infarcts, white matter lesions (WML), cerebral microbleeds, prominent perivascular
spaces, and secondary brain atrophy (Yulu Shi & Wardlaw, 2016; Wardlaw et al., 2009). These
parenchymal lesions are the consequences of SVD after significant damage has already occurred,
thus these imaging markers may not be ideal surrogate markers of early microvascular alterations.
1.3. Non-contrast In Vivo Neurovascular Imaging
The microvasculature involved in SVD includes
small arteries/arterioles (~10µm-1mm), capillaries
(<10µm), and venules (~10-50µm) (Charidimou,
Pantoni, & Love, 2016). For the purposes of this
dissertation, the lenticulostriate arteries (LSAs) were
chosen as the focus for imaging with high resolution and
high fidelity at both ultra-high magnetic field strength
(UHF 7T) and clinically available field strength (3T).
The LSAs play a significant role in SVD because they supply the subcortical regions that are
involved in executive function and memory. As shown in Figure 1.3, these perforating arterioles
that are on the order of 500 microns branch off in a nearly perpendicular manner directly from the
high flow middle cerebral artery (MCA) (Marinković, Gibo, Milisavljević, & Ćetković, 2001).
Consequently, the LSAs are susceptible to either form microaneurysms that may rupture or
become occluded by an embolus or atheroma that may cause a lacunar infarct, both of which would
result in severe negative functional effects (Figure 1.4) (Yulu Shi & Wardlaw, 2016).
Figure 1.3 Cadaver study of three
lenticulostriate arteries branching off the
middle cerebral artery (1). Adapted from:
Marinković, S., et al. Clinical Anatomy. 2001;
3: 190-195
4

Although these vessels are generally believed to be inaccessible to existing in vivo MRI,
recent development of high-resolution MRI with sub-millimeter spatial resolution has enabled at
least the visualization of the small artery/arteriole end of the microvascular spectrum. For example,
as displayed in Figure 1.5A, high resolution time of flight MR angiography (TOF MRA) at UHF
7T allows the visualization of the fine perforating arteries (or arterioles on the order of a few
hundred microns) (Cho et al., 2008). TOF MRA is a non-contrast “bright blood” imaging technique
Perforating arteriolar
atheroma
Figure 1.4 Possible mechanisms occurring at the lenticulostriate arteries that cause a lacunar
infarct. Adapted from: Shi, Y and Wardlaw, JM, Stroke and Vascular Neurology, 2016
Figure 1.5 TOF MRA at 7T allows visualization of fine perforating arteries (A, 10
mm thin maximum intensity projection); 3D T1w TSE with isotropic 0.5 mm
resolution at 3T comparably delineates the lenticulostriate arteries in a healthy
subject (B, 10 mm thin minimum intensity projection).
5
that is optimized for flow enhancement. It employs multiple repetitive radio frequency (RF)-pulses
with larger flip angles to suppress background signal in order to produce high signal contrast from
“fresh” inflowing blood (Figure 1.6). However, TOF techniques are somewhat insensitive to in-
plane flow because maximal
enhancement of flow can only occur
when the vessel is perpendicular to the
plane of imaging (Saloner, 1995).
Furthermore, in order to translate the
MRA technique from 7T to 3T for
clinical studies, prolonged scan time (>
10 minutes) is required and/or resolution must be compromised.
A new high-resolution “black blood” MRI technique emerged in the last decade for
intracranial vessel wall imaging at 3T using a 3D turbo spin-echo (TSE) sequence with a variable
flip angle (VFA) scheme (Qiao et al., 2011; Qiao et al., 2014).
This method is able to achieve isotropic 0.4-0.5mm spatial
resolution, and the “black blood” contrast is due to the
inherent flow suppression as displayed in Figure 1.5B. In this
sequence, the flowing spins dephase along the long echo
train, resulting in a flow void or “black blood” effect. TSE-
VFA is also able to acquire near whole brain coverage with
high resolution in less than 10 minutes. Implementing this
sequence at 7T produces even greater performance in terms
of delineating LSAs with the higher signal-to-noise ratio
Figure 1.6 TOF MRA uses repetitive RF-pulses to saturate the
background in the image slice. Inflowing spins remain fully
magnetized.
Figure 1.7 T1-weighted TSE-VFA at 3T
(top) and 7T (bottom).
6
(SNR) that can be attained with higher field strength (Figure 1.7). Since the LSAs are located near
the center of the head where there are minimal effects of B1 inhomogeneity, they are in the ideal
location for optimal contrast between blood and the surrounding subcortical tissue.
In addition to these two techniques, there are
several other directions of development in the
neurovascular imaging field, especially at UHF (De
Cocker et al., 2018). To visualize and measure blood
flow and pulsatility in small vessels, phase-contrast
(PC)-MRA can be applied at 7T as a functional
assessment of healthy or diseased arterioles (Kang et
al., 2016; van Ooij et al., 2013). Arterial spin labeling
(ASL) is an appealing perfusion imaging approach
that utilizes magnetically labeled arterial blood water
as an endogenous tracer (Wu, St Lawrence, Licht, &
Wang, 2010). ASL has previously been applied in a
range of cerebrovascular disorders and dementia
(Alsop, Dai, Grossman, & Detre, 2010; Detre, Rao,
Wang, Chen, & Wang, 2012), and it is able to directly
visualize and quantify microvascular flow for vessels
beyond the resolution of MRI because the
background tissue signal is removed by pair-wise
subtraction of label and control images. ASL is an entirely non-invasive, repeatable imaging
technique that provides absolute cerebral blood flow (CBF, Figure 1.8) which is highly
Figure 1.8 Three views of 3D GRASE pseudo-
continuous ASL CBF map at 3T with isotropic
2.5mm resolution.
Figure 1.9 Axial slice of 0.3mm2 in-plane
resolution minimum intensity projection
susceptibility weighted image (SWI).
7
reproducible across time scales from minutes, hours, to days (Floyd, Ratcliffe, Wang, Resch, &
Detre, 2003; J. Wang et al., 2003). Although ASL is currently preferentially performed at 3T due
to inhomogeneous B1 magnetic fields at 7T, parallel transmission capability and upgraded head
coils with larger coverage will soon enable UHF ASL with boosted SNR. Lastly, susceptibility
weighted imaging (SWI) at UHF has particularly good performance in the visualization of cerebral
veins (Figure 1.9). SWI employs the magnetic susceptibility differences between a tissue of
interest (i.e. venous blood) and the surrounding tissue (i.e. background) (Deistung et al., 2008).
Especially at UHF, local field inhomogeneities produced by low oxygenated venous blood creates
phase shifts and signal dephasing, thereby producing exceptional venous vessel contrast and detail
(Duyn et al., 2007).
1.4. Machine Learning and Deep Learning Applications for Segmentation Problems
Over the last few years, machine learning (ML) has emerged as a groundbreaking tool for
medical image computing (Zaharchuk, Gong, Wintermark, Rubin, & Langlotz, 2018), and deep
learning (DL) in particular has become a popular method given the rapid improvements in
computing resources. Conventional ML models use algorithms created with careful engineering
and extensive domain expertise to parse data, extract specific features from said data, and detect
or classify patterns from the input. DL is an advanced implementation of ML in which
representation learning is utilized, where the algorithm is fed with raw data but can automatically
discover representations needed for detection based on simple but non-linear modules, removing
the need for considerable domain expertise (LeCun, Bengio, & Hinton, 2015).
8
In particular, DL has shown great
potential in the application of
medical image segmentation. Two
network architectures have become
widely used for segmentation
problems: 3D U-Net and
HighRes3DNet. The 3D U-Net
(Figure 1.10) is one of the most
popular neural network architectures
in medical image computing for the
general segmentation of volumetric
objects within an image (Çiçek,
Abdulkadir, Lienkamp, Brox, &
Ronneberger, 2016; Ronneberger,
Fischer, & Brox, 2015). Essentially, the U-Net architecture relies heavily on data augmentation to
maximize the efficiency of the annotated samples through a fully convolutional network. The
contracting path is used to capture context by employing upsampling operators instead of pooling
operators to increase the resolution of the output. The symmetric expanding path combines the
high-resolution features with the upsampled output to precisely localize the classified object.
Another architecture that has gained popularity in recent years is the more compact HighRes3DNet
(Wenqi Li et al., 2017). HighRes3DNet was designed for the segmentation of fine structures in
volumetric images, a great fit for our goal of automatically segmenting small vessels. In contrast
to the 3D U-Net, HighRes3DNet (Figure 1.11) integrates high spatial resolution feature maps
Figure 1.11 The HighRes3DNet architecture which uses dilated
convolutions and residual connections for dense segmentation.
Adapted from Li, et al., IPMI, 2017
Figure 1.10 The 3D U-Net architecture where blue boxes are feature
maps. Adapted from Çiçek, et al., MICCAI, 2016.
9
throughout its layers, and it uses dilated convolutions for arbitrary enlargement of receptive fields
for more accurate dense predictions and detailed segmentation maps along object boundaries.
Despite the advancements in DL methods, the machine is limited in what it can learn based on
the data it is fed. As such, previous work in automatic segmentation of cerebral blood vessels is
often limited to large vessels of the Circle of Willis from TOF MRA, which are easier to roughly
segment via thresholding (Livne et al., 2019; Phellan, Peixinho, Falcão, & Forkert, 2017). In this
dissertation, we collected a unique set of black blood MRI data from a precious cohort of elderly
Latinx individuals to serve as training data of small vessels. LSAs were meticulously segmented
by hand to provide the label supervision for training these novel neural networks. The feasibility
of performing automatic segmentation on black blood MR images was tested to demonstrate that
small vessels can also be automatically segmented for morphological analysis in a clinical cohort.
1.5. Overview of Studies
The purpose of this dissertation is to bridge the gap between UHF imaging and clinical
implementation of novel imaging technologies by systematically optimizing high-resolution 3D
black blood MRI on both 3T and 7T scanners to directly visualize the small lenticulostriate arteries,
and also by developing automatic tools to characterize the morphological differences that occur
with aging and/or vascular risk factors.
High Resolution 3D T1-weighted Turbo Spin Echo with Variable Flip Angles for the
Visualization of Lenticulostriate Arteries at 3 and 7 Tesla
In Chapter 2, a high-resolution 3D T1-weighted turbo spin echo sequence with a variable
flip angle scheme (T1w TSE-VFA) was implemented and optimized to improve the contrast of the
LSAs through “black blood” imaging at both 3T and 7T. A greater number of LSA branches can
be detected compared to those by time-of-flight MR angiography (TOF MRA) at 7T, especially in
10
the medial portion of the MCA. We demonstrated that the CNR of LSAs was comparable between
7T and 3T imaging, and T1w TSE-VFA showed significantly higher CNR than TOF MRA at the
stem portion of the LSAs branching off the medial middle cerebral artery in both participants
between 19-35 years old and those over 60 years old.
Vessel Segmentation and Morphology Metrics: A Deep Learning Application
In Chapter 3, the development of LSA segmentation methods for quantification of
morphological differences with age is described. In the first portion of the study, manual
segmentation with ITK-SNAP is used for identifying age differences. Then, the manual
segmentation labels are applied as supervision in an application of deep learning in order to more
accurately and automatically perform the vessel segmentation task. The number of detected LSAs
by both TSE-VFA and TOF MRA was significantly reduced in aged subjects, while the mean
vessel length measured on 7T TSE-VFA showed significant difference between the two age
groups. This study demonstrates the feasibility of high performance automatic small vessel
segmentation from optimized black blood MR images using deep learning.
High Resolution 3D T1-weighted Black Blood MRI of Human Lenticulostriate Arteries as an
Imaging Biomarker for Vascular Cognitive Impairment and Dementia
Chapter 4 describes a study in which the black blood MR images are evaluated for clinical
utility in the evaluation of vascular cognitive impairment and dementia. A four-point LSA
delineation (LSAD) rating that qualitatively assesses LSA branch number and tortuosity was
developed and tested for performance in distinguishing patients with higher vascular risk and
subsequent reduced cognitive function. In the unique Los Angeles Latino Eye Study cohort,
patients with poorer LSA quality tended to exhibit reduced cognitive flexibility and executive
function, and they were more frequently diagnosed with very mild dementia (Global Clinical
11
Dementia Rating 0.5). In addition, the interaction term of LSAD rating and regional perfusion
(CBF) in the middle cerebral artery perforator territory was evaluated as an imaging-based measure
of vessel quality and function relative to cognitive performance. Participants with higher vessel
quality and higher perfusion tended to perform better on cognitive tasks involving executive
function. These findings indicate that preserving the vessel quality is an important aspect of
slowing the progression of cognitive decline.

12
2. CHAPTER 2: High Resolution 3D T1-weighted Turbo Spin Echo with Variable Flip
Angles for the Visualization of Lenticulostriate Arteries at 3 and 7 Tesla
2.1. Abstract
Objectives
The lenticulostriate arteries (LSAs) with small diameters of a few hundred microns take origin
directly from the high flow middle cerebral artery (MCA), making them especially susceptible to
damage (e.g. by hypertension). This study aims to present high resolution (isotropic ~0.5 mm),
black blood MRI for the visualization and characterization of LSAs at both 3T and 7T.
Materials and Methods
T1-weighted 3D turbo spin-echo with variable flip angles (T1w TSE-VFA) sequences were
optimized for the visualization of LSAs by performing extended phase graph (EPG) simulations.
Twenty healthy volunteers (15 under 35 years old, 5 over 60 years old) were imaged with the T1w
TSE-VFA sequences at both 3T and 7T.
Results
LSAs can be clearly delineated using optimized 3D T1w TSE-VFA at 3T and 7T, and a greater
number of LSA branches can be detected compared to those by time-of-flight MR angiography
(TOF MRA) at 7T. The CNR of LSAs was comparable between 7T and 3T. T1w TSE-VFA
showed significantly higher CNR than TOF MRA at the stem portion of the LSAs branching off
the medial middle cerebral artery.
Conclusion
The high-resolution black-blood 3D T1w TSE-VFA sequence offers a new method for the
visualization and delineation of LSAs at both 3T and 7T, which may be applied for several
pathological conditions related to the damage of LSAs.
13
2.2. Introduction
Small arteries and arterioles, particularly the lenticulostriate arteries (LSAs) are known to
be involved in silent strokes, which contribute to progressive cognitive impairment in elderly
persons. Detailed anatomical studies on the LSAs have been performed following the discovery of
miliary aneurysms or microaneurysms along the LSAs by Charcot and Bouchard (J. Charcot &
Bouchard, 1868). These studies revealed that the LSAs take origin directly from the high flow
middle cerebral artery (MCA) and consist of either single vessels with small outer diameters of
only 0.08-1.4 mm (Marinković et al., 2001) or branches off common stems with outer diameters
of 0.6-1.8 mm (Umansky et al., 1985). This abrupt size and flow change make them especially
susceptible to damage (e.g. by hypertension) (Dichgans & Leys, 2017). The LSAs supply
important subcortical areas including the caudate nucleus, globus pallidus, putamen, and part of
the posterior limb of the internal capsule (L. Alexander, 1942; Beevor, 1907; Duret, 1873), and
can be divided into medial and lateral groups. The lateral group commonly includes a LSA that
Charcot called “the artery of the cerebral hemorrhage” which may rupture and result in an
intracerebral hemorrhage and subsequent damage to the subcortical regions (J. M. Charcot, 1883).
Additionally, LSAs are "end arteries", meaning that the regions they supply have little or no
collateral blood supply. When occluded, it produces a lacunar infarct in the tissue they supply. The
origin of LSAs is also a common site of MCA aneurysm (Umansky et al., 1985).
High resolution time-of-flight MR angiography (TOF MRA) at ultrahigh magnetic field of
7T has been applied for the visualization of LSAs (Cho et al., 2008; Hendrikse, Zwanenburg,
Visser, Takahara, & Luijten, 2008). However, ultrahigh magnetic field is not commonly available
in clinical practice. High-resolution black blood MRI is a technique originally developed for
imaging intracranial vessel wall and plaques using 3D turbo spin-echo (TSE) sequences. Recently,
14
TSE sequences (A. L. Alexander et al., 1998; E. R. Melhem, Jara, & Yucel, 1997) with variable
flip angles (VFA) have been developed for black blood angiography (Yoneyama, Nakamura,
Tabuchi, Takemura, & Obara, 2012) and vessel wall imaging (Z. Fan et al., 2016; Qiao et al., 2011;
Qiao et al., 2014; van der Kolk et al., 2011). The long echo train of the TSE technique offers two
advantages for visualizing small vessels: 1) adequate flow suppression by inherent dephasing of
flowing signals (black blood MRI); 2) high spatial resolution (isotropic 0.5-0.6mm) and near
whole-brain coverage in a clinically acceptable time (<10min). These features suggest that TSE-
VFA may also be suitable for imaging cerebral small vessels such as LSAs and other perforating
arteries.
The purpose of this study was to present high resolution (isotropic ~0.5mm) 3D T1-
weighted TSE-VFA as a new approach for visualizing LSAs of young and aged subjects at
standard clinical magnetic field strength of 3 Tesla and ultrahigh magnetic field of 7 Tesla. The
imaging parameters of T1w TSE-VFA sequences were optimized through extended phase graph
(EPG) simulations, and the visualization of LSAs was compared with that by TOF MRA at 7T.
2.3. Materials and Methods
2.3.1. Extended Phase Graph Simulation
High-resolution black blood MRI with 3D TSE-VFA has been introduced recently for
imaging intracranial vessel wall and plaques by utilizing an optimal VFA scheme to achieve a
longer echo train length (ETL) for more effective flow suppression and a higher signal-to-noise
ratio (SNR) efficiency compared to standard 3D TSE sequences (Busse et al., 2008; Mugler, 2014;
Park, Mugler, Horger, & Kiefer, 2007; Qiao et al., 2011). However, existing 3D TSE-VFA
techniques for vessel wall imaging were targeted for the suppression of relatively fast blood flow
in large vessels as well as for maximizing the contrast between vessel wall and the surrounding
15
cerebrospinal fluid (CSF). In order to enhance the visualization of small vessels such as LSAs and
optimize the contrast between LSAs and surrounding brain tissue (white matter and deep gray
matter), the spin evolutions of 3D T1w TSE-VFA were simulated using the EPG.
In this study, we employed the default VFA scheme provided by the vendor for generating
T1w contrast for black blood MRI (Fig. 2.1A) (Zheng et al., 2016). The signal evolution curves,
or the magnetization transfer function (MTF), for various ETLs were generated using EPG
simulation software (Hargreaves, 2012) developed in MATLAB (Mathworks, MA, USA) (Fig.
2.1B). The simulation considered the effects of the flip angles (FA) of the radiofrequency (RF)
pulses, T1/T2 relaxations, and flow velocities on echo intensities (or the transverse magnetization
component that develops during free precession). Point spread function (PSF) analysis was
performed to evaluate the image blurring at higher field strength (Fig. 2.1C, top) or with a longer
echo train (Fig. 2.1C, bottom), considering elliptical k-space coverage within the ky-kz plane
(Busse et al., 2008), in which the second echo is the central echo with interleaved linear ordering
on a non-separable grid (Fig. 2.2B). The Fourier transform of the MTF is the PSF along the phase
encode direction, and the full width at half maximum (FWHM) measurement can be used as a
proxy for the effective resolution achieved in the image.
Literature values of T1 and T2 relaxation times were used: T1/T2=1084/69 ms and
T1/T2=1220/47 ms for white matter (WM) at 3T and 7T, respectively; T1/T2=1332/99 ms and
T1/T2=1644/47 ms for deep gray matter (GM) at 3T and 7T, respectively; T1/T2=1932/275 ms
and T1/T2=2587/68 ms for arterial blood at 3T and 7T, respectively (Blockley et al., 2008; Cox,
2008; Dobre, Uğurbil, & Marjanska, 2007; Krishnamurthy et al., 2014; Wenbo Li et al., 2016;
Stanisz et al., 2005; Wright et al., 2008). As the LSAs are located among the white matter and deep
16
gray matter, EPG simulations were performed for both tissues to achieve optimal results.
Using the EPG algorithm, we can observe the magnetization evolution and echo formation
along multiple RF pulses. In order to incorporate slow flow dephasing of arterial blood signal
within the LSAs into the EPG simulation, the following equation was used to calculate the accrued
interpulse phase of the transverse magnetization, Δφ(n):
[1]
where γ is the gyromagnetic ratio (42.58 MHz/T), v is the average velocity of blood flow in small
arteries (i.e. 4.5-8.2 cm/s) (Bouvy et al., 2016; Schnerr et al., 2017), n is the sequential order along
the echo train, τ is half of the echo spacing, and G is the applied gradient amplitude. Laminar flow
was employed to approximate the flow distribution in LSAs (Guyton & Hall, 1991). The flow
Figure 2.1 The variable flip angle schemes from the 3T Siemens Prisma (A, top) and the 7T Siemens Terra (A, bottom)
T1w TSE_VFA sequence. Signal evolution curves were created using extended phase graph simulations for 3T
parameters (B, top) and 7T parameters (B, bottom). Normalized point spread functions along the phase encode
direction were calculated by taking the Fourier transform of the k-space ordered magnetization transfer function
(MTF, in B) to evaluate T2 blurring due to field strength (C, top) and echo train length (C, bottom).
17
distribution p(v), or the area of the blood vessel cross-section containing blood traveling at
velocities between v and v+dv is given by (Maccotta, Detre, & Alsop, 1997)
[2]
where vmax is the maximum flow velocity. It can be shown that Eq. [1] still applies to laminar flow,
with v being the average velocity of blood flow or half of vmax (see Appendix A for derivation).
Based on the reported mean flow velocities in LSAs (4.5-8.2 cm/s) (Bouvy et al., 2016; Schnerr et
al., 2017) and the first gradient moment (M1), the accrued interpulse phase can be estimated. In
this experiment, the first gradient moments for the echo train along the 3 axes were: M1readout =
0.0936 mT·s2/m, M1phase-encode = 0.0225 mT·s2/m, M1slice = -6.8754 mT·s2/m at 3T; and M1readout
= 0.0917 mT·s2/m, M1phase-encode = 0.0213 mT·s2/m, M1slice = -6.3806 mT·s2/m at 7T respectively.
The gradient amplitude along the phase-encode (A-P) direction was used in the simulation to
represent the most conservative estimate of flow induced phase accrual.

Figure 2.2 The pulse sequence diagram (A) for T1-weighted TSE-VFA, demonstrating the gradient pulses involved in
the elliptical k-space coverage (B) with interleaved linear re-ordering. For each echo train, the central echo in k-
space is the second echo.

The calculated accrued phase term was added to the phase of each configuration state of
the spin system associated with a refocusing RF pulse in the TSE sequence, as described previously
18
(Weigel, 2015). As employed for in vivo scanning, a centric ordering scheme on a non-separable
grid was used for k-space encoding and the second echo was used as the TE (Figure 2.2).
Following the echo train, the magnetization was rotated to the negative z-axis by a +90 RF pulse
along the y-axis, which consequently suppressed the CSF signal (Becker & Farrar, 1969; Zhaoyang
Fan et al., 2017; Elias R. Melhem, Itoh, & Folkers, 2001; Van Uijen & Den Boef, 1984). Since
CSF still has remaining transverse magnetization at the end of the echo train due to its long T2,
the +90 RF pulse tips the magnetization to the negative z-axis, causing CSF signal attenuation at
the beginning of the following RF pulse train. Although CSF signal attenuation is not typically
desired for intracranial angiography, this additional +90  RF pulse along the y-axis was evaluated
for its effects on the LSA contrast. The simulations were performed with varying TRs and ETLs
to optimize the contrast between arterial blood in LSAs and surrounding WM and deep GM
respectively. In addition, the effect of B1 variations on the contrast between LSAs and WM was
evaluated by repeating EPG simulations with B1 variations between -20% and 20%.
2.3.2. Subjects
A total of 24 healthy volunteers were recruited for this study after they provided written
informed consent following a protocol approved by the Institutional Review Board (IRB) of the
University of Southern California. Of these volunteers, 3 were excluded due to failures to complete
the study protocol. Within the remaining 21 volunteers, 3 subjects (3 males, age 23.7 ± 2.5 years)
participated in the pilot study to optimize the T1w TSE-VFA sequence. The rest 17 healthy
volunteers participated in the evaluation study of the optimized T1w TSE-VFA, including 12
participants (7 males, 27 ± 3.5 years) between 19-35 years of age and 5 participants (2 males, 64.2
± 1.9 years) more than 60 years of age, herein referred to as the young and aged group, respectively.
One more middle-aged participant (female, 40 years) was scanned with T2-weighted TSE-VFA to
19
evaluate the contribution of perivascular space (PVS) to the T1w TSE-VFA images. All
participants were screened for history of or concurrent neurological or psychiatric disorder or
systemic disease.
2.3.3. MRI Experiments
All MRI scans were performed on a Siemens 3T Prisma scanner with a product 32-channel
head receive coil and body transmit coil, and on a Siemens 7T Terra scanner (Erlangen, Germany)
with a single-channel transmit/32-channel receive (1Tx/32Rx) head coil (Nova Medical,
Wilmington, MA, USA). The imaging protocols for T1w TSE-VFA were based on the clinical
protocols of the SPACE (Sampling Perfection with Application optimized Contrast using different
flip angle Evolution) sequence (Lichy et al., 2005) at 3 and 7T respectively. Pilot scans were
performed at both 3T and 7T to evaluate the parameter optimization of TR and ETL (number of
echoes) based on the EPG simulations. Additionally, various imaging parameters were evaluated
including the use of slab-selective vs. non-selective excitation, saturation bands, magnetization
restoration at the end of the echo train, and imaging slab orientation (sagittal vs. coronal). ECG
triggering was also tested to evaluate the potential effect of flow pulsatility on the delineation of
LSAs.
For the evaluation study, each participant underwent back-to-back scans on the 3T Prisma
and 7T Terra systems, respectively, on the same day with counterbalanced order. The imaging
protocol included a 3D MPRAGE structural scan at 3T (TR/TE=2300/2.98 ms, matrix=240x256,
resolution=1x1x1 mm3, scan time=5:12 min) and a 3D MP2RAGE structural scan at 7T
(TR/TE=4500/3.43 ms, matrix = 320x320, resolution=0.7x0.7x0.7 mm3, scan time=9:46 min). A
high-resolution 3D time of flight (TOF) sequence for bright blood MR angiography (MRA) was
performed at 7T for the visualization of the LSAs according to Kang et al. (TR/TE=12/4.67 ms,
20
matrix=548x672, resolution=0.3x0.3x0.3 mm3, GRAPPA factor = 3 in the phase-encode direction,
FA = 20 degrees, scan time=9:25 min) (Cho et al., 2008; C.-K. Kang et al., 2009). The optimized
imaging protocols for T1w TSE-VFA were performed at 3T and 7T, respectively. Imaging
parameters for 3T T1w TSE-VFA were: TR/TE=1000/12 ms, turbo factor = 44, matrix
size=756x896, resolution=0.51x0.51x0.64 mm3, sagittal slab with 160 slices and 10%
oversampling, GRAPPA factor = 2 in the phase-encode direction, total imaging time =8:39 min, 2
sagittal saturation bands were placed at the left and right temporal regions to suppress the out-of-
FOV signals. Imaging parameters for 7T T1w TSE-VFA were: TR/TE=1200/13 ms, turbo factor
= 40, matrix size= 320x320, resolution=0.5 mm3 isotropic, sagittal slab with 288 slices and no slice
oversampling, GRAPPA factor = 3 in the phase-encode direction, total imaging time =10:05 min.
No saturation bands were applied due to specific absorption rate (SAR) limitations at 7T. For both
3T and 7T T1w TSE-VFA, centric ordering with radial k-space sampling pattern (within ky-kz
plane) was employed to minimize potential head motion effects. Following the echo train, the
magnetization was rotated to the negative z-axis by a +90 RF pulse along the y-axis to suppress
the CSF signal (Zhaoyang Fan et al., 2017; Van Uijen & Den Boef, 1984). The total time of the
3T and 7T study was typically 1.5 hours. The scan protocol parameters are summarized in Table
2.1. To avoid degradation of the images due to motion, additional steps were implemented to
physically limit motion including packing the head coil with cushions and applying paper tape
across the coil on the forehead skin for tactile feedback.
21

2.3.4. Image Analysis
CNR Quantification
All images were reviewed to ensure the absence of visible motion corruption or other
artifacts. 3D T1w TSE-VFA images were reoriented in the coronal view, and LSAs were
visualized by thin-slab (10 mm) minimum intensity projection (minIP) for qualitative review. The
contrast to noise ratio (CNR) between arterial blood in the LSA and surrounding WM was
calculated using Fiji software (Schindelin et al., 2012) by drawing a line across multiple dark
vessels in an axial slice of the raw image data with slices containing the lateral stems, medial stems,
lateral mid-length, and lateral distal portions, respectively. The signal intensity along the lines (i.e.,
signal profile in Figure 2.9 inset) was plotted. The signal intensities along the valleys (mean values
in the lower quartile) were considered measures of the arterial blood signal. A 20 mm2 circular
Table 2.1. Summary of imaging parameters for sequences
22
region of interest (ROI) was drawn on the homogeneous WM that was adjacent to the basal ganglia
region with matched locations on 3T and 7T images, respectively. The CNR was calculated by
taking the signal intensity difference between WM and arterial blood of LSAs divided by the
standard deviation of signals in the WM ROI (Zhang et al., 2019).
The raw images of TOF MRA at 7T were co-registered with 3D T1w TSE-VFA images at
7T in each subject using FreeSurfer (Fischl et al., 2002; Fischl et al., 2004) and Elastix version 4.8
(Klein, Staring, Murphy, Viergever, & Pluim, 2010; Shamonin et al., 2014) in Laboratory of Neuro
Imaging (LONI) Pipeline (Rex, Ma, & Toga, 2003). The CNR of TOF MRA was calculated along
the same signal profile lines and WM ROI used for T1w TSE-VFA, except that the signal
intensities along the peaks (mean values in the upper quartile) were taken as the arterial blood
signal.
2.3.5. Statistical Analysis
Statistical analysis was performed with STATA 13.1 (College Station, Texas). CNR
measurements were compared across the three techniques (T1w TSE-VFA at 3T and 7T, and TOF
MRA at 7T) using within-subject ANOVA, followed by post-hoc paired t-test for pairwise
comparison. A p-value ≤ 0.05 (two-sided) was considered statistically significant.
2.4. Results
2.4.1. EPG Simulations
The VFA schemes (Zheng et al., 2016) employed for T1w TSE are displayed in Figure
2.1A. The signal evolution along an echo train as well as the corresponding point-spread functions
(PSF) considering the elliptical k-space coverage (within the ky-kz plane, Figure 2.2B) (Busse et
al., 2008) are displayed in Figure 2.1B and 2.1C, respectively, for 3T (Figure 2.1, top row) and
7T (Figure 2.1, bottom row). Using the T1w VFA schemes provided by the vendor, the PSF is
23
sharper at 7T (FWHM=1.47 voxels) compared to 3T (FWHM=1.59 voxels). The employed VFA
schemes resulted in comparable T2 blurring along the echo train or comparable PSFs across
different ETLs at both 3 and 7T (Fig. 2.1C).
Taking into account phase accrual due to various blood flow velocities in the LSAs (4.5-
8.2 cm/s), the optimal CNR between arterial blood and white matter is achieved with a TR of 1000
ms for the ETL of 33 and 41 or a TR of 1100 ms for the ETL of 60 at 3T, and a TR of 1200 ms for
the ETL of 33, 41, and 60 at 7T, respectively (Figure 2.3A&B). The simulated CNR curves
between arterial blood and deep gray matter at 3 and 7T respectively are shown in Figure 2.3C&D.
The optimal CNR between arterial blood and deep gray matter is achieved with a TR of 1200 ms
for the ETL of 33 and 41 or a TR of 1300 ms for the ETL of 60 at 3T, and a TR of 1300 ms for all
3 tested ETLs at 7T. Based on the simulation results to achieve the maximal CNR with an imaging
time of ≤ 10min, the ETL of 40 with TR of 1000 ms and the ETL of 44 with TR of 1200 ms were
employed for in vivo imaging at 3T and 7T, respectively (Figure 2.4). The effect of B1 variations
on the contrast between LSAs and WM was also simulated, and the results showed that the contrast
varies approximately linearly with the magnitude of B1 field, and the B1 effect is larger at 7T
compared to 3T (Figure 2.11A).
24

2.4.2. Optimization of T1w TSE-VFA

Figure 2.3 Plots of the signal difference between arterial blood and either white matter (WM) or deep gray matter
(GM) as a function of TR for 3 different echo train lengths (ETLs) and blood flow velocities (4.5 cm/s for aged (A,
C), and 8.2 cm/s for young (B, D)) respectively. The optimal TR is 1000 ms for T1w TSE-VFA at 3T (top row) and
1200 ms at 7T (bottom row) for signal difference between WM and blood. The optimal TR is 1200 ms for T1w
TSE-VFA at 3T and 1300 ms at 7T for signal difference between deep GM and blood (dashed lines).
Figure 2.4 Coronal 10 mm thin slice minimum intensity projections of pilot scans using echo train length (ETL) of 40
or 60, and TR of 600 ms and 1000 ms at 3T (top two rows), and TR of 1200 ms at 7T (bottom row), set to the same
window level. The pilot scan results confirm trends observed in the EPG simulation, in which the best contrast is
observed for ETL = 40 and TR = 1000 ms at 3T. Due to SAR limitations for short TR at 7T, the theoretical optimal
TR of 1200 ms was used, and the contrast observed with ETL=44 is better than that of ETL=60.
25
Pilot studies were performed to optimize the T1w TSE-VFA protocol based on simulation
results. Figure 2.4 shows a comparison of thin slab (10 mm) minIP images of LSAs displayed
with the same window/level settings using various imaging protocols (TR=600 or 1000 ms and
ETL=40 or 60 at 3T; TR = 1200 ms and ETL = 44 or 60 at 7T). The experimental result matched
well with EPG simulation results of Figure 2.3, resulting in the optimal TR of 1000 ms and ETL
of 40 at 3T, and the optimal TR of 1200 ms and ETL of 44 at 7T. Improved sharpness of LSAs
with the optimized T1w TSE-VFA protocol can be clearly seen. The +90°y pulse at the end of the
echo train was evaluated for the effects on CNR. As shown in Figure 2.5, the +90°y pulse further
suppresses the blood signal in the distal portions of the LSAs at the cost of also suppressing the
CSF signal. However, the improved CNR between the LSAs and surrounding tissue enabled
improved manual segmentation. The effect of ECG triggering is shown in Figure 2.6 of minIP
images of LSAs acquired without and with ECG triggering, respectively. Improved delineation of
LSAs especially the distal branches can be seen with ECG triggering at the price of increased scan
time (10-12 minutes depending on the cardiac rate). Sagittal and coronal slab orientations were
compared, and the results are shown in Figure 2.7 with the table of sequence parameters in Table
2.2. Coronal slab orientation has the potential advantage of more efficient coverage of LSAs and
reduced imaging time (7min 31sec); however, LSAs branching off the medial MCA were often
missed compared to those observed using sagittal slab orientation, probably due to reduced SNR.
Therefore, our final imaging protocol utilized sagittal slab orientation with TR=1000 ms/ETL=44
and TR=1200 ms/ETL=40 at 3T and 7T, respectively. ECG triggering was ultimately not
employed due to prolonged scan time and potential variation in LSA contrast due to individual
differences in cardiac rate.
26

Figure 2.6 Thin 10mm minimum intensity projection images of a pilot subject scanned
with the +90°y Magnetization Restore Pulse (left) and without the the +90°y
Magnetization Restore Pulse (right). The magnetization restore pulse further
suppresses the signal in the distal portions of the LSAs (red arrows) while also
suppressing the CSF signal. However, the improved CNR of vessel and tissue enables
improved manual segmentation.
Figure 2.5 Thin 10mm minimum intensity projection images of a pilot
subject without (left) and with ECG triggering (right). Despite the improved
delineation of LSAs at distal regions, the prolonged scan time was a strong
limitation.
Figure 2.7 Thin 10mm minimum intensity projection images of a pilot
subject scanned using a sagittal acquisition (left) and a coronal acquisition
(right). While the coronal acquisition offers more efficient coverage of LSAs
and reduced imaging time, LSAs branching off the medial portion of MCA
were often missed (red arrow).
27

Table 2.2 Summary of Sagittal vs. Coronal TSE-VFA Parameters

Sagittal Acq. Coronal Acq.
Bandwidth (Hz/pixel)
360 360
Resolution (mm
3
)
0.51 x 0.51 x 0.64 0.5 x 0.5 x 0.5
TE/TR (ms)
12/1000 12/1000
Orientation
Sagittal Coronal
FOV Read
230 230
FOV Phase
84.4% 73.7%
Turbo Factor
44 44
Echo train duration (ms)
162 162
Slices
160 160
Slice Oversampling
1.1 1.1
Y Partial Fourier
0.61 0.61
Z Partial Fourier
0.78 0.78
Accel. Factor PE
2 2
Echo Spacing (ms)
5.78 5.78
Acquisition Time
8:39 7:31

28
2.4.3. Evaluation of T1w TSE-VFA

Figure 2.8 Coronal 10 mm thin slice minimum intensity projections of both young and aged subject TSE-VFA scans
at 3.0 Tesla (top row) and 7.0 Tesla (middle row). The bottom row shows coronal 10 mm thin slice maximum intensity
projection of 7T TOF MRA. TSE-VFA can seemingly resolve more LSAs than 7T TOF MRA, especially for the LSAs
located in the medial group along the middle cerebral artery (white arrows).
Figure 2.8 shows thin slab minIP images of LSAs from two representative young and two
aged subjects using T1w TSE-VFA at 3T and 7T, as well as corresponding maximum intensity
projection (MIP) images using TOF MRA at 7T. Qualitatively, LSAs are more clearly delineated
with increased field strength, particularly in the distal portions of the vessels. T1w VFA-TSE at
both 3T and 7T are able to resolve more LSAs than 7T TOF MRA (p<0.05), especially the LSAs
located in the medial portion of the MCA (Figure 2.8, white arrows). As summarized in the bar
plot of Figure 2.9, the CNR between LSAs and surrounding tissue are comparable between images
acquired at 7T versus 3T for mid-length and distal portions of the LSAs. Compared to TOF MRA
at 7T, the CNR of the stems of the LSAs originating from the lateral portion of the MCA in young
subjects was significantly lower using T1w TSE-VFA (3T, p=0.045; 7T, p = 0.034). However, the
CNR of the stems of the LSAs originating from the medial portion of the MCA in both young and
29
aged subjects was significantly higher using T1w TSE-VFA compared to 7T TOF MRA (young
3T, p= 0.001; young 7T, p=0.001; aged 3T, p=0.003; aged 7T, p<0.001).

2.5. Discussion
2.5.1. Clinical Value of LSA Imaging
In this study, we presented high resolution (isotropic ~0.5mm), black blood T1w TSE-VFA
for the visualization and characterization of LSAs at both 3 and 7T. To date, very few techniques
are available for in vivo imaging of LSAs. Digital subtraction angiography (DSA) and X-ray
computed tomography angiography (CTA) have been applied for the characterization of LSAs in
clinical populations (Gotoh et al., 2012; Kammerer, Mueller-Eschner, Berkefeld, & Tritt, 2017).
However, DSA is an invasive procedure and both DSA and CTA involve radiation exposure,
Figure 2.9 CNR measures between blood and WM background in 3T VFA-TSE, 7T VFA-TSE, and 7T
TOF images for different portions of the LSAs for age 19-35 years and age > 60 years respectively.
The image intensity values were obtained by plotting the profiles across the regional planes (inset)
and taking the mean of the top quartile as the tissue background signal and the mean of the bottom
quartile of the signal plot profile as the vessel blood signal. Standard deviation of the noise was
acquired from a region of interest (dotted circle) in tissue with relatively uniform contrast. * indicates
significance (p < 0.05), ** indicates significance (p ≤ 0.001).
30
precluding their use in healthy and/or preclinical populations. The sensitivity of DSA and CTA for
visualizing LSAs is moderate with reported number of LSAs in the range of 1-7 (Gotoh et al.,
2012). In contrast, the presented black blood MRI with T1w TSE-VFA is not only noninvasive,
but also more sensitive in delineating LSAs with detected number of vessels (6-18) consistent with
textbook descriptions of LSAs (Marinković et al., 2001; Umansky et al., 1985). Our technique may
be useful for the clinical diagnosis of subcortical lacunar stroke and hemorrhages and may be
applied as a screening tool for LSA (micro)aneurysms in healthy populations.
2.5.2. T1w TSE VFA and 7T TOF MRA
TOF MRA at 7T has been proposed for visualizing and quantifying LSAs, which was
considered impractical at lower field strengths (Cho et al., 2008). In this study, we presented
optimized imaging protocols of T1w TSE-VFA at both 3T and 7T for visualizing LSAs. Compared
to the reference standard of 7T TOF MRA, T1w TSE-VFA at both 3T and 7T were able to detect
more primary stems of LSAs, although 7T TOF MRA enabled longer segmentation of the few
LSAs that were detected. The CNR of 7T TOF MRA was greater than that of T1w TSE-VFA in
relatively thick LSAs, however the saturation effect of TOF MRA on slow flowing spins led to
compromised delineation of smaller LSAs. In contrast, our EPG simulation showed that small
arteries with flow velocity in the range of 4.5-8.2 cm/s (Bouvy et al., 2016; Schnerr et al., 2017)
can be reliably visualized by T1w TSE-VFA due to combined effects of longer T1/T2 values of
arterial blood and flow induced phase dispersion during TSE readout at both 3T and 7T. These
simulation results were further verified by in vivo experimental data, suggesting T1w TSE-VFA
may outperform 7T TOF MRA for visualizing (small) medial LSAs at both clinical field strength
and ultrahigh magnetic field. Black blood MRI also allows simultaneous assessment of vessel wall
and parenchymal lesions. It is worth noting that the TOF MRA protocol in our experiment was not
31
as fully optimized as that of TSE-VFA, so the comparison results between the two techniques may
need to be interpreted with caution.
For black blood MRI with T1w TSE-VFA, the CNR of LSAs was comparable at 3T with
that at 7T, although significantly more LSA secondary branches could be delineated at 7T. The
spatial resolution of T1w TSE-VFA was slightly lower at 3T (voxel size = 0.51x0.51x0.64mm3)
than 7T (0.5x0.5x0.5mm3) in the present study. The reduced T2 blurring and sharper PSF at 7T
enabled easier identification of LSAs with improved delineation of distal portions of the vessels.
For automated segmentation of LSAs, the sharpness and improved SNR at 7T may be more
advantageous. In addition, the 7T Terra MR system used in the present study is the first FDA
approved ultrahigh field MR system. All developed MR pulse sequences and imaging protocol can
therefore be directly translated to clinical imaging.
2.5.3. Aging Effects on LSAs
In the present study, significant differences between the two age groups were found for the
number of detected LSAs. These results are consistent with recent findings of a reduced number
of perforators in the basal ganglia of patients with lacunar infarction compared to matched controls
(Geurts Lennart, Zwanenburg Jaco, Klijn Catharina, Luijten Peter, & Biessels Geert, 2019). A
recent phase-contrast (PC) MRI study assessed flow pulsatility of LSAs at 7T and found increased
pulsatility index and reduced damping factor of LSA blood flow in aged subjects, suggesting
increased stiffness and reduced vascular compliance of LSAs with aging (Schnerr et al., 2017).
The reported mean (minimum–maximum) flow velocity was 8.2 (6.2-10.1) cm/s and 4.5 (2.9-6.1)
cm/s in young and aged subjects, and the number of LSAs that can be reliably identified for flow
measurement was 3-7 and 3-4 in young and aged subjects, respectively. Our data of reduced vessel
number and length in aged subjects are highly consistent with the PC MRI study, suggesting that
32
reduced flow velocity and pulsatility index may contribute to impaired visualization of LSAs in
T1w TSE-VFA scans.
Due to the relatively long acquisition time, a limitation of the black blood MRI sequence
is the possibility for motion, especially in the less compliant aged subjects. Motion artifacts such
as blurring and ghosting were observed in the TSE-VFA images of the excluded subjects who
failed to complete the MRI scans. Motion artifacts can also affect the manual segmentation in aged
subjects, since even a slight shift of the head can affect the visualization of the 80-1400 µm
diameter LSAs. Due to the use of packing in the head coil and applying paper tape across the
forehead skin for tactile feedback, the black blood MRI data included in our analyses were free of
visible motion or other artifacts.
2.5.4. Limitations of the Study

Figure 2.10 (A) Co-registered 7T 10mm thin slice intensity projection images from a 40-year-old female participant
(left to right: 10mm minimum intensity projection (minIP) of T1w TSE-VFA and 10mm maximum intensity
projections (MIP) of T2w TSE-VFA and TOF, respectively). (B) Raw images of 7T T1w TSE-VFA without and with
overlays from bright TOF (red) and T2w TSE-VFA (yellow) signals, which indicate blood vessels and cerebrospinal
fluid in perivascular spaces, respectively. Although CSF also appears dark in T1w TSE-VFA thereby enlarging the
apparent thickness of the LSAs, the perivascular spaces filled with CSF appear more prominently toward the middle
rather than the stem of the LSAs. There are also LSA branches/stems that can only be visualized in T1w TSE-VFA
but not in T2w TSE-VFA or TOF images.
33
There are several limitations of this study. First, T1w TSE-VFA also suppresses CSF due
to its long T1/T2 values; therefore, the delineated LSAs may include perivascular space (PVS).
However, we have compared T1w TSE-VFA and T2w TSE-VFA (Figure 2.10), which delineates
PVS; and the locations of LSAs in black blood MRI did not completely match those of PVS in the
basal ganglia area. Another possibility is to combine complementary flow suppression techniques
like DANTE (Viessmann, Li, Benjamin, & Jezzard, 2017; Xie et al., 2016) with the T1w VFA-
TSE sequence to further suppress the signal of moving spins in arterial blood and improve black-
blood contrast. Second, the image contrast of LSAs may be affected by B1 inhomogeneity,
especially at 7T. Nevertheless, the central location of LSAs at the “bright spot” of B1+ field due
to dielectric effects at 7T is favorable for enhancing the CNR of LSAs, based on our simulation
(Figure 2.11). The region of interest for estimating the noise was taken from a relatively uniform
white matter region within this “bright spot”, although the number of voxels in this region may not
be as large as ideally required for the estimation of the standard deviation (Greenwood &
Sandomire, 1950; Kellman & McVeigh, 2005). With the acquisition of an 8Tx/32Rx head coil
(Nova Medical, Wilmington, MA) at our institute, we also determined that parallel transmission
B1 shimming can significantly improve B1+ field homogeneity for future implementation of T1w
TSE-VFA (Figure 2.11). Third, the scan time of T1w TSE-VFA was relatively long (8:39 min at
3T and 10:05 min at 7T) making black blood MRI susceptible to motion artifacts. In order to
reduce scan time, we have attempted zoomed TSE (S. J. Ma et al., 2017) and accelerated TSE with
2D CAIPI acquisition (Fritz et al., 2016). However, these two techniques caused reduced SNR and
CNR, hindering the delineation of LSAs. In the future, prospective motion correction techniques
using optical and/or RF tracking (Callaghan et al., 2015) may be applied in conjunction with sparse
sampling techniques such as compressed sensing. Due to SAR limitations at 7T, the imaging
34
protocols and resolutions of TSE-VFA were not identical at 3 and 7T. However, they were
developed based on the best clinical practice at each field strength.

Figure 2.11 The effect of B1 variation on contrast between arterial blood and background white matter (A). Contrast
varies approximately linearly with the magnitude of the B1 field, and this effect is larger at 7T (red) compared to 3T
(blue). However, the LSAs are located in the central “bright spot” of the B1+ field due to standing wave shading
artifacts (dielectric effects) at 7T, which is favorable for enhancing the CNR of LSAs. This issue can also be addressed
with the use of pTx B1 shimming, which reduces the dielectric effects toward the base of the brain (B).
2.6. Conclusion
High-resolution black-blood 3D T1w TSE-VFA sequence offers a new method for the
visualization and quantification of LSAs at both 3T and 7T, which may serve as a promising
imaging marker of pathological conditions related to damaged LSAs.

35

3. CHAPTER 3: Vessel Segmentation and Morphology Metrics: A Deep Learning
Application
3.1. Abstract
Objectives
The morphology of the lenticulostriate arteries (LSAs) can provide insight into the degenerative
processes of SVD; however, these vessels are difficult to visualize and segment from clinical MRI
images at 3T. In chapter 2, we proposed a “black-blood” MRI technique to visualize LSAs with
sub-millimeter spatial resolution using 3D turbo spin echo with variable flip angles (T1w-VFA-
TSE) at standard clinical field strength of 3T(Samantha J. Ma, Sarabi, et al., 2019). In this study,
we manually delineated the LSAs in 3D to characterize the small vessel morphology at 3T and 7T.
Using the manual segmentation as supervision, we developed and evaluated a deep learning (DL)-
based algorithm to semi-automatically segment the LSAs from the 3D T1w-VFA-TSE (black
blood) images acquired at 3T.
Materials and Methods
36
Seventeen healthy volunteers (12 under 35 years old, 5 over 60 years old) were imaged with the
previously optimized T1-weighted 3D turbo spin-echo with variable flip angles (T1w TSE-VFA)
at both 3T and 7T. LSAs were manually segmented using ITK-SNAP. Automated Reeb graph
shape analysis was performed to extract features including vessel length and tortuosity. All
quantitative metrics were compared between the two field strengths and two age groups using
ANOVA. The manual segmentations from the healthy volunteers were then combined with manual
segmentations from 11 older patients (67.5 ±4.4 years old) from a vascular risk cohort to form a
dataset for automatic segmentation. The DL models were evaluated compared to optimally
oriented flux segmentation.
Results
LSAs can be clearly delineated using optimized 3D T1w TSE-VFA at 3T and 7T, and a greater
number of LSA branches can be detected compared to those by time-of-flight MR angiography
(TOF MRA) at 7T. The mean vessel length and tortuosity were greater on TOF MRA compared
to TSE-VFA. The number of detected LSAs by both TSE-VFA and TOF MRA was significantly
reduced in aged subjects, while the mean vessel length measured on 7T TSE-VFA showed
significant difference between the two age groups. The visual analysis of the three automatic
segmentation methods (U-Net, HighRes3DNet, and OOF) demonstrated that it is feasible to
perform DL-based automatic segmentation of the LSAs using the optimized black blood MR
images. HighRes3DNet had the best performance with respect to similarity with the manual
segmentation label, with average Dice coefficient ~0.54, average 95HD ~27.86 voxels, and AVD
~1.79 voxels, respectively.
Conclusion
37
Using manual segmentations as supervision, 3D segmentation of LSAs is demonstrated to be
feasible with relatively high performance and can serve as a potentially effective tool to aid
quantitative morphometric analyses in patients with cerebral small vessel disease.
3.2. Introduction
Cerebral small vessels are largely inaccessible to existing clinical in vivo imaging
technologies. As such, early cerebral microvascular morphological changes in small vessel disease
(SVD) are difficult to evaluate. In previous studies, high resolution time-of-flight MR angiography
(TOF MRA) at ultrahigh magnetic field of 7T has been applied for the characterization of
lenticulostriate arteries (LSAs) (Cho et al., 2008; Hendrikse et al., 2008). The morphology of these
LSAs (e.g. branch number, radius, tortuosity) showed significant differences between subjects
with subcortical vascular dementia (SVaD) and healthy controls (Seo et al., 2012) as well as
between hypertensive and normotensive subjects (C. K. Kang et al., 2009). These data demonstrate
the potential for quantifying the morphology of LSAs as imaging biomarkers of hypertensive small
vessel disease, which primarily affects subcortical regions.
Since ultra-high field MRI is not commonly available in the clinical setting, the purpose of
this study was to utilize our previously optimized high resolution (isotropic ~0.5mm) 3D T1-
weighted TSE-VFA acquired at 3 and 7T to segment the LSAs and characterize the morphological
changes that may occur with age. Quantitative metrics of LSA morphology were derived by Reeb
graph analysis of segmented LSA shapes which were compared between the two age groups and
two field strengths, respectively. Unlike previous morphology analyses of LSAs that used two
dimensional maximum intensity projections from TOF MRA at 7T (C.-K. Kang et al., 2009) or
two dimensional minimum intensity projections from TSE-VFA at 3T (Zhang et al., 2019), we
implemented a novel three-dimensional shape analysis of LSAs derived from T1-weighted TSE-
38
VFA black-blood images to quantify the morphology of the vessels.
In recent years, machine learning (ML) methods have been applied to aid diagnoses (Asadi,
Dowling, Yan, & Mitchell, 2014; van Os et al., 2018) or automatically segment structures in
medical images (Livne et al., 2019). Machine learning models use algorithms to parse data, learn
features from that data, and make informed decisions based on the learning with some user
guidance. Deep learning (DL) is an advanced ML method that takes the intelligent decision making
one step further by using the neural network architecture, which resembles human visual
perception (LeCun et al., 2015). Deep learning is able to capture the hierarchical and complex
features of an input image automatically and can identify, classify, and quantify patterns in medical
images (Shen, Wu, & Suk, 2017; Zaharchuk et al., 2018). In the second part of this study, we
developed and evaluated a DL-based algorithm to demonstrate the feasibility of automatically
segmenting the LSAs in black blood MR images, using the manual segmentation as supervision.
3.3. Materials and Methods
3.3.1. Subjects
To evaluate the ability to observe age-related differences, the volunteer cohort from the
previous study described in Chapter 2 was imaged with the previously optimized T1-weighted 3D
turbo spin-echo with variable flip angles (T1w TSE-VFA) at both 3T and 7T. This cohort included
a total of 17 healthy volunteers, with 12 participants (7 male, 27 ± 3.5 years) between 19- 35 years
old and 5 participants (2 male, 64.2 ± 1.9 years) more than 60 years of age, herein referred to as
the young and aged group, respectively. All participants provided written informed consent
following a protocol approved by the Institutional Review Board (IRB) of the University of
Southern California.
39
In order to develop the automatic DL-based segmentation, 3 Tesla T1w TSE-VFA images
from 11 patients in the Los Angeles Latino Eye Study cohort (1 male, 67.5 ±4.4 years old) were
also acquired. The 3T black blood MRI images collected from this total of 28 participants formed
the training dataset for DL model development.
3.3.2. MRI Experiment
All MRI scans were performed on a Siemens 3T Prisma scanner with a product 32-channel
head receive coil and body transmit coil, and on a Siemens 7T Terra scanner (Erlangen, Germany)
with a single-channel transmit/32-channel receive (1Tx/32Rx) head coil (Nova Medical,
Wilmington, MA, USA). The scan protocol and sequence parameters are described in Chapter 2,
Table 2.1.
3.3.3. Image Analysis: Vessel Segmentation and Morphology Metrics

Figure 3.1 Lenticulostriate artery morphological quantification workflow begins with manual vessel segmentation
using ITK-SNAP on the raw TSE-VFA image (a). The vessel volumes are reconstructed, and a mesh surface is created
in preparation for shape analysis (b). Quantitative measures such as vessel length and tortuosity are calculated from
the Reeb graph (c).

We developed a three-step method to extract quantitative 3D measurements of LSAs for
both TSE-VFA and TOF MRA (Figure 3.1). As the first step, manual vessel segmentation was
40
performed by an experienced neuroradiologist (Y.C.) using ITK-SNAP (Yushkevich et al., 2006).
To increase the accuracy of vessel masks, raw images of all axial, coronal, and sagittal views were
used in ITK-SNAP as shown in Figure 3.1a. However, geometric and topological outliers occur
frequently during manual segmentation due to limited image resolution or anatomical variability
across subjects. As the second step to perform quantitative shape analysis on these data, we applied
the surface reconstruction method (Y. Shi et al., 2010) that uses the Laplace-Beltrami (LB)
spectrum for outlier removal without shrinkage (Figure 3.1b). The LB spectrum can be viewed as
a generalization of the Fourier basis onto general surfaces. Building upon this intuitive
understanding, the mesh reconstruction method (Y. Shi et al., 2010) iteratively projects the mask
boundary onto a subspace of low frequency LB eigenfunctions and removes outliers with large
changes during the projection process. As shown in Figure 3.1, our method successfully obtained
a smooth surface for each vessel (Figure 3.1b, vessel surface reconstruction). This fully automatic
and iterative method allows the generation of a smooth surface representation of LSAs without
shrinking other parts of the mask, and it ensures that all surfaces have a consistent genus-zero
topology, which means all surfaces have no handle or hole in the reconstructed mesh (Xiao,
Chenyang, & Prince, 2003).
As the last step, the Reeb graph analysis method (Yonggang Shi et al., 2008) was applied
to model and extract geometrical measurements of 3D vessel shapes such as vessel length and
tortuosity for 3T and 7T images, respectively (Figure 3.1c). The Reeb graph is an abstract graph
describing the neighboring relation of the level contours of a function defined on a surface.
Following our work on brain surfaces (Y. Shi, Lai, Toga, & Alzheimer's Disease Neuroimaging,
2013), we used the first non-constant LB eigenfunction as the feature function f for Reeb analysis.
More specifically, we sampled 25 evenly spaced level contours of f as plotted on the vessel surface
41
in Figure 3.1c. The centroid of each contour was used to explicitly represent the graph nodes. For
the purpose of measuring the longest branch in each vessel, we assumed that the Reeb graph
followed a chain topology, where each node on the graph only had up to two neighbors. To
estimate the vessel length (Figure 3.1c), we extracted the center line or medial core of the 3D
vessel surface by connecting the centroids of level contours, then summed up the distances of
every two adjacent centroids. For vessel tortuosity, we used the common distance metric measure,
which provides a ratio of the estimated vessel length to the Euclidean distance between the two
end points of the curve. Vessel thickness could also be estimated; however, we chose not to report
the results due to potential errors of small vessel size on the order of one voxel.
3.3.4. Statistical Analysis
Statistical analysis was performed with STATA 13.1 (College Station, Texas). Morphological
metrics (vessel length, tortuosity) were subject to Repeated-Measures ANOVA to evaluate the
effect of age group, gender, field strength, and hemisphere. The total branch number was also
compared across the three techniques (T1w TSE-VFA at 3T and 7T, and TOF MRA at 7T) using
within-subject ANOVA. A p-value ≤ 0.05 (two-sided) was considered statistically significant.
42
3.3.5. Deep Learning Model Development
A flowchart of the deep learning experiment is shown in Figure 3.2, including modules
of pre-processing, data input, DL model architecture, and evaluation.
Image Preprocessing
To prepare for input into the automated segmentation models, the raw images underwent
several pre-processing steps. First, the images were denoised via non-local means filtering
(Manjón et al., 2008). The filtered images were then cropped to a volume encompassing the LSAs
and separated by left and right hemispheres to avoid the ventricular structures for a total of 56
image volumes. To improve the specificity of the training, an LSA regional mask was created by
dilating the manual segmentation labels and taking the common covered region as the mask. The
dataset was divided into a training set with 21 subjects (42 volumes) and a test set with 7 subjects
(14 volumes).
Network Architecture and Training
Based on our previous success using this model in an application of penumbral tissue
segmentation in acute ischemic stroke (K. Wang et al., 2020), we applied the HighRes3DNet
Figure 3.2 Flowchart of the pre-processing, data input, network training, and evaluation of deep learning models.
Black blood images were used for manual segmentation in ITK-SNAP, which served as supervision. The images were
cropped to the subcortical region of interest, underwent non-local means filtering, and split into hemispheres to
increase the sample size. The models with 3D U-net and Highres3DNet were trained with the training set and then
evaluated on the test set (n=14 hemispheres from 7 subjects) relative to OOF performance. ReLU = rectified Linear
Unit.
43
network (Wenqi Li et al., 2017), which offers a compact end-to-end 3D convolutional neural
network structure that maintains high-resolution multi-scale features. The HighRes3DNet
architecture included 20 trainable layers with dilated convolution and dilating factors of 1, 2, and
4, respectively. Residual connections were employed for every two convolutional layers. The
architecture was adapted from and trained within the NiftyNet (Gibson et al., 2018) platform on 2
Nvidia GeForce GTX 1080 Ti GPUs. Black blood images and the LSA regional masks were used
as input, and manual segmentation labels served as the supervision. 48x48x48 volumes (batch
size=4) were randomly extracted from the 3D preprocessed images for training. Volume-level
augmentation was utilized, including rotation with a random angle in the range of [-10°, 10°] for
each of the three orthogonal planes and random spatial rescaling with a scaling factor in the range
of [0.9,1.1]. The training process was performed with 40,000 iterations to enable the training
process to reach steady state, with Dice loss (Milletari, Navab, & Ahmadi, 2016) as the loss
function and the Adam optimizer (Kingma & Ba, 2014) for computing graph gradients. The
hyperparameters of the network (eg, layer number, volume size, and iteration number) were
determined based on pretraining on a fixed validation data set.
For comparison, 3D U-Net (Çiçek et al., 2016) with comparable configuration parameters
was also tested through the NiftyNet platform. The 3D U-Net architecture is a popular specialized
convolutional neural network with an encoding down-sampling path and an upsampling decoding
path that was purposely designed for general segmentation tasks in biomedical images. In addition,
vessel segmentation was performed in MATLAB (Mathworks, Natick, MA) using 3D optimally-
oriented flux (OOF) (M. Law & A. Chung, 2008) filtering, which relies on image gradients to
estimate local vessel orientations.
Model Performance Assessment
44
The model performance was assessed on an inference dataset that was not seen by the DL
models. The vessel probability maps of the inference images that were output from the model were
binarized by thresholding at the standard of 0.5. The binary vessel segmentation was then
evaluated using three different measures: Dice coefficient, 95% percentile Hausdorff distance
(95HD), and average Hausdorff distance (AVD) (Livne et al., 2019). The 95HD and AVD metrics
are resistant to outliers and represent the longest distances in voxels between two segmentation
results, hence, smaller values indicate better performance. Dice coefficient ranges from [0,1]
unitless values where better performance is indicated by larger values. The metrics were calculated
for OOF, U-Net, and HighRed3DNet using MATLAB and the EvaluateSegmentation software
(Taha & Hanbury, 2015), respectively.
3.4. Results
3.4.1. Evaluation of T1w TSE-VFA Manual Segmentation and Age
Figure 3.3 shows 3D renderings of manually segmented left LSAs on T1w TSE-VFA
images at both 3T and 7T. Each primary vessel is assigned a different color label, with shorter
secondary branches being given separate labels from the primary vessel. The total bilateral number
of primary vessels counted using T1w TSE-VFA at 3T (mean ± SD: 11.00 ± 3.02) and 7T (11.65
± 2.55) was significantly higher than that by TOF MRA at 7T (6.47 ± 1.68, p=0.0001).
Additionally, significantly more secondary branches could be detected in 7T T1w TSE-VFA (3.75
± 1.60, p=0.0075) compared to that at 3T (2.25 ± 1.71). Figure 3.4 shows a few more examples
of the 3D renderings where more branching is detected with 7T imaging. Metrics of vessel
morphology including vessel length and tortuosity were quantified by Reeb graph analyses and
subjected to ANOVA (see Table 2 for mean ± SD values of the metrics at 3T and 7T in two age
groups, respectively). The effect of hemisphere and gender were not significant for both metrics.
45
The effect of age group was significant for mean vessel length (p=0.0257), which was reduced in
aged subjects for 7T TSE-VFA. The mean tortuosity was increased in aged subjects but the effect
was not significant (p=0.25 for 7T TOF), as shown in the boxplots in Figure 3.5.

Figure 3.3 Three-dimensional renderings of left LSAs from the manual segmentations of an aged subject (A,C,E) and
a young subject (B,D,F). VFA-TSE at both 3T and 7T enabled the identification of more vessels in aged subjects,
especially in the medial region. 7T VFA-TSE enabled the identification of more branches in general in the younger
subjects (i.e. orange and turquoise branch vessels in D). Despite longer segmentation of vessels using 7T TOF images,
fewer vessels could be identified. As summarized by the box plot of vessel count, significantly more vessels were
detected in young subjects than in aged subjects (**, p<0.01), and more vessels were detected using VFA-TSE
compared to 7T TOF MRA (*, p<0.05).
46

Figure 3.4 Additional examples of the 3D renderings of the
manual vessel segmentation for two younger subjects with larger
number of vessels and branches detected. Despite the high vessel
density, the resolution was sufficient to distinguish secondary
branches from primary stems.
Table 3.1 Summary of Reeb graph metrics for 3T and 7T VFA-TSE manual segmentations of young and aged subjects
(mean ± SD)
47

Figure 3.5 Comparison of mean vessel length and tortuosity of lenticulostriate arteries in young (age 19-35 years)
and aged (age > 60 years) groups for each modality. * indicates significance p<0.05; ** indicates significance
p<0.01.

3.4.2. Deep Learning Segmentation of LSAs
Figures 3.6 and 3.7 show examples of the 3D projections of the segmentation results for
each method. The visual analysis of the three methods demonstrated that it is feasible to perform
automatic segmentation of the LSAs using the optimized black blood MR images. However, as
seen in Figure 3.7, the false positives appear to be true continuations of the LSAs. In this case,
despite the low performance, all three segmentation methods seem to detect distal portions of the
vessels that the human eye may have missed during the manual segmentation. For OOF, 3D U-
Net, and HighRes3DNet, the average Dice coefficient was 0.27±0.10, 0.30±0.16, and 0.54±0.07,
respectively. The average 95HD was 36.05±11.83, 38.05±10.37, and 27.86±8.94 voxels,
respectively. The AVD was 5.14±1.93, 5.64±2.33, and 1.79±0.80 voxels, respectively.

48

Figure 3.6 3D projections of segmentation results using each method. The figure illustrates an exemplary result for
one subject. Labels are shown in the first column, and segmentation by OOF, 3D U-Net, and HighRes3DNet in the
top row of the other three columns, respectively. The bottom row shows the error maps, where red voxels indicate
true positives, green voxels false positives, and blue voxels false negatives. Overall, HighRes3DNet achieved the
highest performance. (3D interpolations may not translate real voxel-to-voxel differences.)
Figure 3.7 3D projections of segmentation results using each method. This figure illustrates the tendency for manual
segmentation to miss the distal portion of the LSAs that deep learning and OOF can seemingly detect.
49
In Figure 3.8, boxplots showing the distribution of each metric are plotted for each method.
HighRes3DNet shows superior performance (p<0.01 for Dice and AHD, p<0.05 for 95HD) in
terms of segmenting the LSAs in reference to manual segmentation.
3.5. Discussion
3.5.1. Aging Effects on LSAs
In this study, the two age groups exhibited differences regarding the morphology of the
detected LSAs. In addition to finding significantly fewer LSAs in the aged group, there was a trend
of decreased mean vessel length and increased mean tortuosity in the aged group. Our data is
consistent with past studies quantifying the morphology of LSAs showing significant differences
between subjects with subcortical vascular dementia (SVaD) and healthy controls (Seo et al., 2012)
as well as between hypertensive and normotensive subjects (C. K. Kang et al., 2009). With the
presence of vascular risk factors such as diabetes, hypertension, and hyperlipidemia, the LSA
metrics such as vessel number and length may further decrease compared to healthy controls.
However, increased LSA tortuosity has been reported in vascular dementia compared to age
matched controls (Seo et al., 2012). While this trend of increased tortuosity of LSAs in aged
subjects was observed in the 7T TOF scans of our study, there was no significant difference in
Figure 3.8 Boxplots of the performance metrics for each segmentation method. Dice similarity is measured in a range
[0,1], while 95% Hausdorff distance and average Hausdorff distance are measures in voxels. HighRes3DNet
produced significantly superior performance for segmentation results relative to the manual segmentation label
compared with OOF (p<0.01 for Dice and AHD) and 3D U-Net (p<0.05 for 95HD). * indicates p<0.05, ** indicates
p<0.01.
50
tortuosity when evaluated using TSE-VFA images at 3 or 7T. As the morphologic metrics in our
study were derived from manually segmented LSAs, it is possible that the distal portions of LSAs
were not well-detected by the human eye. As a result, tortuosity was mainly quantified in larger
LSA stems, leading to similar tortuosity in aged subjects (tortuosity is defined as the ratio between
the actual path length divided by the linear distance). It is also possible that tortuosity of LSAs
may increase and subsequently contribute to altered blood brain barrier permeability at the distal
ends of these perforators in aged subjects at risk of developing cerebral small vessel disease and
dementia compared to age matched controls (Shao et al., 2019). This hypothesis is being tested in
our lab on a cohort of aged subjects.
3.5.2. Manual Segmentation and Shape Quantification
Using manual segmentation and automated surface modeling, we performed fully 3D
analyses of the LSA morphometry in this study. The vessel counts observed for each modality
using manual segmentation in this study align with the findings of number of stems (two to ten)
and tortuosity observed in previous 7T TOF studies (C.-K. Kang et al., 2009; C. K. Kang et al.,
2009; Seo et al., 2012). The surface reconstruction and Reeb graph analysis methods provided a
consistent representation for the comparison of vessel geometry across subjects and vessel
branches. Significant differences were detected between age groups based on the measures from
Reeb graph modeling. While we used manually delineated masks for our quantification, the 3D
shape analysis tools can be directly applied to automated segmentation results when they become
available. This will be a main direction for our future research. The manually segmented masks
form a reasonable training data for supervised learning algorithms if we consider that there are
typically 10+ vessel branches from each subject. For medical image segmentation, patch-based
deep learning segmentation algorithms have achieved success in various tasks (Kamnitsas et al.,
51
2017; Wachinger, Reuter, & Klein, 2018). We will leverage these frameworks for our research on
automated LSA segmentation as they fit very well with the vessel segmentation problem, which is
inherently based on contrasts in a local neighborhood.
3.5.3. Automatic DL Segmentation
As demonstrated in this study, machine learning and in particular DL algorithms are ideally
suited for vessel segmentation problems. Although filtering methods such as Frangi vesselness
filters (Frangi, Niessen, Vincken, & Viergever, 1998) or optimally oriented flux (OOF) (M. W.
Law & A. C. Chung, 2008) can capture vessel-like structures with high sensitivity, these filters are
often indiscriminate to other structures such as perivascular spaces or ventricles, which also would
appear somewhat dark in the T1w TSE-VFA images. The Dice similarity, 95 percentile Hausdorff
distance, and Average Hausdorff distance metrics used to evaluate the performance in reference
to manual segmentation in this study should be interpreted with caution because the manual
segmentation process is still limited to human reading of the images. Based on the OOF results,
filtering methods are a sensitive, feasible approach to initiate the manual segmentation process or
even serve as pre-training for deep learning. With further hyperparameter optimization, the
HighRes3DNet model is a promising method for specific LSA segmentation in black blood
images.
3.5.4. Limitations
There are several limitations in this study. The manual segmentation of LSAs is time
consuming and subjective. Depending on human observation and the limitation of the human eye,
some very distal portions of the vessels may be missed in the segmentation process. Additionally,
vessels may be over-painted in ITK-SNAP, resulting in vessels that appear thicker than their actual
morphology. Since the images likely exhibit some partial volume effects, the delineation of very
52
thin segments of the vessels may be difficult to the human eye. However, as there is no gold
standard for delineating the LSAs on T1w TSE-VFA images, manual segmentation is the best
possible label that could be used as supervision for deep learning model training. Another
limitation is the number of data samples with completed segmentations. In addition to manually
separating the two hemispheres into separate images, data augmentation was employed to increase
performance. In the future, with fine tuning of the application of OOF or Frangi vesselness filters,
a rough segmentation can be performed to possible pre-train the network with more samples. As
this is a prospective ongoing study, more data will be collected and included in the training dataset.
3.6. Conclusion
The segmentation of LSAs in high-resolution black-blood 3D T1w TSE-VFA may serve
as a new method for the direct visualization and quantification of small vessel morphological
changes with age. Considering that automatic segmentation is highly feasible, evaluating the shape
and number of LSAs can be a promising imaging marker of pathological conditions related to
damaged LSAs.

53
4. CHAPTER 4: High Resolution 3D T1-weighted Black Blood MRI of Human
Lenticulostriate Arteries as an Imaging Biomarker for Vascular Cognitive Impairment
and Dementia
4.1. Abstract
Objectives
The clinical differentiation of Alzheimer’s disease (AD) from vascular cognitive impairment and
dementia (VCID) due to cerebral small vessel disease (SVD) is blurred (Gorelick et al., 2011;
Schneider et al., 2007). A new high-resolution black-blood MRI technique has emerged for sub-
millimeter spatial resolution using 3D turbo spin echo with variable flip angles (T1w-VFA-
TSE)(Z. Fan et al., 2016; Qiao et al., 2011; Qiao et al., 2014). In this study, this 3D TSE sequence
was applied for the visualization of perforating lenticulostriate arteries (LSA) to explore early
microvascular changes related to SVD.
Materials and Methods
52 volunteers (13 male, 69±7 years) from the Los Angeles Latino Eye Study (LALES) cohort were
scanned on a Siemens 3T Prisma scanner using a 20-channel head coil. A 3D T1w TSE sequence
with VFA was chosen for small vessel imaging with its inherent excellent contrast between vessels
and tissue due to good blood suppression properties. All subjects were evaluated for diabetes,
hypertension, and hyperlipidemia risk. The NIH Toolbox evaluated cognition and motor abilities.
The Global Clinical Dementia Rating (CDR) was evaluated for 16 subjects. To evaluate the LSA
quality in the T1w-TSE-VFA images, a 4-point rating scale was developed and evaluated by two
independent raters. The average LSA delineation (LSAD) rating was correlated pairwise with the
patient information, the NIH Toolbox scores, and Global CDR. Test-retest repeatability was
assessed by intraclass correlation coefficient (ICC).
54
Results
The Cohen’s kappa was found to be 0.77, indicating decent agreement between raters for the
LSAD rating. The ICC was 0.72 for LSA ratings between repeated scans, indicating a decent level
of test-retest repeatability. The Flanker, PSMT, and PCPS standard scores were positively
correlated with LSAD rating (p<0.05), indicating stronger cognitive flexibility, executive function,
and episodic memory in subjects with less tortuous small vasculature. Lower LSAD in subjects
with Global CDR of 0.5 (p<0.001), high cholesterol (p<0.01), or diabetes (p<0.01) suggests LSA
count and tortuosity may serve as a potential imaging marker for mild cognitive impairment
(CDR=0.5).
Conclusions
Morphological features of LSAs may serve as a potential imaging marker of cerebral SVD such
as VCID.
4.2. Introduction
As therapies improve with new biotechnology, people are living longer, and the burden of
cognitive impairment in an increasingly aging population has become an important issue. While
Alzheimer’s disease (AD) is the most common cause of dementia, the contribution of vascular
factors to cognitive impairment and dementia is becoming increasingly recognized (Gorelick et al.,
2011). AD and cerebrovascular diseases share common risk factors such as hypertension, obesity,
and diabetes; and these conditions coexist in 40–50% of clinically diagnosed AD, resulting in
mixed AD-vascular dementia becoming the most common cause of cognitive impairment in the
aged (Iadecola, 2016). The clinical differentiation of AD from vascular cognitive impairment and
dementia (VCID) is blurred (Schneider et al., 2007). Cerebral small vessel disease (SVD) is the
most common vascular cause of dementia, a major contributor to mixed dementia, and the cause
55
of about one fifth of all strokes worldwide (Norrving, 2008; Pantoni, 2010). The aging population
worldwide and the increase in vascular disease with age has led to projections of major growth in
VCID over the next 30 years (Gorelick et al., 2011). However, the underlying mechanisms of small
vessel disease remain poorly understood, resulting in no specific guidelines for its prevention and
treatment.
The large knowledge gap in SVD is partly because cerebral small vessels, including
arterioles, capillaries, and venules, are inaccessible to existing in vivo imaging technologies.
Clinical diagnosis of SVD relies on conventional MRI findings including lacunar infarcts, white
matter lesions (WML), cerebral microbleeds, prominent perivascular spaces, and atrophy
(Wardlaw et al., 2009). Large epidemiological studies have shown that silent cerebral infarction
and WMLs are associated with both non-memory-related cognitive deficits (e.g. executive
function and perceptual speed) (Mayda & Decarli, 2009; Rosenberg et al., 2015), and memory
impairment (Breteler et al., 1994; Debette et al., 2010; DeCarli et al., 1995). Recent evidence
suggests that WML are a core feature of AD (Lee et al., 2016), and the progression of WML is a
better predictor of cognitive impairment than baseline WML burden (Silbert, Howieson, Dodge,
& Kaye, 2009). However, these parenchymal lesions are the consequences of SVD rather than the
surrogate markers of microvascular changes, and they are not useful for guiding early interventions
to change the course of VCID. Therefore, it is of paramount importance to identify and develop
imaging markers of early microvascular changes related to SVD for the design of future clinical
trials to prevent and treat VCID.
Arteriolosclerosis (or hypertensive SVD) is a systemic condition that affects the brain,
kidney, and retina; and is strongly associated with aging, diabetes, and hypertension.
Arteriolosclerosis is the most common form of sporadic SVD and typically affects small
56
perforating arteries of the deep gray nuclei and deep white matter (Charidimou et al., 2016). In this
study, we present an imaging-based evaluation of the lenticulostriate arteries (LSAs) and perfusion
in the middle cerebral artery (MCA) perforator territory (S.J. Ma, Yan, Jann, & Wang, 2018) as it
pertains to cognitive function and vascular risk in a cohort of elderly Latino subjects.
4.3. Materials and Methods
4.3.1. Subjects and Clinical Evaluation
A total of 55 volunteers (13 males, 69 7 years) were recruited for this study from the Los
Angeles Latino Eye Study (LALES) cohort after they provided written informed consent following
a protocol approved by the Institutional Review Board (IRB) of the University of Southern
California. The LALES participants are of Hispanic descent, and may have a history of diabetes,
hypertension, and/or hypercholesterolemia. All participants underwent a blood draw and returned
for a repeat scan approximately 6 weeks from the first visit.
Of these participants, 38 completed their interviews to determine Clinical Dementia Rating
(CDR) and history of diabetes, hyperlipidemia, and hypertension; were evaluated with
neuropsychological testing; and were tested with the NIH Toolbox Cognition module. The
neuropsychological testing included the Montreal Cognitive Assessment (MoCA), Craft Stories,
Spanish English Verbal Learning Test (SEVLT), Numbers Span Forward and Backwards, Trail
Making Test, and Digit Symbol Coding. The NIH Toolbox Cognition module includes the Flanker
Inhibitory Control and Attention Test (Flanker), the Dimensional Change Card Sort Test (DCCS),
Picture Sequence Memory Test (PSMT), and Pattern Comparison Processing Speed Test (PCPS).
23 participants additionally completed the NIH Toolbox Motor module which included the 9-Hole
Pegboard Dexterity Test (Pegboard), Standing Balance Test, 4-Meter Walk Gait Speed Test, and
2-Minute Walk Endurance Test.
57
4.3.2. MRI Protocol and Analysis
All MRI scans were performed on a Siemens 3T Prisma scanner (Siemens, Erlangen,
Germany) using a 20-channel head coil and body transmit coil. Our optimized 3D T1-weighted
turbo spin echo with variable flip angles (T1w TSE-VFA) sequence was chosen for small vessel
imaging with its inherent excellent contrast between vessels and tissue due to good blood
suppression properties. The sequence parameters are described in the methods section of Chapter
2. In addition, background suppressed 3D pseudo continuous arterial spin labeling with gradient
and spin echo (GRASE) readout was acquired with TR/TE=4300/36.76ms, label time=1.5s, post
label delay=2000ms, slices=48 , resolution=2.5mm3 isotropic, label/control pairs=8 & 1 M0
image, scan time=4:40min. Quantitative cerebral blood flow (CBF) maps were generated with
custom software using Interactive Data Language (IDL, Boulder, CO, USA) that included motion
correction, pairwise subtraction between label and control images, pairwise subtraction between
label and control images, averaging to generate the mean difference image and the calculation of
quantitative CBF maps(D. J. Wang et al., 2013; D. J. Wang et al., 2012). The LSA/MCA perforator
(MCAperf) territory (Tatu, Moulin, Bogousslavsky, & Duvernoy, 1998) region of interest was
extracted using our in-house Matlab scripts (Samantha J. Ma, Jann, et al., 2019).
4.3.3. LSA Delineation Rating Scale
To evaluate the LSA quality in the T1w TSE-VFA black blood images, a 4-point LSA
delineation (LSAD) rating scale was developed as described in Table 4.1 and evaluated by two
independent raters. The LSAD rating scale is based on the number and observable tortuosity on
10mm thin minimum intensity projection (minIP) images from T1w TSE-VFA.
58

4.3.4. Statistical Analysis
Cohen’s kappa determined interrater agreement. The mean LSAD rating between raters
and MCAperf CBF were correlated pairwise with Pearson correlation. Intraclass correlation (ICC)
analysis was used to evaluate test-retest repeatability. We evaluated a mixed effects model to
determine associations of LSAD with neurocognitive assessments and vascular risk factors,
controlling for age and gender. A two-sided Wilcoxon rank-sum test determined if the median
LSAD rating was significantly different between categories for patient vascular risk factors and
Global CDR. A p-value of 0.05 was considered significant. In addition, the interaction term
between the mean LSAD rating and MCAperf CBF (LSAD*MCAperfCBF) was used in a
multilevel mixed effect generalized model to predict performance on cognitive tests, adjusted for
age, gender, and global CBF.
4.4. Results
4.4.1. Evaluation of LSAD Rating
Figure 4.1 shows examples of T1w TSE-VFA black blood minIP images for each point of
the rating scale. The Cohen’s kappa was 0.77, indicating substantial interrater agreement for
LSAD. The ICC was 0.72 for LSAD between repeat scans. Figure 4.2 shows the relation between
Table 4.1 Criteria for 4-point LSA Delineation Rating Scale
59
MCAperf CBF and LSAD. Better vessel quality was associated with higher measured local
perfusion or CBF. Only right MCAperf CBF significantly correlates with LSAD (p=0.039).
Figure 4.1 Examples of various subjects along the LSA Delineation Rating Scale with test-retest reliability of LSA
delineation. In the male subject with LSAD rating of 1 (far left), the vessels are hardly visible, and he had diabetes
and a CDR score of 0.5. The healthy aged female subject with LSAD rating of 4 (far right) has more than 6 relatively
straight vessels on each side.
Figure 4.2 Added variable plots depicting the relation between CBF in MCAperf territory
(yellow ROI) and LSAD rating for vessel quality. Although bilateral MCAperf CBF (top,
adjusted for age, gender, and global CBF) has a significantly positive relationship with LSAD
rating (p=0.0448), the Pearson correlation was only significant in the right hemisphere
(p=0.039), where our predominantly right-handed cohort may not have as good vascular
reserve capacity in early cerebral SVD.
60
4.4.2. LSAD Associations with Neurocognitive Assessment and Vascular Risk
Figure 4.3 shows added variable plots of the mixed effect model results for various tests.
LSAD was significantly positively correlated with the uncorrected standard scores for Flanker
(coefficient=4.58, p=0.003), PSMTa (3.15, p=0.018), PSMTb (4.67, p=0.001), PSMTc (5.12,
p=0.003), and PCPS (3.58, p=0.023). The non-dominant Pegboard raw score, which measures
completion times, was also significantly negatively correlated with LSAD (-2.56, p=0.001), with
better manual dexterity for higher LSAD. Higher LSAD was also significantly correlated with
lower blood-based measures of cholesterol (-9.36, p=0.038) and hemoglobin A1c (-0.30, p=0.042).
For neuropsychological testing, LSAD had significant associations with MoCA z-score (0.51,
p=0.011), Craft Story Verbatim Mean (1.9, p=0.020), Craft Story Paraphrase Mean (1.42,
p=0.019), SEVLT Long Delay Cued Recall (1.38, p=0.016), Numbers Span Forward z-score (-
0.24, p=0.008), and total correct Digit Symbol Coding (3.4, p=0.036).
Figure 4.4 includes boxplots of LSAD for Global CDR of 0 (normal) vs. 0.5 (very mild
dementia) as well as history of hyperlipidemia and diabetes. Wilcoxon rank-sum test indicated the
rejection of the equal median hypothesis for Global CDR (p<0.001), history of hyperlipidemia
(p=0.002), and history of diabetes (p=0.004).
61

Figure 4.3 Added variable plots depicting the significant relation between LSAD rating for vessel quality and
hemoglobin A1c, Flanker uncorrected and fully corrected scores, PSMTb age-corrected score, PCPS uncorrected
score, and pegboard non-dominant score. In general, patients with higher vessel quality performed better on cognitive
tests and exhibited lower HbA1c.

Figure 4.4 Box plots demonstrating the distribution of LSAD rating history of hyperlipidemia, history of diabetes,
and global CDR. Patients with a history of vascular risk factors and mild cognitive decline had significantly lower
LSAD ratings.
62
4.4.3. LSAD*MCAperf CBF Association with Cognition and Vascular Risk
The interaction term, LSAD*MCAperf CBF, was significantly positively correlated with
MoCA z-score (p=0005), Flanker (p=0.007), PSMTa (p=0.011), PSMTb (p=0.001), and Non-
dominant Pegboard (p=0.034). Figure 4.5 shows the added variable plots of the above significant
associations corrected for age, gender, and global CBF. Figure 4.6 displays the box plots of the
distribution of LSAD*MCAperf CBF for Global CDR (p=0.0051) and history of hyperlipidemia
(p=0.011). In both cases, normal subjects had significantly higher median LSAD*MCAperf CBF,
or better vessel quality and local perfusion, than affected subjects.

Figure 4.5 Added variable plots depicting the significant relations between LSAD*MCAperf CBF (corrected for
age, gender, and global CBF) and MoCA z-score (p=0.005), Flanker (p=0.007), PSMTa (p=0.011), and PSMTb
(p=0.001) uncorrected standard scores, as well as Pegboard Non-dominant Score (p=0.034). Higher scores
indicate higher level of cognitive or motor ability, which is observed in the subjects with higher LSAD*MCAperf
CBF or overall better vessel quality and perfusion.
63

Figure 4.6 Box plots of the distribution of LSAD*MCAperfCBF for Global CDR 0 (normal) vs. 0.5 (very mild
dementia) (p=0.0051) and Normal vs. History of Hyperlipidemia (p=0.011). In both cases, normal subjects had
significantly higher median LSAD*MCAperfCBF, or better vessel quality and perfusion, than affected subjects.

4.5. Discussion
4.5.1. Clinical Utility of Black Blood MRI and the LSAD Rating Scale
The microvasculature involved in SVD includes small arteries/arterioles (~10μm-1mm),
capillaries (<10μm) and venules (~10-50μm)(Charidimou et al., 2016). While it is generally
believed these small vessels are not accessible to MRI, we demonstrated that high resolution black
blood MRI with sub-millimeter spatial resolution is feasible at clinically available field strengths
for visualizing the small artery/arteriole portion of the microvascular spectrum. Characterizing the
LSAs from images acquired at clinical field strength with regards to cognition and vascular risk
factors is important in the development of an earlier imaging marker of VCID.
While previous studies have looked at lenticulostriate arteries in aging using UHF (Cho et
al., 2008; C. K. Kang et al., 2009; Seo et al., 2012), UHF TOF MRA is not always feasible for
older patients due to the higher prevalence of metal implants, reduced thermo-autoregulation (SAR
sensitivity), and higher sensitivity to vertigo. However, there is insight to be gained by looking at
the relationship between the morphology of small vessels as one ages healthily or ages with
64
vascular risk factors. The 3D T1w TSE-VFA sequence used in this study can be applied for the
visualization, segmentation, and morphological quantification of perforating lenticulostriate
arteries, as demonstrated in Chapter 3. More importantly, 3D TSE-VFA is an FDA approved
sequence installed on all major MR vendor platforms, allowing direct translation to clinical trials
or clinical practice.
The LSAD rating system provides a simple and fast metric for observing the quality of the
LSAs visible in the T1w TSE-VFA minIP images. Observationally, the number of LSAs detected
– 2 to 12 vessels in each hemisphere – were on par with cadaver anatomical studies (Marinković
et al., 2001); thus, using a vessel number cutoff of 6 vessels was reasonable. In this study, there
were significant pairwise correlations between LSAD rating and various cognitive assessments. In
general, subjects with a higher number of relatively less tortuous LSAs tended to have higher
performance on assessments of executive function and attention, cognitive flexibility, as well as
information processing speed. These results agreed with prior studies of cerebral small vessel
disease and rate of decline in specific cognitive domains (Callahan, Sharma, Ramirez, Stuss, &
Black, 2017; Prins et al., 2005).
4.5.2. LSA Structure and Cerebrovascular Function
During this study, we found that high-resolution black-blood T1w-VFA-TSE can be used
in conjunction with pCASL to characterize the status of LSAs with subcortical perfusion in
patients with vascular risk factors. While subcortical perfusion from pCASL CBF maps is
quantitative, it may be susceptible to noise; thus, utilizing the LSAD rating as a qualitative
weighting enables us to create a good representation of morphology and cerebrovascular function
for evaluation. In this cohort, there was a positive relationship between vessel quality and local
perforator-supplied perfusion. With more numerous and less tortuous LSAs, the bilateral CBF was
65
significantly higher, suggesting stronger blood supply to the subcortical regions of the brain.
Nevertheless, upon separation by hemisphere, we only found a significant correlation between the
right MCAperf CBF and LSAD. Observationally, more LSAs could typically be delineated by our
readers in the left hemisphere than the right hemisphere. Although further longitudinal analysis is
needed, this unilateral effect was possibly observed because the left MCA perforator had better
reserve capacity in our primarily right-handed cohort. Perhaps the right MCA perforator vessels
may be affected earlier in the disease process; however, more longitudinal analysis will be needed
better understand this trend. When LSAD rating and MCAperf CBF are evaluated together as an
interaction term, they may serve as a potential imaging-based marker to identify early small vessel
and local perfusion changes related to vascular cognitive impairment and dementia.
4.5.3. Limitations
The qualitative LSAD rating scale is unfortunately still subjective in nature. There was
somewhat of a learning curve when training the readers in identifying relative tortuosity, thereby
explaining the mediocre interrater reliability. In future, the LSAD rating scale can be better defined
by also listing the number of turns required to be considered as tortuous. Furthermore, not all
participants completed the clinical interviews and neuropsychological assessments. There may
have been a slight bias toward individuals who were willing to maintain attention and
concentration for the duration of the 2-3 hour cognitive evaluations. As more participant data is
collected in this ongoing study, the current findings will be verified to evaluate if this imaging-
based biomarker can be successfully translated to the clinical setting. The majority of participants
were cognitively normal or exhibited only very mild dementia, thus severe cerebral small vessel
disease-related cerebrovascular alterations could not be properly assessed. Nevertheless, as these
participants exhibit significant vascular risk factors such as diabetes, hypertension, and/or
66
hypercholesterolemia, we have a unique opportunity to study a spectrum of the longitudinal effects
of vascular risk factors on cognition in future analyses.
4.6. Conclusion
In this study, we demonstrated that morphological features of LSAs may serve as a
potential imaging marker of cerebral SVD with respect to VCID. Furthermore, the interaction term
between LSAD rating and MCA perforator CBF can be a valuable marker of both LSA quality
and function. In future, more data samples will be collected, and quantitative shape analysis will
be performed with the techniques developed in Chapter 3 to confirm these preliminary clinical
findings.

67
5. CHAPTER 5: Conclusion and Ongoing Work
In this dissertation, we implemented and systematically optimized high-resolution 3D
black blood MRI at both ultra-high field and clinical field strength for the novel application of
vascular cognitive impairment and dementia. Using this imaging technique enables the clear
visualization and delineation of lenticulostriate arteries relative to the whole brain structure. With
the development of both a qualitative rating method and quantitative shape analyses, we
demonstrated the achievability of directly evaluating the morphology of small perforating arteries
in relation to cognitive function and overall microvascular health. Furthermore, we successfully
utilized deep learning algorithms to automatically segment the LSAs from the black blood MR
images using manual segmentation as supervision. In future work, we aim to further improve the
automatic segmentation algorithm by exploiting the benefits of UHF imaging for higher image
quality. Through our continued rapport with the precious LALES cohort and collaboration with
the MarkVCID consortium, we are also working on the translation of the proposed black blood
MRI technique for use in the clinical space with applications in cerebral SVD staging and vascular
dementia prevention and/or monitoring.
5.1. Future Directions in Ultra-High Field
As demonstrated in Chapter 2, acquisition at UHF universally increases the sensitivity for
MRI, producing images with both higher SNR and CNR. However, there are still distinct
challenges to address such as RF field inhomogeneity and tissue heating, which may extend
acquisition times and ultimately hinder high-quality imaging.
5.1.1. Patient Safety
The turbo spin echo pulse sequence relies on a series of refocusing pulses with a z-direction
phase-encoding gradient for each echo to significantly reduce imaging time, especially for three
68
dimensional MRI. Despite the usage of lower variable flip angles in the proposed sequence, this
series of refocusing pulses can still result in high energy deposition or specific absorption rate
(SAR), particularly at UHF or a higher static magnetic field. The dielectric properties at UHF result
in shorter RF wavelengths, which increase the likelihood of foci of RF heating occurring (Fiedler,
Ladd, & Bitz, 2018). Compared to lower field strengths, standard local SAR limits must be
considered for all transmit RF coils, and these limits are a necessary constraint to consider when
optimizing the sequence parameters. While the purpose of this dissertation was to methodically
evaluate the T1-weighted TSE-VFA sequence at 3T and 7T with comparable parameters, there is
still room for improvement when imaging at UHF. Future efforts will include more UHF-specific
simulations to reduce SAR with novel variable flip angle schemes and implementation of parallel
imaging techniques such as CAIPIRINHA (controlled aliasing in parallel imaging results in higher
acceleration) to reduce total scan time and improve patient comfort (Fritz et al., 2016).
5.1.2. Parallel Transmission (pTx)
Our conventional homogeneous bird-cage type coil driven by a single RF-pulse waveform
does not possess spatial degrees of freedom, thus it works best for uniform excitations. While
usable at 3T or lower field strengths, field non-uniformities arise when the wavelength of the RF
Figure 5.1 Destructive excitation field interference with a conventional circularly
polarized single transmission (1Tx) coil causes characteristic strong center
brightening, while parallel transmission (pTx) allows more spatial degrees of freedom
for shape-specific B1 shimming to achieve more uniform B1 field.
69
approaches the dimension of the human head, which is observed at ultra-high field strength,
creating destructive excitation field interference and characteristic strong center brightening
(Figure 5.1). The resulting non-uniform images with spatially varying contrast are not ideal for
diagnostic evaluation; consequently, changes in hardware must be introduced to address these
technical hurdles. Parallel transmission (pTx) allows more spatial degrees of freedom for tailored
RF pulses or shape specific B1 shimming for a more uniform field (Katscher & Bornert, 2006;
Padormo, Beqiri, Hajnal, & Malik, 2016). According to antenna theory whenever the current
distribution over a cylindrical surface satisfies sinusoidal angular dependence, a resonant condition
exists, and a homogeneous magnetic field can be created inside the conductor (Figure 5.2). With
more spatial degrees of freedom, we have an opportunity to tailor the excitation pulse to mitigate
non-uniform flip angle problems due to inhomogeneous B1 field and reduce SAR. Moreover, we
can consider anatomy-specific excitation to reduce image encoding needs by reducing the effective
field of view for potentially more clinically useful, highly specific, and localized imaging.
5.2. Ongoing Development of the Deep Learning Model
In Chapter 3, we demonstrated that HighRes3DNet could successfully segment the LSAs
with decent performance. With additional fine tuning of model hyperparameters and the addition
of transfer learning, there is ample room for the current algorithm to continue to improve in
Figure 5.2 With pTx, it is possible to carefully control the homogeneity of RF excitation. Each RF source produces a
standing wave, which creates these shading artifacts (dielectric effects) in the periphery. By simultaneously
transmitting opposing subfields, the net B1 field experienced by the tissue would be more homogeneous.
70
performance. Specifically, hyperparameters such as the learning rate, batch size, momentum, and
weight decay among many others can be optimized to achieve the best performance on the Dice,
average Hausdorff distance, or 95% percentile Hausdorff distance, respectively. Several
approaches can be employed to search for the best hyperparameter configuration, including
Random Search (Bergstra & Bengio, 2012) or Bayesian Sequential Model Based Optimization
(Hutter, Hoos, Leyton-Brown, & Murphy, 2010). Moreover, k-fold cross-validation (likely k=10)
will be employed to reduce the bias of the model toward one specific test dataset. Since we have a
limited sample size, k-fold cross-validation can help to provide a less optimistic estimate of the
DL model performance (Kuhn & Johnson, 2013).
In terms of transfer learning or the use of pretrained models, data augmentation is a well-
known method to decrease overfitting of the model. There have been prior studies in segmentation
of commonly visible large vessels in bright blood TOF MRA (Livne et al., 2019; Phellan et al.,
2017) that can be used as pre-training to give the DL model a general context about the vessel
objects in the image. It is common practice to import and use models from published literature,
such as DeepVesselNet (Tetteh et al., 2018), to speed up the convergence of DL models compared
to training from scratch as in Chapter 3. It is also possible to generate a larger dataset of synthetic
vessels to augment our small training dataset using Variational Autoencoders (VAE) and
Generative Adversarial Networks (GAN) to create 3D geometries describing vessel-like structures
based on real participant anatomies (Danu, Nita, Vizitiu, Suciu, & Itu, 2019). From the results of
Chapter 3 we found that OOF, while highly sensitive to any curvilinear structure in the image, is
not particularly specific to blood vessels in the presence of dark CSF or other tissue structures. We
plan to use this sensitivity to our advantage by incorporating this information in the pre-training,
since manual segmentation can sometimes miss the finer vessel details. As we continue to improve
71
our DL algorithm for automatic LSA segmentation, we will also include more UHF images as
additional training data to reduce overfitting.
5.3. Clinical Translation
Moving forward, we will segment and perform shape analysis on all the LALES participant
data based on the improved automatic segmentations to evaluate the quantitative morphological
measures as a more specific imaging-based biomarker of vascular cognitive impairment and
dementia. The LALES cohort is largely cognitively normal but exhibit significant vascular risk
factors such as diabetes, hypertension, and hypercholesterolemia; therefore, we have a unique
opportunity to study the longitudinal effects of vascular risk on cognition. However, because the
black blood MRI sequence takes approximately ten minutes to acquire, clinical adoption may be
slow. Nevertheless, we will concentrate on working closely with neurologists and clinicians to
translate the developed biomarker to the clinical setting by maintaining sufficient image quality
while reducing scan time and SAR-related risks at UHF.
Incidentally, this close collaboration has resulted in potentially important findings
regarding LSA microaneurysms, which can be potentially debilitating upon rupture. For example,
in one case during this study, we were able to identify a fibrin clot due to blood brain barrier
leakage in a hypertensive and diabetic patient using black blood MRI at 3T (Figure 5.3). An UHF
multi-modality approach (Figure 5.4) was employed to rule out the presence of a dangerous LSA
microaneurysm (Barisano et al., 2018). Further collaboration with clinicians will enable us to
translate and confirm the clinical utility of our proposed technique. As we continue to collect
images and clinical outcomes data from patients with early or mild effects of vascular-related
cognitive impairment, it is our intention to optimize our proposed biomarker toolkit so it can guide
earlier interventions in order to slow the progression of vascular dementia.
72

Figure 5.3 A globe-like structure on the LSA observed in a 64 yo hypertensive and diabetic female patient. The globe
structure could be: (1) With aging and brain inflammation, the leakiness and increased BBB permeability allows for
the passage of metabolic waste products, including fibrin and other blood products in the PVS, obstructing the CSF-
ISF flow and PVSs, causing a “fibrin globe” and subsequent dilation of the PVS. (2) True lenticulostriate
microaneurysm, related to hypertension and other vascular risk factors, visible on TOF. (3) Pseudo aneurysm, blood
products in the LSA vessel wall or lipohyalinosis.Adapted from Barisano, G., Ma, S.J., et al, ISMRM, 2018.

Figure 5.4 Multi-modal sub-millimeter resolution UHF MRI enabled the determination of a fibrin clot due to BBB
leakage in the perivascular space. Adapted from Barisano, G., Ma, S.J., et al, ISMRM, 2018.

73
References
Alexander, A. L., Buswell, H. R., Sun, Y., Chapman, B. E., Tsuruda, J. S., & Parker, D. L. (1998).
Intracranial black-blood MR angiography with high-resolution 3D fast spin echo. Magn
Reson Med, 40(2), 298-310.
Alexander, L. (1942). The vascular supply of the strio-palladium. Res Publ Assoc Res Nerv Ment
Dis, 21, 77-132.
Alsop, D. C., Dai, W., Grossman, M., & Detre, J. A. (2010). Arterial spin labeling blood flow
MRI: its role in the early characterization of Alzheimer's disease. Journal of Alzheimer's
disease : JAD, 20(3), 871-880. doi:10.3233/JAD-2010-091699
Asadi, H., Dowling, R., Yan, B., & Mitchell, P. (2014). Machine Learning for Outcome Prediction
of Acute Ischemic Stroke Post Intra-Arterial Therapy. PLoS One, 9(2), e88225.
doi:10.1371/journal.pone.0088225
Barisano, G., Ma, S. J., Wang, D. J., Shi, Y., Toga, A. W., & Law, M. (2018). Imaging the
perivascular spaces and lenticulostriate arteries at 3T and 7T to provide insight into the
pathophysiology of neurodegenerative and neurovascular diseases. Paper presented at the
Proc. Intl. Soc. Mag. Reson. Med, Paris, France.
Becker, E. D., & Farrar, T. C. (1969). Driven equilibrium Fourier transform spectroscopy. A new
method for nuclear magnetic resonance signal enhancement. J Am Chem Soc, 91(27), 7784-
7785.
Beevor, C. E. (1907). The cerebral arterial supply. Brain, 30, 403-425.
Bergstra, J., & Bengio, Y. (2012). Random Search for Hyper-Parameter Optimization. Journal of
Machine Learning Research, 13, 281-305.
Blockley, N. P., Jiang, L., Gardener, A. G., Ludman, C. N., Francis, S. T., & Gowland, P. A.
(2008). Field strength dependence of R1 and R 2* relaxivities of human whole blood to
prohance, vasovist, and deoxyhemoglobin. Magnetic Resonance in Medicine, 60(6), 1313-
1320. doi:10.1002/mrm.21792
Bouvy, W. H., Geurts, L. J., Kuijf, H. J., Luijten, P. R., Kappelle, L. J., Biessels, G. J., &
Zwanenburg, J. J. M. (2016). Assessment of blood flow velocity and pulsatility in cerebral
perforating arteries with 7‐T quantitative flow MRI. Nmr in Biomedicine, 29(9), 1295-
1304. doi:10.1002/nbm.3306
Breteler, M. M., van Amerongen, N. M., van Swieten, J. C., Claus, J. J., Grobbee, D. E., van Gijn,
J., . . . van Harskamp, F. (1994). Cognitive correlates of ventricular enlargement and
cerebral white matter lesions on magnetic resonance imaging. The Rotterdam Study.
Stroke, 25(6), 1109-1115.
74
Busse, R. F., Brau, A. C., Vu, A., Michelich, C. R., Bayram, E., Kijowski, R., . . . Rowley, H. A.
(2008). Effects of refocusing flip angle modulation and view ordering in 3D fast spin echo.
Magn Reson Med, 60(3), 640-649. doi:10.1002/mrm.21680
Callaghan, M. F., Josephs, O., Herbst, M., Zaitsev, M., Todd, N., & Weiskopf, N. (2015). An
evaluation of prospective motion correction (PMC) for high resolution quantitative MRI.
Frontiers in Neuroscience, 9, 97. doi:10.3389/fnins.2015.00097
Callahan, B. L., Sharma, M. J., Ramirez, J., Stuss, D. T., & Black, S. E. (2017). THE IMPACT
OF SMALL VESSEL DISEASE BURDEN ON EXECUTIVE FUNCTIONING IN
INDIVIDUALS WITH OR AT RISK FOR DEMENTIA. Alzheimer's & Dementia: The
Journal of the Alzheimer's Association, 13(7), P916-P917. doi:10.1016/j.jalz.2017.07.342
Charcot, J., & Bouchard, C. (1868). Nouvelles recherches sur la pathogenie de l'hemorrhagie
cerebrale. Arch Physiol Norm Pathol, 1, 110–127, 643–665.
Charcot, J. M. (1883). Lectures VII, VIII, and IX, in Hadden WB (ed): Lectures on Localization
of Cerebral and Spinal Diseases. London: The New Sydenham Society, pp 46–97.
Charidimou, A., Pantoni, L., & Love, S. (2016). The concept of sporadic cerebral small vessel
disease: A road map on key definitions and current concepts. Int J Stroke, 11(1), 6-18.
doi:10.1177/1747493015607485
Cho, Z. H., Kang, C. K., Han, J. Y., Kim, S. H., Kim, K. N., Hong, S. M., . . . Kim, Y. B. (2008).
Observation of the lenticulostriate arteries in the human brain in vivo using 7.0T MR
angiography. Stroke, 39(5), 1604-1606. doi:10.1161/STROKEAHA.107.508002
Çiçek, Ö., Abdulkadir, A., Lienkamp, S., Brox, T., & Ronneberger, O. (2016). 3D U-Net: Learning
Dense Volumetric Segmentation from Sparse Annotation.
Cox, E. F., Gowland, P.A. (2008). Measuring T2 and T2' in the brain at 1.5T, 3T and 7T using a
hybrid gradient echo-spin echo sequence and EPI. Paper presented at the Proc. Intl. Soc.
Mag. Reson. Med., Toronto, Canada.
Danu, M., Nita, C., Vizitiu, A., Suciu, C., & Itu, L. M. (2019, 9-11 Oct. 2019). Deep learning
based generation of synthetic blood vessel surfaces. Paper presented at the 2019 23rd
International Conference on System Theory, Control and Computing (ICSTCC).
De Cocker, L. J., Lindenholz, A., Zwanenburg, J. J., van der Kolk, A. G., Zwartbol, M., Luijten,
P. R., & Hendrikse, J. (2018). Clinical vascular imaging in the brain at 7T. NeuroImage,
168, 452-458. doi:10.1016/j.neuroimage.2016.11.044
Debette, S., Beiser, A., DeCarli, C., Au, R., Himali, J. J., Kelly-Hayes, M., . . . Seshadri, S. (2010).
Association of MRI markers of vascular brain injury with incident stroke, mild cognitive
impairment, dementia, and mortality: the Framingham Offspring Study. Stroke, 41(4), 600-
606. doi:10.1161/STROKEAHA.109.570044
75
DeCarli, C., Murphy, D. G., Tranh, M., Grady, C. L., Haxby, J. V., Gillette, J. A., . . . et al. (1995).
The effect of white matter hyperintensity volume on brain structure, cognitive
performance, and cerebral metabolism of glucose in 51 healthy adults. Neurology, 45(11),
2077-2084.
Deistung, A., Rauscher, A., Sedlacik, J., Stadler, J., Witoszynskyj, S., & Reichenbach, J. R. (2008).
Susceptibility weighted imaging at ultra high magnetic field strengths: theoretical
considerations and experimental results. Magn Reson Med, 60(5), 1155-1168.
doi:10.1002/mrm.21754
Detre, J. A., Rao, H., Wang, D. J., Chen, Y. F., & Wang, Z. (2012). Applications of arterial spin
labeled MRI in the brain. Journal of magnetic resonance imaging : JMRI, 35(5), 1026-
1037. doi:10.1002/jmri.23581
Dichgans, M., & Leys, D. (2017). Vascular Cognitive Impairment. Circ Res, 120(3), 573-591.
doi:10.1161/CIRCRESAHA.116.308426
Dobre, M. C., Uğurbil, K., & Marjanska, M. (2007). Determination of blood longitudinal
relaxation time (T1) at high magnetic field strengths. Magnetic Resonance Imaging, 25(5),
733-735. doi:https://doi.org/10.1016/j.mri.2006.10.020
Duret, H. (1873). Note sur la circulation du corps strié. Progres Medical, 1, 259-260.
Duyn, J. H., van Gelderen, P., Li, T. Q., de Zwart, J. A., Koretsky, A. P., & Fukunaga, M. (2007).
High-field MRI of brain cortical substructure based on signal phase. Proc Natl Acad Sci U
S A, 104(28), 11796-11801. doi:10.1073/pnas.0610821104
Fan, Z., Yang, Q., Deng, Z., Li, Y., Bi, X., Song, S., & Li, D. (2016). Whole-brain intracranial
vessel wall imaging at 3 Tesla using cerebrospinal fluid-attenuated T1-weighted 3D turbo
spin echo. Magn Reson Med. doi:10.1002/mrm.26201
Fan, Z., Yang, Q., Deng, Z., Li, Y., Bi, X., Song, S., & Li, D. (2017). Whole-Brain Intracranial
Vessel Wall Imaging at 3 Tesla Using Cerebrospinal Fluid–Attenuated T1-Weighted 3D
Turbo Spin Echo. Magnetic Resonance in Medicine, 77(3), 1142-1150.
doi:10.1002/mrm.26201
Fiedler, T. M., Ladd, M. E., & Bitz, A. K. (2018). SAR Simulations & Safety. NeuroImage, 168,
33-58. doi:10.1016/j.neuroimage.2017.03.035
Fischl, B., Salat, D. H., Busa, E., Albert, M., Dieterich, M., Haselgrove, C., . . . Dale, A. M. (2002).
Whole brain segmentation: automated labeling of neuroanatomical structures in the human
brain. Neuron, 33(3), 341-355.
Fischl, B., Salat, D. H., van der Kouwe, A. J., Makris, N., Segonne, F., Quinn, B. T., & Dale, A.
M. (2004). Sequence-independent segmentation of magnetic resonance images.
NeuroImage, 23 Suppl 1, S69-84. doi:10.1016/j.neuroimage.2004.07.016
76
Floyd, T. F., Ratcliffe, S. J., Wang, J., Resch, B., & Detre, J. A. (2003). Precision of the CASL-
perfusion MRI technique: global and regional cerebral blood flow within vascular
territories at one hour and one week. J Magn Reson Imaging, 18, 649-655.
Frangi, A. F., Niessen, W. J., Vincken, K. L., & Viergever, M. A. (1998). Multiscale vessel
enhancement filtering. In A. C. a. S. L. D. W.M. Wells (Ed.), Lecture Notes in Computer
Science (Vol. 1496, pp. 130-137). Berlin, Germany: Springer Verlag.
Fritz, J., Fritz, B., Thawait, G. G., Meyer, H., Gilson, W. D., & Raithel, E. (2016). Three-
Dimensional CAIPIRINHA SPACE TSE for 5-Minute High-Resolution MRI of the Knee.
Investigative Radiology, 51(10), 609-617. doi:10.1097/rli.0000000000000287
Furuta, A., Ishii, N., Nishihara, Y., & Horie, A. (1991). Medullary arteries in aging and dementia.
Stroke, 22(4), 442-446. doi:10.1161/01.str.22.4.442
Geurts Lennart, J., Zwanenburg Jaco, J. M., Klijn Catharina, J. M., Luijten Peter, R., & Biessels
Geert, J. (2019). Higher Pulsatility in Cerebral Perforating Arteries in Patients With Small
Vessel Disease Related Stroke, a 7T MRI Study. Stroke, 50(1), 62-68.
doi:10.1161/STROKEAHA.118.022516
Gibson, E., Li, W., Sudre, C., Fidon, L., Shakir, D. I., Wang, G., . . . Vercauteren, T. (2018).
NiftyNet: a deep-learning platform for medical imaging. Computer Methods and Programs
in Biomedicine, 158, 113-122. doi:https://doi.org/10.1016/j.cmpb.2018.01.025
Gorelick, P. B., Scuteri, A., Black, S. E., Decarli, C., Greenberg, S. M., Iadecola, C., . . .
Anesthesia. (2011). Vascular contributions to cognitive impairment and dementia: a
statement for healthcare professionals from the american heart association/american stroke
association. Stroke, 42(9), 2672-2713. doi:10.1161/STR.0b013e3182299496
Gotoh, K., Okada, T., Satogami, N., Yakami, M., Takahashi, J. C., Yoshida, K., . . . Togashi, K.
(2012). Evaluation of CT angiography for visualisation of the lenticulostriate artery:
difference between normotensive and hypertensive patients. The British Journal of
Radiology, 85(1019), e1004-e1008. doi:10.1259/bjr/67294268
Greenwood, J. A., & Sandomire, M. M. (1950). Sample Size Required for Estimating the Standard
Deviation as a Per Cent of its True Value. Journal of the American Statistical Association,
45(250), 257-260.
Guyton, A. C., & Hall, J. E. (1991). Textbook of Medical Physiology. Philadelphia: W. B.
Saunders.
Hargreaves, B. (Producer). (2012). EPG Lecture. [Lecture] Retrieved from
http://web.stanford.edu/~bah/software/epg/
Hendrikse, J., Zwanenburg, J. J., Visser, F., Takahara, T., & Luijten, P. (2008). Noninvasive
depiction of the lenticulostriate arteries with time-of-flight MR angiography at 7.0 T.
Cerebrovasc Dis, 26(6), 624-629. doi:10.1159/000166838
77
Hutter, F., Hoos, H. H., Leyton-Brown, K., & Murphy, K. (2010). Time-Bounded Sequential
Parameter Optimization. Learning and Intelligent Optimization, 6073, 281-298. doi:Doi
10.1007/978-3-642-13800-3_30
Iadecola, C. (2016). Vascular and Metabolic Factors in Alzheimer's Disease and Related
Dementias: Introduction. Cell Mol Neurobiol. doi:10.1007/s10571-015-0319-y
Kammerer, S., Mueller-Eschner, M., Berkefeld, J., & Tritt, S. (2017). Time-resolved 3D
Rotational Angiography (4D DSA) of the Lenticulostriate Arteries: Display of Normal
Anatomic Variants and Collaterals in Cases with Chronic Obstruction of the MCA. Clin
Neuroradiol, 27(4), 451-457. doi:10.1007/s00062-017-0578-8
Kamnitsas, K., Ledig, C., Newcombe, V. F. J., Simpson, J. P., Kane, A. D., Menon, D. K., . . .
Glocker, B. (2017). Efficient multi-scale 3D CNN with fully connected CRF for accurate
brain lesion segmentation. Medical Image Analysis, 36, 61-78.
doi:https://doi.org/10.1016/j.media.2016.10.004
Kang, C.-K., Park, C.-W., Han, J.-Y., Kim, S.-H., Park, C.-A., Kim, K.-N., . . . Cho, Z.-H. (2009).
Imaging and analysis of lenticulostriate arteries using 7.0-Tesla magnetic resonance
angiography. Magnetic Resonance in Medicine, 61(1), 136-144.
doi:doi:10.1002/mrm.21786
Kang, C. K., Park, C. A., Lee, D. S., Lee, Y. B., Park, C. W., Kim, Y. B., & Cho, Z. H. (2016).
Velocity measurement of microvessels using phase-contrast magnetic resonance
angiography at 7 Tesla MRI. Magn Reson Med, 75(4), 1640-1646. doi:10.1002/mrm.25600
Kang, C. K., Park, C. A., Lee, H., Kim, S. H., Park, C. W., Kim, Y. B., & Cho, Z. H. (2009).
Hypertension correlates with lenticulostriate arteries visualized by 7T magnetic resonance
angiography. Hypertension, 54(5), 1050-1056.
doi:10.1161/HYPERTENSIONAHA.109.140350
Katscher, U., & Bornert, P. (2006). Parallel RF transmission in MRI. NMR Biomed, 19(3), 393-
400. doi:10.1002/nbm.1049
Kellman, P., & McVeigh, E. R. (2005). Image reconstruction in SNR units: a general method for
SNR measurement. Magn Reson Med, 54(6), 1439-1447. doi:10.1002/mrm.20713
Kingma, D., & Ba, J. (2014). Adam: A Method for Stochastic Optimization. International
Conference on Learning Representations.
Klein, S., Staring, M., Murphy, K., Viergever, M. A., & Pluim, J. P. W. (2010). Elastix: A Toolbox
for Intensity-Based Medical Image Registration. IEEE Transactions on Medical Imaging,
29(1), 196-205. doi:10.1109/TMI.2009.2035616
Krishnamurthy, L. C., Liu, P., Xu, F., Uh, J., Dimitrov, I., & Lu, H. (2014). Dependence of blood
T2 on oxygenation at 7 T: In vitro calibration and in vivo application. Magnetic Resonance
in Medicine, 71(6), 2035-2042. doi:doi:10.1002/mrm.24868
78
Kuhn, M., & Johnson, K. (2013). Applied Predictive Modeling (1 ed.). New York: Springer-Verlag
New York.
Law, M., & Chung, A. (2008). Three Dimensional Curvilinear Structure Detection Using
Optimally Oriented Flux.
Law, M. W., & Chung, A. C. (2008). Three dimensional curvilinear structure detection using
optimally oriented flux. Proc. 10th Eur. Conf. Comput. Vision,, 368-382.
Lawrence, A. J., Brookes, R. L., Zeestraten, E. A., Barrick, T. R., Morris, R. G., & Markus, H. S.
(2015). Pattern and Rate of Cognitive Decline in Cerebral Small Vessel Disease: A
Prospective Study. PLoS One, 10(8), e0135523. doi:10.1371/journal.pone.0135523
LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436-444.
doi:10.1038/nature14539
Lee, S., Viqar, F., Zimmerman, M. E., Narkhede, A., Tosto, G., Benzinger, T. L., . . . Dominantly
Inherited Alzheimer, N. (2016). White matter hyperintensities are a core feature of
Alzheimer's disease: Evidence from the Dominantly Inherited Alzheimer Network. Ann
Neurol. doi:10.1002/ana.24647
Li, W., Grgac, K., Huang, A., Yadav, N., Qin, Q., & van Zijl Peter, C. M. (2016). Quantitative
theory for the longitudinal relaxation time of blood water. Magnetic Resonance in
Medicine, 76(1), 270-281. doi:doi:10.1002/mrm.25875
Li, W., Wang, G., Fidon, L., Ourselin, S., Cardoso, M. J., & Vercauteren, T. (2017). On the
Compactness, Efficiency, and Representation of 3D Convolutional Networks: Brain
Parcellation as a Pretext Task.
Lichy, M. P., Wietek, B., Mugler III, J. P., Horger, W., Menzel, M., Siegmann, K., . . . Kiefer, B.
(2005). Whole-body applications of isotropic high-resolution T2-weighted MRI with a
single slab 3D-TSE based sequence optimized for high sampling efficiency, called SPACE
- initial clinical experiences, Miami Beach, Florida, USA.
Livne, M., Rieger, J., Aydin, O. U., Taha, A. A., Akay, E. M., Kossen, T., . . . Madai, V. I. (2019).
A U-Net Deep Learning Framework for High Performance Vessel Segmentation in Patients
With Cerebrovascular Disease. Frontiers in Neuroscience, 13(97).
doi:10.3389/fnins.2019.00097
Ma, S. J., Gahm, J. K., Yan, L., Shen, Y., Shao, X., Shi, Y., & Wang, D. J. J. (2017). Evaluation
of Whole Brain and ZOOMit T1-weighted Turbo-Spin Echo (TSE) for Visualization of
Human Lenticulostriate Arteries at 3.0 T: A Preliminary Study. Paper presented at the
International Society of Magnetic Resonance in Medicine Annual Meeting, Honolulu, HI,
USA.
Ma, S. J., Jann, K., Barisano, G., Shao, X., Yan, L., Casey, M., . . . Wang, D. J. J. (2019).
CHARACTERIZATION OF LENTICULOSTRIATE ARTERIES USING ARTERIAL
SPIN LABELING AND HIGH-RESOLUTION 3D BLACK-BLOOD MRI AS AN
79
IMAGING MARKER IN VASCULAR COGNITIVE IMPAIRMENT AND DEMENTIA.
Alzheimer's & Dementia: The Journal of the Alzheimer's Association, 15(7), P75-P76.
doi:10.1016/j.jalz.2019.06.4928
Ma, S. J., Sarabi, M. S., Yan, L., Shao, X., Chen, Y., Yang, Q., . . . Wang, D. J. J. (2019).
Characterization of lenticulostriate arteries with high resolution black-blood T1-weighted
turbo spin echo with variable flip angles at 3 and 7 Tesla. NeuroImage, 199, 184-193.
doi:https://doi.org/10.1016/j.neuroimage.2019.05.065
Ma, S. J., Yan, L., Jann, K., & Wang, D. J. J. (2018). High Resolution Black-blood T1-weighted
Turbo Spin Echo with Variable Flip Angles for Visualization of Small Perforating Arteries
at 3 and 7 Tesla. Paper presented at the Proc. Intl. Soc. Mag. Reson. Med., Paris, France.
Maccotta, L., Detre, J. A., & Alsop, D. C. (1997). The efficiency of adiabatic inversion for
perfusion imaging by arterial spin labeling. NMR Biomed, 10(4-5), 216-221.
Manjón, J. V., Carbonell-Caballero, J., Lull, J. J., García-Martí, G., Martí-Bonmatí, L., & Robles,
M. (2008). MRI denoising using Non-Local Means. Medical Image Analysis, 12(4), 514-
523. doi:https://doi.org/10.1016/j.media.2008.02.004
Marinković, S., Gibo, H., Milisavljević, M., & Ćetković, M. (2001). Anatomic and clinical
correlations of the lenticulostriate arteries. Clinical Anatomy, 14(3), 190-195.
doi:doi:10.1002/ca.1032
Mayda, A., & Decarli, C. (2009). Vascular cognitive impairment: prodrome to VaD? In:. In E. T.
Wahlund L-O, Gauthier S (Ed.), Vascular Cognitive Impairment in Clinical Practice.
Cambridge, (pp. 11-31). Cambridge: Cambridge University Press.
Melhem, E. R., Itoh, R., & Folkers, P. J. M. (2001). Cervical Spine: Three-dimensional Fast Spin-
Echo MR Imaging—Improved Recovery of Longitudinal Magnetization with Driven
Equilibrium Pulse. Radiology, 218(1), 283-288. doi:10.1148/radiology.218.1.r01ja38283
Melhem, E. R., Jara, H., & Yucel, E. K. (1997). Black blood MR angiography using multislab
three-dimensional TI-weighted turbo spin-echo technique: imaging of intracranial
circulation. AJR Am J Roentgenol, 169(5), 1418-1420. doi:10.2214/ajr.169.5.9353471
Milletari, F., Navab, N., & Ahmadi, S. A. (2016). V-Net: Fully Convolutional Neural Networks
for Volumetric Medical Image Segmentation. Proceedings of 2016 Fourth International
Conference on 3d Vision (3dv), 565-571. doi:10.1109/3dv.2016.79
Mugler, J. P., 3rd. (2014). Optimized three-dimensional fast-spin-echo MRI. J Magn Reson
Imaging, 39(4), 745-767. doi:10.1002/jmri.24542
Norrving, B. (2008). Lacunar infarcts: no black holes in the brain are benign. Pract Neurol, 8(4),
222-228. doi:10.1136/jnnp.2008.153601
Padormo, F., Beqiri, A., Hajnal, J. V., & Malik, S. J. (2016). Parallel transmission for ultrahigh-
field imaging. NMR Biomed, 29(9), 1145-1161. doi:10.1002/nbm.3313
80
Pantoni, L. (2010). Cerebral small vessel disease: from pathogenesis and clinical characteristics to
therapeutic challenges. Lancet Neurol, 9(7), 689-701. doi:10.1016/S1474-4422(10)70104-
6
Park, J., Mugler, J. P., 3rd, Horger, W., & Kiefer, B. (2007). Optimized T1-weighted contrast for
single-slab 3D turbo spin-echo imaging with long echo trains: application to whole-brain
imaging. Magn Reson Med, 58(5), 982-992. doi:10.1002/mrm.21386
Patterson, C. (2018). World Alzheimer Resport 2018 The state of the art of dementia research:
New frontiers. Alzheimer's Disease International (ADI), London.
Phellan, R., Peixinho, A., Falcão, A., & Forkert, N. D. (2017). Vascular Segmentation in TOF
MRA Images of the Brain Using a Deep Convolutional Neural Network. Paper presented
at the International Workshop on Large-Scale Annotation of Biomedical Data and Expert
Label Synthesis.
Prins, N. D., van Dijk, E. J., den Heijer, T., Vermeer, S. E., Jolles, J., Koudstaal, P. J., . . . Breteler,
M. M. B. (2005). Cerebral small-vessel disease and decline in information processing
speed, executive function and memory. Brain, 128(9), 2034-2041.
doi:10.1093/brain/awh553
Qiao, Y., Steinman, D. A., Qin, Q., Etesami, M., Schar, M., Astor, B. C., & Wasserman, B. A.
(2011). Intracranial arterial wall imaging using three-dimensional high isotropic resolution
black blood MRI at 3.0 Tesla. J Magn Reson Imaging, 34(1), 22-30.
doi:10.1002/jmri.22592
Qiao, Y., Zeiler, S. R., Mirbagheri, S., Leigh, R., Urrutia, V., Wityk, R., & Wasserman, B. A.
(2014). Intracranial plaque enhancement in patients with cerebrovascular events on high-
spatial-resolution MR images. Radiology, 271(2), 534-542. doi:10.1148/radiol.13122812
Rex, D. E., Ma, J. Q., & Toga, A. W. (2003). The LONI Pipeline Processing Environment.
NeuroImage, 19(3), 1033-1048. doi:https://doi.org/10.1016/S1053-8119(03)00185-X
Ronneberger, O., Fischer, P., & Brox, T. (2015). U-Net: Convolutional Networks for Biomedical
Image Segmentation. Medical Image Computing and Computer-Assisted Intervention, Pt
Iii, 9351, 234-241. doi:10.1007/978-3-319-24574-4_28
Rosenberg, G. A., Wallin, A., Wardlaw, J. M., Markus, H. S., Montaner, J., Wolfson, L., . . .
Hachinski, V. (2015). Consensus statement for diagnosis of subcortical small vessel
disease. J Cereb Blood Flow Metab. doi:10.1038/jcbfm.2015.172
Saloner, D. (1995). The AAPM/RSNA physics tutorial for residents. An introduction to MR
angiography. RadioGraphics, 15(2), 453-465. doi:10.1148/radiographics.15.2.7761648
Schindelin, J., Arganda-Carreras, I., Frise, E., Kaynig, V., Longair, M., Pietzsch, T., . . . Cardona,
A. (2012). Fiji: an open-source platform for biological-image analysis. Nature Methods, 9,
676.doi:10.1038/nmeth.2019
https://www.nature.com/articles/nmeth.2019#supplementary-information
81
Schneider, J. A., Arvanitakis, Z., Bang, W., & Bennett, D. A. (2007). Mixed brain pathologies
account for most dementia cases in community-dwelling older persons. Neurology, 69(24),
2197-2204. doi:10.1212/01.wnl.0000271090.28148.24
Schnerr, R. S., Jansen, J. F. A., Uludag, K., Hofman, P. A. M., Wildberger, J. E., van
Oostenbrugge, R. J., & Backes, W. H. (2017). Pulsatility of Lenticulostriate Arteries
Assessed by 7 Tesla Flow MRI—Measurement, Reproducibility, and Applicability to
Aging Effect. Frontiers in Physiology, 8, 961. doi:10.3389/fphys.2017.00961
Seo, S. W., Kang, C. K., Kim, S. H., Yoon, D. S., Liao, W., Worz, S., . . . Cho, Z. H. (2012).
Measurements of lenticulostriate arteries using 7T MRI: new imaging markers for
subcortical vascular dementia. J Neurol Sci, 322(1-2), 200-205.
doi:10.1016/j.jns.2012.05.032
Shamonin, D., Bron, E., Lelieveldt, B., Smits, M., Klein, S., & Staring, M. (2014). Fast Parallel
Image Registration on CPU and GPU for Diagnostic Classification of Alzheimer's Disease.
Frontiers in Neuroinformatics, 7(50). doi:10.3389/fninf.2013.00050
Shao, X., Ma, S. J., Casey, M., D’Orazio, L., Ringman, J. M., & Wang, D. J. J. (2019). Mapping
water exchange across the blood–brain barrier using 3D diffusion-prepared arterial spin
labeled perfusion MRI. Magnetic Resonance in Medicine, 81(5), 3065-3079.
doi:10.1002/mrm.27632
Shen, D., Wu, G., & Suk, H. I. (2017). Deep Learning in Medical Image Analysis. Annu Rev
Biomed Eng, 19, 221-248. doi:10.1146/annurev-bioeng-071516-044442
Shi, Y., Lai, R., Krishna, S., Sicotte, N., Dinov, I., & Toga, A. W. (2008, 23-28 June 2008).
Anisotropic Laplace-Beltrami eigenmaps: Bridging Reeb graphs and skeletons. Paper
presented at the 2008 IEEE Computer Society Conference on Computer Vision and Pattern
Recognition Workshops.
Shi, Y., Lai, R., Morra, J. H., Dinov, I., Thompson, P. M., & Toga, A. W. (2010). Robust Surface
Reconstruction via Laplace-Beltrami Eigen-Projection and Boundary Deformation. IEEE
Transactions on Medical Imaging, 29(12), 2009-2022. doi:10.1109/TMI.2010.2057441
Shi, Y., Lai, R., Toga, A. W., & Alzheimer's Disease Neuroimaging, I. (2013). Cortical surface
reconstruction via unified Reeb analysis of geometric and topological outliers in magnetic
resonance images. IEEE Trans Med Imaging, 32(3), 511-530.
doi:10.1109/TMI.2012.2224879
Shi, Y., & Wardlaw, J. M. (2016). Update on cerebral small vessel disease: a dynamic whole-brain
disease. Stroke and Vascular Neurology, 1(3), 83-92. doi:10.1136/svn-2016-000035
Silbert, L. C., Howieson, D. B., Dodge, H., & Kaye, J. A. (2009). Cognitive impairment risk: white
matter hyperintensity progression matters. Neurology, 73(2), 120-125.
doi:10.1212/WNL.0b013e3181ad53fd
82
Stanisz, G. J., Odrobina, E. E., Pun, J., Escaravage, M., Graham, S. J., Bronskill, M. J., &
Henkelman, R. M. (2005). T1, T2 relaxation and magnetization transfer in tissue at 3T.
Magnetic Resonance in Medicine, 54(3), 507-512. doi:doi:10.1002/mrm.20605
Taha, A. A., & Hanbury, A. (2015). Metrics for evaluating 3D medical image segmentation:
analysis, selection, and tool. BMC Medical Imaging, 15(1), 29. doi:10.1186/s12880-015-
0068-x
Tatu, L., Moulin, T., Bogousslavsky, J., & Duvernoy, H. (1998). Arterial territories of the human
brain. Cerebral hemispheres, 50(6), 1699-1708. doi:10.1212/wnl.50.6.1699
Tetteh, G., Efremov, V., Forkert, N. D., Schneider, M., Kirschke, J., Weber, B., . . . Menze, B.
(2018). DeepVesselNet: Vessel Segmentation, Centerline Prediction, and Bifurcation
Detection in 3-D Angiographic Volumes.
Umansky, F., Gomes, F. B., Dujovny, M., Diaz, F. G., Ausman , J. I., Mirchandani, H. G., &
Berman, S. K. (1985). The perforating branches of the middle cerebral artery. Journal of
Neurosurgery, 62(2), 261-268. doi:10.3171/jns.1985.62.2.0261
van der Kolk, A. G., Zwanenburg, J. J., Brundel, M., Biessels, G. J., Visser, F., Luijten, P. R., &
Hendrikse, J. (2011). Intracranial vessel wall imaging at 7.0-T MRI. Stroke, 42(9), 2478-
2484. doi:10.1161/STROKEAHA.111.620443
van Ooij, P., Zwanenburg, J. J., Visser, F., Majoie, C. B., vanBavel, E., Hendrikse, J., &
Nederveen, A. J. (2013). Quantification and visualization of flow in the Circle of Willis:
time-resolved three-dimensional phase contrast MRI at 7 T compared with 3 T. Magn
Reson Med, 69(3), 868-876. doi:10.1002/mrm.24317
van Os, H. J. A., Ramos, L. A., Hilbert, A., van Leeuwen, M., van Walderveen, M. A. A., Kruyt,
N. D., . . . , t. M. C. R. I. (2018). Predicting Outcome of Endovascular Treatment for Acute
Ischemic Stroke: Potential Value of Machine Learning Algorithms. Frontiers in
Neurology, 9(784). doi:10.3389/fneur.2018.00784
Van Uijen, C. M. J., & Den Boef, J. H. (1984). Driven-equilibrium radiofrequency pulses in NMR
imaging. Magnetic Resonance in Medicine, 1(4), 502-507.
doi:doi:10.1002/mrm.1910010409
Viessmann, O., Li, L., Benjamin, P., & Jezzard, P. (2017). T2-Weighted intracranial vessel wall
imaging at 7 Tesla using a DANTE-prepared variable flip angle turbo spin echo readout
(DANTE-SPACE). Magn Reson Med, 77(2), 655-663. doi:10.1002/mrm.26152
Wachinger, C., Reuter, M., & Klein, T. (2018). DeepNAT: Deep convolutional neural network for
segmenting neuroanatomy. Neuroimage, 170, 434-445.
doi:https://doi.org/10.1016/j.neuroimage.2017.02.035
Wang, D. J., Alger, J. R., Qiao, J. X., Gunther, M., Pope, W. B., Saver, J. L., . . . Investigators, U.
S. (2013). Multi-delay multi-parametric arterial spin-labeled perfusion MRI in acute
83
ischemic stroke - Comparison with dynamic susceptibility contrast enhanced perfusion
imaging. Neuroimage Clin, 3, 1-7. doi:10.1016/j.nicl.2013.06.017
Wang, D. J., Alger, J. R., Qiao, J. X., Hao, Q., Hou, S., Fiaz, R., . . . Investigators, U. S. (2012).
The value of arterial spin-labeled perfusion imaging in acute ischemic stroke: comparison
with dynamic susceptibility contrast-enhanced MRI. Stroke, 43(4), 1018-1024.
doi:10.1161/STROKEAHA.111.631929
Wang, J., Aguirre, G. K., Kimberg, D. Y., Roc, A. C., Li, L., & Detre, J. A. (2003). Arterial spin
labeling perfusion fMRI with very low task frequency. Magn Reson Med, 49(5), 796-802.
Wang, K., Shou, Q., Ma, S. J., Liebeskind, D., Qiao, X. J., Saver, J. L., . . . Wang, D. J. (2020).
Deep Learning Detection of Penumbral Tissue on Arterial Spin Labeling in Stroke. Stroke,
51, 00-00. doi:10.1161/STROKEAHA.119.027457
Wardlaw, J. M., Doubal, F., Armitage, P., Chappell, F., Carpenter, T., Munoz Maniega, S., . . .
Dhillon, B. (2009). Lacunar stroke is associated with diffuse blood-brain barrier
dysfunction. Ann Neurol, 65(2), 194-202. doi:10.1002/ana.21549
Weigel, M. (2015). Extended phase graphs: Dephasing, RF pulses, and echoes - pure and simple.
Journal of Magnetic Resonance Imaging, 41(2), 266-295. doi:doi:10.1002/jmri.24619
Wright, P. J., Mougin, O. E., Totman, J. J., Peters, A. M., Brookes, M. J., Coxon, R., . . . Gowland,
P. A. (2008). Water proton T 1 measurements in brain tissue at 7, 3, and 1.5T using IR-
EPI, IR-TSE, and MPRAGE: results and optimization. Magnetic Resonance Materials in
Physics, Biology and Medicine, 21(1), 121. doi:10.1007/s10334-008-0104-8
Wu, W. C., St Lawrence, K. S., Licht, D. J., & Wang, D. J. (2010). Quantification issues in arterial
spin labeling perfusion magnetic resonance imaging. Top Magn Reson Imaging, 21(2), 65-
73. doi:10.1097/RMR.0b013e31821e570a
Xiao, H., Chenyang, X., & Prince, J. L. (2003). A topology preserving level set method for
geometric deformable models. IEEE Transactions on Pattern Analysis and Machine
Intelligence, 25(6), 755-768. doi:10.1109/TPAMI.2003.1201824
Xie, Y., Yang, Q., Xie, G., Pang, J., Fan, Z., & Li, D. (2016). Improved black-blood imaging using
DANTE-SPACE for simultaneous carotid and intracranial vessel wall evaluation.
Magnetic Resonance in Medicine, 75(6), 2286-2294. doi:10.1002/mrm.25785
Yoneyama, M., Nakamura, M., Tabuchi, T., Takemura, A., & Obara, M. (2012). Optimization of
3D-variable refocusing flip angle RARE imaging for high-resolution volumetric black-
blood angiography. Radiol Phys Technol, 5(2), 270-276. doi:10.1007/s12194-012-0161-6
Yushkevich, P. A., Piven, J., Hazlett, H. C., Smith, R. G., Ho, S., Gee, J. C., & Gerig, G. (2006).
User-guided 3D active contour segmentation of anatomical structures: Significantly
improved efficiency and reliability. NeuroImage, 31(3), 1116-1128.
doi:https://doi.org/10.1016/j.neuroimage.2006.01.015
84
Zaharchuk, G., Gong, E., Wintermark, M., Rubin, D., & Langlotz, C. P. (2018). Deep Learning in
Neuroradiology. AJNR Am J Neuroradiol. doi:10.3174/ajnr.A5543
Zhang, Z., Fan, Z., Kong, Q., Xiao, J., Wu, F., An, J., . . . Zhuo, Y. (2019). Visualization of the
lenticulostriate arteries at 3T using black-blood T1-weighted intracranial vessel wall
imaging: comparison with 7T TOF-MRA. Eur Radiol, 29(3), 1452-1459.
doi:10.1007/s00330-018-5701-y
Zheng, R., Li, Y., Sun, H., Lu, X., Sun, B. F., Wang, R., . . . Wang, H. (2016). Deep RNA
sequencing analysis of syncytialization-related genes during BeWo cell fusion.
Reproduction. doi:10.1530/REP-16-0343

85
APPENDIX A
Accrued Interpulse Phase with Laminar Flow Distribution
Laminar flow distribution p(v), or the area of the blood vessel cross-section containing
blood traveling at velocities between v and v+dv, can be incorporated to determine the interpulse
phase accrual Δφ(n).
Using Eq. [2] in Eq. [1], Δφ(n) can be written as:

[A1]

Considering that there should only be velocities at or less than vmax, the velocity function v in [A1]
can be solved such that
[A2]
[A3]
Therefore, Eq. [1] in Chapter 2 still applies in the case of a laminar flow distribution, with v being
the average velocity of blood flow or half of vmax.

Abstract (if available)

Abstract Cerebral small vessel disease (SVD) is the most common vascular cause of dementia and the cause of about one fifth of all strokes worldwide. Still, the underlying mechanisms of SVD remain poorly understood, resulting in no specific guidelines for its staging, treatment, and prevention. The knowledge gap in cerebral SVD is partly because cerebral small vessels, including arterioles and capillaries, are largely inaccessible to existing, clinically available in vivo imaging technologies. Prior development in magnetic resonance imaging (MRI) techniques has demonstrated the feasibility of non-invasively visualizing cerebral small vessels, such as the lenticulostriate arteries (LSAs), using Time of Flight (“bright blood”) MR angiography at ultra-high field strengths. However, this flow-related enhancement or “bright blood” technique cannot capture the finer details at distal portions of the small vessels, and thus does not translate well to lower field strength, which is more commonly used in the clinical setting. This dissertation aims to bridge the gap between ultra-high field imaging and clinical implementation of novel imaging technologies by systematically optimizing high-resolution 3D black blood MRI on both 3T and 7T scanners to directly visualize the small lenticulostriate arteries and developing automatic tools to characterize the morphological differences that occur with aging and/or vascular risk factors. ❧ In this dissertation, Chapter 1 provides a general introduction to the research outlined in the following chapters. It covers the clinical significance of cerebral small vessel disease in the progression of vascular cognitive impairment and dementia, the current state of neurovascular imaging, and the emergence of machine learning for segmentation problems. Chapter 2 describes a study in which a high-resolution 3D T1-weighted turbo spin echo sequence with a variable flip angle scheme (T1w TSE-VFA) was optimized to improve the contrast of the LSAs through “black blood” imaging. A greater number of LSA branches can be detected compared to those by time-of-flight MR angiography (TOF MRA) at 7T. The CNR of LSAs was comparable between 7T and 3T. T1w TSE-VFA showed significantly higher CNR than TOF MRA at the stem portion of the LSAs branching off the medial middle cerebral artery. Chapter 3 presents the development of LSA segmentation methods for quantification of morphological differences with age. This study not only used manual segmentation for identifying age differences, but it also used the manual segmentation labels as supervision in an application of deep learning in order to more accurately and automatically perform the vessel segmentation task. The number of detected LSAs by both TSE-VFA and TOF MRA was significantly reduced in aged subjects, while the mean vessel length measured on 7T TSE-VFA showed significant difference between the two age groups. This study also demonstrated the feasibility of high performance automatic small vessel segmentation from optimized black blood MR images using deep learning. Lastly, Chapter 4 describes a study in which the black blood MR images were evaluated for clinical utility in the staging of cerebral small vessel disease. A four-point LSA delineation rating that qualitatively assesses LSA branch number and tortuosity was developed and tested for performance in distinguishing patients with higher vascular risk and subsequent reduced cognitive function. In the unique Los Angeles Latino Eye Study cohort, patients with poorer LSA quality tended to exhibit reduced cognitive flexibility and executive function, and they were more frequently diagnosed with very mild dementia (Global Clinical Dementia Rating 0.5). Taken together, findings from these studies indicate that directly looking at the small vessels can provide insight into the earlier vascular changes that could lead to cognitive decline in vascular dementia. Rather than relying on imaging features that occur as a later consequence of small vessel disease, there is value in observing the morphology and understanding how it may indicate neurovascular degeneration in the presence of vascular risk factors.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

3D vessel mapping techniques for retina and brain as an early imaging biomarker for small vessel diseases

PDF

Mapping water exchange rate across the blood-brain barrier

PDF

Improving the sensitivity and spatial coverage of cardiac arterial spin labeling for assessment of coronary artery disease

PDF

Quantitative MRI for oxygenation imaging in cerebrovascular diseases

Asset Metadata

Creator Ma, Samantha Jenny (author)

Core Title Characterization of lenticulostriate arteries using high-resolution black blood MRI as an early imaging biomarker for vascular cognitive impairment and dementia

School Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Biomedical Engineering

Publication Date 02/28/2020

Defense Date 12/10/2019

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag aging,black blood MRI,deep learning,lenticulostriate artery,magnetic resonance imaging,morphology,OAI-PMH Harvest,subcortical vascular dementia,ultrahigh magnetic field

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Wang, Danny J.J. (committee chair), Nayak, Krishna (committee member), Shi, Yonggang (committee member), Toga, Arthur (committee member), Yan, Lirong (committee member)

Creator Email masj@usc.edu,samanthajennyma@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-275978

Unique identifier UC11674112

Identifier etd-MaSamantha-8219.pdf (filename),usctheses-c89-275978 (legacy record id)

Legacy Identifier etd-MaSamantha-8219.pdf

Dmrecord 275978

Document Type Dissertation

Rights Ma, Samantha Jenny

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA