University of Southern California Dissertations and Theses

Mapping water exchange rate across the blood-brain barrier

(USC Thesis Other)

Mapping water exchange rate across the blood-brain barrier

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Mapping water exchange rate across the blood-brain barrier
by
Xingfeng Shao
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulllment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(Biomedical Engineering)
May 2019
Copyright 2019 Xingfeng Shao
Dedication
This dissertation is dedicated to my mom and dad, especially to my wife Yvonne.
ii
Acknowledgements
First of all, I would like to express my special thanks to my advisor Prof. Danny JJ Wang
for recruiting me as his Ph.D. student and for the continuous motivation and encouragement
during the past ve years. His guidance not only helped me develop the skills for conducting
MRI research, but also cultivated my ability to unearth the solution for concrete problems. His
immense knowledge in physics and neuroscience always benets me and broadens my horizons. I
would also like to thank the rest of my dissertation committee members: Prof. Krishna Nayak,
Prof. Vasilis Marmarelis, and Prof. Lirong Yan, for reviewing the manuscript and providing
insightful comments with their prociency and expertise.
I would like to thank all the students and research fellows in Dr. Wang's lab, in USC Stevens
Neuroimaging and Informatics Institute, in USC BME department and in UCLA MRI lab, for the
enlightening discussions in medical physics and for the support for conducting experiments. It is
your encouragement helped me get out of the hard times during the past ve years, and you have
witnessed my growth in research and changes in life. I would also like to thank my alma mater
Tsinghua Univerity for the precious gift she brought to me: the ability of critical and independent
thinking. I'm also grateful to the International Society of Magnetic Resonance in Medicine for
the education and training. It is a great honor for me to be a part of the society and witness the
rapid development of MRI techniques and its indispensable eect on modern medicine.
And nally, I would like to thank my parents for their understanding and encouragement
throughout my Ph.D study in the U.S. This dissertation cannot be nished without their moral
and material support. To my beloved wife Yvonne, you make my life so delightful and I would
iii
never nish my Ph.D. without your tremendous support and love. Thank you for appearing in
my life and accompanying me all the time.
iv
Curriculum Vitae
Education:
- Ph.D., Biomedical Engineering, University of Southern California, CA, US (2016-2019)
- M.S., Bioengineering, University of California, Los Angeles, CA, US (2014-2016)
- B.Eng., Engineering Physics, Tsinghua University, Beijing, China (2010-2014)
Awards:
- Young investigator award (W. S. Moore award) nalist, ISMRM 26th Annual Meeting.
(2018)
- Summa Cum Laude Merit Award, ISMRM 24th/26th Annual Meeting. (2016, 2018)
- Magna Cum Laude Merit Award, ISMRM 26th Annual Meeting. (2018)
Oral presentations:
- A constrained slice-dependent background suppression scheme for simultaneous multi-slice
pseudo-continuous arterial spin labeling. Power Pitch presentation, ISMRM 2016.
- Prospective motion correction for 3D GRASE pCASL with volumetric navigators. ISMRM
2017.
- Measuring human placental blood
ow with multi-delay 3D GRASE pseudo-continuous ar-
terial spin labeling at 3T. Young investigator award competition, ISMRM 2018.
v
- Improved reliability for mapping water exchange across blood-brain barrier by diusion
prepared three-dimensional pseudo-continuous arterial spin labeling. ISMRM 2018.
- Quantication of
ow hemodynamics using non-contrast enhanced 4-dimensional dynamic
magnetic resonance angiography. ISMRM 2018.
- Technological innovations of arterial spin labeling'. MGH/HST Martinos Center for Biomed-
ical Imaging, 2018.
- 7T high-resolution arterial spin labeling reveals layer dependent cerebral blood
ow. ISMRM
2019.
- Mapping water exchange across the blood-brain barrier using three-dimensional diusion-
prepared arterial spin labeled perfusion MRI. Power Pitch presentation, ISMRM 2019.
Journal publications:
- Wang Y, Shao X, Martin T, et al. Phase-cycled simultaneous multi-slice balanced SSFP
imaging with CAIPIRINHA for ecient banding reduction [J]. Magnetic resonance in medicine,
2016, 76(6): 1764-1774.
- Shao X, Wang Y, Moeller S, et al. A constrained slice-dependent background suppression
scheme for simultaneous multi-slice pseudo-continuous arterial spin labeling [J], Magnetic
Resonance in Medicine, 2018, 79 (1), 394-400.
- Yu S, and Shao X, et al. ASPECTS Based Reperfusion Status on Arterial Spin Labeling Is
Associated with Clinical Outcome in Acute Ischemic Stroke Patients [J]. Journal of Cerebral
Blood Flow and Metabolism, 2018, 38 (3), 382-392.
- Shao X, Liu D, Martin T, et al. Measuring human placental blood
ow with multi-delay 3D
GRASE pseudocontinuous arterial spin labeling at 3T [J]. Journal of Magnetic Resonance
Imaging, 2018, 47 (6), 1667-1676.
vi
- Martin T, Wang Y, Rashid S, Shao X, et al. Highly Accelerated SSFP Imaging with Con-
trolled Aliasing in Parallel Imaging and integrated-SSFP (CAIPI-iSSFP) [J]. Investigative
Magnetic Resonance Imaging, 2017, 21(4): 210-222.
- Shen Y, Yan L, Shao X, et al. Improved sensitivity of cellular MRI using phase-cycled
balanced SSFP of ferumoxytol nanocomplex-labeled macrophages at ultrahigh eld[J]. In-
ternational journal of nanomedicine, 2018, 13: 3839.
- Shao X, Ma JS, Casey M, DOrazio L, Ringman J and Wang DJJ. Mapping water ex-
change across the blood-brain barrier using three-dimensional diusion-prepared arterial
spin labeled perfusion MRI [J]. Magnetic Resonance in Medicine, 2018.
- Hu HH, Rusin JA, Peng R, Shao X, et al. Multi-Phase 3D Arterial Spin Labeling Brain
MRI in Assessing Cerebral Blood Perfusion and Arterial Transit Times in Children at 3T[J].
Clinical Imaging, 2019, 53, 210-220.
- Shao X, Zhao Z, Wang DJJ, and Yan L. Quantication of cerebrovascular hemodynamics
using non-contrast enhanced four-dimensional dynamic magnetic resonance angiography[J].
Magnetic Resonance in Medicine, 2019.
- Sung K, Liu D, Shao X, Danyalov A, Chanlaw T, Wang DJJ, Janzen C, and Devaskar
SU.Human Placenta Perfusion During Early Gestation with Pseudo-Continuous Arterial
Spin Labeling MRI [J]. Radiology (in revision).
- Ma SJ, Sarabi MS, Yan L, Shao X, Chen Y, Yang Q, Jann K, Toga A, Shi Y, and Wang
DJ, Characterization of Lenticulostriate Arteries with High Resolution Black-blood T1-
weighted Turbo Spin Echo with Variable Flip Angles at 3 and 7 Tesla [J]. Neuroimage (in
resubmission).
vii
Table of Contents
Dedication ii
Acknowledgements iii
Curriculum Vitae v
List Of Tables x
List Of Figures xi
List Of Abbreviations xviii
Abstract xx
Chapter 1: Introduction 1
1.1 The blood-brain barrier (BBB) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Aquaporin (AQP): the water channel . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 AQP1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.2 AQP4: Maintaining Brain Homeostasis . . . . . . . . . . . . . . . . . . . . 5
1.2.3 AQP4: Glymphatic Function and Clearance of Deleterious
Proteins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Cerebral small vessel disease (SVD) . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Chapter 2: BBB permeability 13
2.1 BBB disruption and leakage of plasma constituents . . . . . . . . . . . . . . . . . . 14
2.1.1 Mechanisms contributing to the BBB leakage . . . . . . . . . . . . . . . . . 14
2.1.2 Leakage of plasma through the BBB causes neuronal damage . . . . . . . . 15
2.2 The role of pericytes in BBB permeability . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Bidirectional water permeability change after BBB
disruption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4 MRI techniques assessing BBB permeability . . . . . . . . . . . . . . . . . . . . . . 21
2.4.1 Dynamic susceptibility contrast MRI . . . . . . . . . . . . . . . . . . . . . . 21
2.4.2 Dynamic contrast enhanced MRI . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4.3 Water: an endogenous tracer . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Chapter 3: When diusion meets perfusion 31
3.1 Microcirculation of the capillary network and intra-voxel incoherent motion (IVIM) 32
3.2 Aterial spin labling (ASL) and perfusion signals . . . . . . . . . . . . . . . . . . . . 34
3.3 MRI pulse sequence design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4 Diusion weighted perfusion signal . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
viii
Chapter 4: Improved SPA modeling of kw 43
4.1 Modeling of water exchange rate kw across BBB . . . . . . . . . . . . . . . . . . . 43
4.2 Estimation of kw with total generalized variation
regularized SPA model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.3 Test and retest reproducibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Chapter 5: Initial evaluation of water exchange across the BBB in elder subjects
at risk of SVD 52
5.1 Clinical assessments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.2 Correction between kw and vascular risk factors . . . . . . . . . . . . . . . . . . . . 53
5.3 Correction between kw and cognitive measurements . . . . . . . . . . . . . . . . . 56
5.3.1 CDR and MoCA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.3.2 NIH toolbox . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.4 Correction between kw and WMH measurements . . . . . . . . . . . . . . . . . . . 62
Chapter 6: Preliminary comparison of water permeability and BBB permeability
to contrast agents 65
6.1 DCE-MRI acquisition and modeling of Ktrans . . . . . . . . . . . . . . . . . . . . . 66
6.2 Correlation between BBB permeability to water (kw) and contrast agent (Ktrans) 68
6.3 Correlation between regional BBB permeability and cognitive measurements . . . . 71
Chapter 7: Discussion and ongoing work 77
7.1 Limitations of the current study . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
7.2 Temporal regularized TGV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
7.3 Alternative methods combing diusion weighting and TSE based sequences . . . . 86
7.4 Deep learning for ASL de-noising . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
7.5 Animal studies to evaluate kw changes in dierent
disease models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
7.5.1 BBB disruption in neurological disease . . . . . . . . . . . . . . . . . . . . . 92
7.5.2 Transient BBB opening using focused ultrasound (FUS) and
microbubbles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
7.5.3 Pericyte-decient and AQP4-decient model . . . . . . . . . . . . . . . . . 94
7.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
Reference List 96
Appendix A
Optimization of the timing of diusion gradients to minimize eddy current artifacts . . 107
Appendix B
Estimation of voxel-wise R
1b
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
ix
List Of Tables
4.1 Average kw and ICC values of test and retest measurements in 8 ROIs related to AD 50
5.1 Summary of clinical assessments performed in this study. . . . . . . . . . . . . . . 54
5.2 Repeated measures mixed-eects linear regression coecients for kw and clinical
measurements. P values are listed in the parentheses. Signicant correlations with
P values smaller than 0.05 and 0.005 are indicated by asterisks. . . . . . . . . . . . 56
5.3 Characterizing level of dementia using CDR scores . . . . . . . . . . . . . . . . . . 58
5.4 Repeated measures mixed-eects linear regression coecients for kw and UDS
measurements. P values are listed in the parentheses. Signicant correlations with
P values smaller than 0.05 and 0.005 are indicated by asterisks. . . . . . . . . . . . 59
5.5 Repeated measures mixed-eects linear regression coecients for kw and NIH
toolbox measurements. P values are listed in the parentheses. Signicant correla-
tions with P values smaller than 0.05 and 0.005 are indicated by asterisks. . . . . . 60
5.6 Repeated measures mixed-eects linear regression coecients for kw and WMH
measurements. P values are listed in the parentheses. Signicant correlations with
P values smaller than 0.05 and 0.005 are indicated by asterisks. . . . . . . . . . . . 62
6.1 Regional kw and Ktrans values. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.2 Correlation coecients between regional kw and Ktrans measurements (P values
were listed in brackets). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
x
List Of Figures
1.1 The cell associations at the BBB. The cerebral endothelial cells form tight junctions
at their margins which seal the aqueous paracellular diusional pathway between
the cells. Pericytes are distributed discontinuously along the length of the cerebral
capillaries and partially surround the endothelium. Both the cerebral endothe-
lial cells and the pericytes are enclosed by, and contribute to, the local basement
membrane which forms a distinct perivascular extracellular matrix (basal lamina
1, BL1), dierent in composition from the extracellular matrix of the glial endfeet
bounding the brain parenchyma (BL2). Foot processes from astrocytes form a com-
plex network surrounding the capillaries and this close cell association is important
in induction and maintenance of the barrier properties. (Figure from Abbott N J,
et al. Neurobiology of disease, 2010, 37(1): 13-25.) . . . . . . . . . . . . . . . . . . 3
1.2 Double immunolabeling of AQP4 (red) and glial brillary acidic protein (GFAP)
(green) in cortex. AQP4 immunolabeling reveals that the entire network of ves-
sels, including capillaries, is covered by astrocytic processes, albeit GFAP negative.
Smaller vessels and capillaries are mostly GFAP negative but display intense la-
beling against the astrocyte-specic channel AQP4. The AQP4 labeling reveals
continuous coverage by astrocytic end feet. (From Simard M. et al. J Neurosci,
2003, 23: 9254-9262.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Schematic of blood water transfer from intra-vascular space to extra-vascular tissue
space. A. Labelled blood water molecules exchange into the extra-vascular space via
all water transport mechanisms including aquaporin-4 (AQP4) water channels in
wild-type (WT) mice. B. In AQP4-decient mice, the water transport mechanisms
are restricted and only occur through cotransport with organic molecules and by
diusion through the lipid bilayer of the plasma membrane. (Figure adapted from
Ohene Y, et al. NeuroImage, 2019, 188: 515-523.) . . . . . . . . . . . . . . . . . . . 7
1.4 Electron micrographs show how Amyloid- deposition is associated with loss of
astrocytic polarity at 8 months. A. Normal linear distribution of AQP4 in wild type,
gold particles indicated by arrows. B. AQP4 (gold particles indicated by arrows)
lose their polarity as Amyloid- (asterisk) accumulates around the vessel (CAA).
C. Increase in AQP4-immunoreactivity in the non-endfeet membranes, indicated by
arrows. (Compare with wild type littermate in A) Abbreviations: Ast, astrocyte;
E, endothelial cell; GFAP, glial brillary acidic protein; L, lumen. Scale bars, 0.5m.
(Figure from Yang J, et al. Journal of Alzheimer's Disease, 2011, 27(4): 711-722.) 9
xi
1.5 Major mechanisms underlying vascular cognitive impairment (VCI). A. Vascular
causes. B. Brain parenchymal lesions associated with VCI. (Figure from Dichgans
M, et al. Circulation research, 2017, 120(3): 573-591.) . . . . . . . . . . . . . . . . 11
1.6 Imaging of Small Vessel Disease: example ndings (upper), schematic representa-
tion (middle) and a summary of imaging characteristics (lower) of MRI features
for changes related to small vessel disease. (Figure from Shi Y, et al. Stroke and
vascular neurology, 2016, 1(3): 83-92.) . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.1 Illustration of how BBB becomes more permeable, causing damage to surrounding
neurons and glial cells. (Figure from Wardlaw J M et al, Stroke, 2003, 34(3):
806-812.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Changes at the neurovascular unit caused by pericyte deciency. In the absence of
pericytes (P, orange) the endothelium (E, pink) shows convoluted junctions (J) and
increased formation of pinocytotic vesicles (arrowheads). Pericyte-deciency aects
polarization of astrocyte end-feet (AF, blue), shown by mislocalization of aquaporin
4 (AQP4, red) and defective deposition of astrocyte-derived basement membrane
(aBM, dark brown). Deposition of endothelium-derived basement membrane (eBM,
beige) is not aected by pericyte-deciency. (Figure from Armulik A, et al. Nature,
2010, 468(7323): 557.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Pericyte coverage correlates with BBB integrity. a, b, Quantication of pericyte
coverage of capillaries in the cerebral neocortex of adult Pdgfb
ret=ret
mice (a),
and R26P
+=0
and R26P
+=+
mice (b). c, d, Capillary diameter in the cerebral
neocortex of adult Pdgfb
ret=ret
mice (c), and R26P
+=0
and R26P
+=+
mice (d),
with or without pericyte coverage. e, Three-dimensional reconstructions of confocal
image z-stacks of adult cerebral neocortex vasculature depicted by collagen IV
(basement membrane) and CD13 (pericyte) staining inPdgfb
ret=ret
, R26P
+=0
and
R26P
+=+
mice. f, Wet/dry weight ratios of control and Pdgfbret/ret mice. g, h,
Quantication of Evans blue in Pdgfb
ret=ret
mice (g) and R26P
+=0
and R26P
+=+
mice (h) in the cerebrum after 16 h of circulation. i, Time course of Evans blue
accumulation in the cerebrum ofPdgfb
ret=ret
animals. y-axis shows optical density
(OD) at 620 nm per gram of tissue. *P<0.03; **P<0.007; ***P<0.0005. All error
bars show means.e.m. (Figure from Armulik A, et al. Nature, 2010, 468(7323):
557.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4 A
ow diagram demonstrating the Pathological progress in the WM induced by
pericyte loss. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5 Expression of AQP4 mRNA in mesencephalic perivascular in ovariectomized ani-
mals. Bright-eld high-magnication photographs showing AQP4 mRNA expres-
sion in perivascular glial processes in control (A), and LPS-injected animals at 6 hr
(B), 24 hr (C), 48 hr (D), and 14 days (E) after endotoxin injection. Scale bar = 50
um. The highest perivascular AQP4 mRNA expression was found at 6 hr and was
progressively decreasing afterwards. (Figure adapted from Toms-Camardiel M, et
al. Journal of neuroscience research, 2005, 80(2): 235-246.) . . . . . . . . . . . . . 22
xii
2.6 Dynamic MRI parametric maps of a patient with medulloblastoma. Circles indicate
region of interest (ROI). The top row (a-e) is preoperative images. The conventional
post GBCA T1-weighted MRI (a) shows a homogenously enhancing tumor in the
posterior fossa (arrow). DSC MRI parametric maps (b-c) demonstrate elevated
vascularity in the tumor. DCE MRI parametric maps (d-e) demonstrate elevated
Ktrans and heterogenous Ve. The bottom row (f-j) was obtained postoperatively
and demonstrates complete tumor resection and substantially reduced vascularity
and permeability. (Figure is adapted from Thompson E M, et al. Journal of neuro-
oncology, 2012, 109(1): 105-114.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.7 MRI and immunostaining results in TgF344-AD and wild-type rats. (a) PSw
is signicantly higher (up to 2-fold) in TgF344-AD rats relative to wild-types
(p<0.05; two-way ANOVA). (b) Trans-BBB leakage of contrast agent (Ktrans) is
unaltered between TgF344-AD rats and wild-types (p=0.477; two-way ANOVA).
(c) Occludin is reduced in TgF344-AD relative to wild-types (p<0.05; two-way
ANOVA), corresponding well with genotype dierences in PSw. (d) No detectable
TgF344AD/wild-type dierences were observed for aquaporin-4 (AQP4). (e) Rep-
resentative occludin and lectin immuno-stains. Aspecic staining of amyloid- was
visually identied on the lectin snapshots and manually segmented as shown. (Fig-
ure adapted from Dickie B R, et al. NeuroImage, 2019, 184: 349-358.) . . . . . . . 26
2.8 Row I and II: M and M (non-labeled ASL) signals from multi-b data tted well
to the bi-exponential model (n=8, R
2
= 0.997, 0.998 respectively). Using the full
model, A1 and kw were estimated to be 0.094 (95% Condence bounds: 0.053, 0.14)
and 70min
1
(95% Condence bounds: 59 min
1
, 86min
1
) respectively. Using
the simplied two-b model, A1 and kw were estimated to be 0.140.02 and 586
min
1
respectively. Row III shows the A1 and kw map before and after mannitol,
along with the Ktrans map and Evans blue brain slice. Mannitol caused BBB
disruption laregely to one hemisphere. kw in the aected hemisphere decreased
to 102 min
1
after mannitol (P <0.01, n=5) (row IV). Ktrans and Evans blue
staining showed similar patterns of BBB disruption. (Figure from Tiwari YV, et
al, Proceedings of ISMRM, p0792, 2015.) . . . . . . . . . . . . . . . . . . . . . . . 29
2.9 Average M signal (symbols) acquired at four post-labeling delays plotted as a
function of diusion-weighting strength. Data were averaged across all pixels in
grey matter (a), or white matter (b), and across four subjects. Each subjects
data were normalized to the signal without diusion weighting (b=0). The errors
bars represent the standard error. The solid lines represent the best t of a bi-
exponential decay model to each M series. (Figure from St. Lawrence K S, et
al. Magnetic Resonance in Medicine, 2012, 67(5): 1275-1284.) . . . . . . . . . . . . 30
2.10 DW-ASL data acquired with PLD = 1.5 s. First row: average ASL signal with-
out diusion weighting DM(b0); Second row: average ASL signal with diusion
weighting DM(bDW); Third row: the ratio images DM(bDW)/DM(b0); Last row:
kw images. All images were smoothed with a gaussian lter with a kernel size of 15
mm for the DW-ASL (kw) images. (Figure from St. Lawrence K S, et al. Magnetic
Resonance in Medicine, 2012, 67(5): 1275-1284.) . . . . . . . . . . . . . . . . . . . 30
xiii
3.1 Molecular diusion and blood pseudo-diusion. Molecular diusion (left) is a ran-
dom process at individual molecular level resulting in a Gaussian distribution of
molecular displacement (free diusion). Pseudo-diusion for blood
ow results from
the collective water
ow in the randomly orientated capillary segments. l is average
displacements, v is velocity and D/D* are diusion coecient and pseudo-diusion
coecient. (Figure from Le Bihan D. NeuroImage, 2017 (in press).) . . . . . . . . 33
3.2 (a) Experiment setup of pCASL scan. Labeling plane (yellow line) is typically
placed perpendicular to the carotid below the cerebellum. Incoming blood is tagged
by
ow driven adiabatic inversion and imaged after entering capillary/tissue space
at PLD. Blue box indicates the imaging volume. (b) Perfusion signal is obtained by
subtraction of control (without ASL tagging) and label (with ASL tagging) images. 35
3.3 Capillary-tissue unit dened by the one-barrier distributed parameter (1BDP)
model. The inner cylinder represents the capillary space and the outer cylinder
represents surrounding brain tissue. The capillary space has cross-sectional area
A
c
, volume V
c
, and tracer concentration C
c
(x;t), where x is the spatial dimension
along the length of the cylinder. The surrounding brain tissue has cross-sectional
area A
b
, volume V
b
, and tracer concentration C
b
(x;t). A permeable membrane
that represents the blood-brain barrier separates the two spaces: exchange of tracer
across the blood-brain barrier is characterized by the PS product. Labeled water

ows into the capillary-tissue unit via arterial blood at a concentration C
a
(t), and
exits via venous blood at a concentration C
v
(t). The rate of blood
ow into the
capillary space is F. (Figure from St. Lawrence K S, et al. MRM, 2000, 44(3):
440-449.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4 (a) Sequence scheme of 3D DW-pCASL. (b) Diusion preparation module: Non-
selective pulses were used to compensate eld inhomogeneity, timing of gradients
was optimized to minimize eddy current. De-phasing gradient was added along y-
axis (4 dephasing per voxel) before tip-up to eliminate phase sensitivity of GRASE
readout. Strong spoiler along three axes were added after tip-up to remove residual
transverse magnetization. (c) GRASE readout: Non-selective excitation was used
to improve the slab prole, re-phasing and rewound de-phasing gradients were
added at two sides of EPI readout to maintain MG condition. . . . . . . . . . . . . 38
3.5 Perfusion map with 6 diusion weightings acquired at PLD = 1500, 1800, and 2100
ms, respectively. Gray scale indicates relative perfusion signal intensity compared
to average perfusion signal acquired with b = 0 s=mm
2
at PLD = 1500 ms. . . . . 40
3.6 Average perfusion signals from 4 subjects with 6 diusion weightings acquired at
PLD = 1500, 1800, and 2100 ms. Error bar indicates the standard deviation of
kw measurements across 4 subjects. Biexponential tting results are shown in the
upper right corner. Capillary/tissue fraction were 24%/76% when PLD = 1500 ms,
15%/85% when PLD = 1800 ms, and 11%/89% when PLD = 2100 ms, respectively 41
xiv
4.1 Comparison of direct modeling with Gaussian smoothing (rst row) and regular-
ized SPA modeling (second row). (a) Perfusion map without diusion weighting
acquired at PLD = 900 ms. (b) Perfusion map with b = 14 s=mm
2
(VENC =
7.5 cm/s to suppress vascular signal) acquired at PLD = 900 ms. (c) ATT map.
(d) Perfusion map without diusion weighting acquired at PLD = 1800 ms. (e)
Perfusion map with b = 50s=mm
2
acquired at PLD = 1800 ms. (f) kw map. Red
arrows indicate the local regions with noise induced spuriously high kw values using
direct modeling (rst row). kw map from regularized SPA modeling was relatively
smooth (second row). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2 kw map of 6 slices from one representative subjects test and retest scans . . . . . . 48
4.3 (a) Average kw values from test-retest experiments using the proposed 3D DP
pCASL sequence. Horizontal and vertical axis indicates the kw measurements
from the rst and second MRI scan, respectively. (b) Average global CBF val-
ues from test-retest experiments. Horizontal and vertical axis indicates the CBF
measurements from the rst and second MRI scan, respectively . . . . . . . . . . . 49
4.4 Average kw values from test-retest experiments using the 2D DW-pCASL sequence.
Horizontal and vertical axis indicates the kw measurements from the rst and
second MRI scan, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.1 (a,b) Bar plot of average kw in normal subjects versus subjects with diabetes (a)
and hypercholesterolemia (b). (c) Bar plot of average kw versus vascular risk
factors. Error bars indicate standard deviation of kw across subjects. . . . . . . . . 57
5.2 (a,b) Bar plot of average kw versus clinical dementia rating scales CDR-SB (a) and
CDR-GS (b). Error bars indicate standard deviation of kw across subjects. . . . . 60
5.3 (a-d) Scatter plots of average kw versus NIH toolbox measurements: Flanker (a),
DCCS (c), PSMTa (b), and PSMTb (d). Slopes andR
2
of linear regressions (with-
out controlling age/sex, indicated by the black dashed lines) are listed in each
scatter plot. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.4 Bar plot of average kw versus Fazekas scale. Error bars indicate standard deviation
of kw across subjects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.5 Scatter plot of average kw versus CBF. kw and CBF from both test and retest
experiments are displayed in the scattor plot. . . . . . . . . . . . . . . . . . . . . . 63
6.1 A representative slice of T1 (a) and Ktrans map measured from a 63-year-old subject. 67
6.2 Scatter plot between regional kw measurements and average Ktrans in MCA per-
forator territory. Dashed lines represent the tted linear regression curve. . . . . . 69
6.3 Scatter plot between regional kw measurements and average Ktrans in Caudate.
Dashed lines represent the tted linear regression curve. . . . . . . . . . . . . . . . 70
6.4 Scatter plots of regional Ktrans (a-f) and kw (g-l) versus Flanker. Dashed lines
represent the tted linear regression curve. . . . . . . . . . . . . . . . . . . . . . . . 74
xv
6.5 Scatter plots of regional Ktrans (a-f) and kw (g-l) versus PSMTa. Dashed lines
represent the tted linear regression curve. . . . . . . . . . . . . . . . . . . . . . . . 75
6.6 Scatter plots of regional Ktrans (a-f) and kw (g-l) versus DCCS. Dashed lines
represent the tted linear regression curve. . . . . . . . . . . . . . . . . . . . . . . . 76
7.1 Two under-sampling patterns and corresponding aliasing appearances are shown
in (a) and (b). Dashed circles indicate un-acquired k-space lines. 2-fold under-
sampling in both phase (ky) and partition (kz) encoding directions results in a
total acceleration factor of 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
7.2 Perfusion map from (a) fully sampled data acquired in 4 segments and (b) single-
shot acquisition with 4-fold under-sampling. Total acquisition time was kept the
same for both protocols. Resolution = 333 mm
3
. . . . . . . . . . . . . . . . . . 83
7.3 Simultaneously calculated CBF, ATT and T1 map from single-shot multi-delay
pCASL. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
7.4 Time dependent 2D CAIPI under-sampling patterns for six consecutive repetitions.
Dashed circles indicate un-acquired k-space lines. 3-fold under-sampling in phase
(ky) and 2-fold under-sampling in partition (kz) encoding directions results in a
total acceleration factor of 6. Spatial and temporal incoherence is maximized by
shifting under-sampling patterns between repetitions . . . . . . . . . . . . . . . . . 85
7.5 A single slice of ASL perfusion images from 18 label/control pairs acquired at PLD
= 1800 ms with diusion weighting (b = 50 s=mm
2
) and GRASE readout. Diu-
sion weighting was induced by (a) traditional bipolar diusion gradients and 120
0
refocusing pulses in GRASE, (b) diusion gradients with M1 motion compensation,
and (c) the proposed diusion preparation module. All images were acquired from
the same subjects who was asked to avoid head motion during the scan. Color scale
indicates relate perfusion signal intensity to average signal intensity of M0. . . . . 87
7.6 Schematic illustration of a CNN framework for ASL denoising. The rst convolu-
tional layer (conv) received two or three subtraction images as input, followed by
four convolutional layers in the global (orange arrow) and local (large purple arrow)
pathways. Dilation factors of four convolutional layers in the global pathway were
2, 4, 8 and 16. Parallel pathways were concatenated at a later step (green arrows).
Residual image was acquired by subtracting the ground truth (GT) from average of
input images. Number of lters is shown in italics. The residual learning approach
was used to speed the learning process and improve the performance of CNNs by
simplifying image generation (Figure from Kim K H, et al. Radiology, 2017, 287(2):
658-666.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
7.7 Schematic illustration of a CNN framework for SPA modeling. Inputs are perfusion
images with and without diusion weightings and a structural image. And output
are kw and ATT maps. The CNN works as a 'black box' which computes kw and
ATT based on pre-trained convolutional networks. . . . . . . . . . . . . . . . . . . 91
xvi
7.8 Typical T2, ADC, CBF, T1, A2, DCE subtraction, Ktrans maps and Evans blue
(EB) slices 2 days after stroke. The ROI denitions for normal (green) and infarcted
tissue (red) are drawn on T2 maps. Scale bar are: T2 (40-120 ms), T1 (1250-2500
ms), ADC (0-0.0001 s/mm
2
), CBF (0-1.5 ml/g/min), A2 (0.4-1.4), DCE subtrac-
tion (0-25,000 signal unit), KTrans (0-0.003 min
1
). (Figure from Tiwari Y V, et
al. JCBFM, 2017, 37(8): 2706-2715.) . . . . . . . . . . . . . . . . . . . . . . . . . . 93
xvii
List Of Abbreviations
ASL : : : : : : : Arterial spin labeling
AD : : : : : : : : Alzheimer's disease
ADC : : : : : : : Apparent diusion coecients
ATT : : : : : : : Arterial transit time
AQP : : : : : : : Aquporin
BBB : : : : : : : Blood-brain barrier
CBF : : : : : : : Cerebral blood
ow
CBV : : : : : : : Cerebral blood volume
CNS : : : : : : : Central nervous system
CSF : : : : : : : Cerebrospinal
uid
CAIPI : : : : : : Controlled aliasing in parallel imaging
CPMG : : : : : Carr Purcell Meiboom Gill
DCE : : : : : : : Dynamic contrast enhanced
DSC : : : : : : : Dynamic susceptibility contrast
Ew : : : : : : : : Water extraction ratio
xviii
GRASE : : : : : GRadient And Spin Echo
IVIM : : : : : : Intra-voxel incoherent motion
ICC : : : : : : : Intraclass correlation coecient
kw : : : : : : : : Water exchange rate
MRI : : : : : : : Magnetic resonance imaging
PCA : : : : : : : Principal component analysis
PET : : : : : : : Positron emission tomography
PLD : : : : : : : Post-labeling delay
PSw : : : : : : : Water permeability surface product
SVD : : : : : : : Small vessel disease
SPA : : : : : : : Single pass approximation
SNR : : : : : : : Signal-to-noise ratio
TJ : : : : : : : : Tight junction
TGV : : : : : : : Total generalized variation
VCID : : : : : : Vascular cognitive impairment and dementia
WMH : : : : : : White matter hyperintensities
WML : : : : : : White matter lesions
xix
Abstract
Blood-brain barrier (BBB) maintains the homeostasis within the brain and the dysfunction of
BBB has been linked to multiple central nervous system diseases and psychiatric disorders. In
this dissertation, we rst performed a thorough literature review on the biological mechanism
of water exchange across the BBB. Previous animal studies have shown that increased water
exchange occurs with loss of pericytes and before BBB leakage to contrast agent in Alzheimer's
Disease (AD), while other studies demonstrated decreased water exchange in ischemic stroke
and experiment induced BBB disruption. A hypothesis was proposed that changes of water
permeability are bidirectional when BBB is prone to leakage: water permeability increases at
early stage of BBB opening and decreases after chronic BBB leakage due to accumulated toxic
substances in extravascular space and brain parenchymal damage.
The purpose of this work is to present a novel MR pulse sequence and regularized modeling
algorithm to quantify the water exchange rate, kw, across the BBB without contrast, and to
evaluate its clinical utility in a cohort of elderly subjects at risk of cerebral small vessel disease
(SVD). Ongoing studies have shown preliminary results about correlations between regional kw
and Ktrans, a measurement of BBB leakage to contrast agent.
A diusion preparation module with spoiling of non-Carr-Purcell-Meiboom-Gill signals was in-
tegrated with pseudocontinuous arterial spin labeling and 3D gradient and spin echo readout. The
tissue/capillary fraction of the arterial spin labeling signal was separated by appropriate diusion
weighting (b = 50s=mm
2
). kw was quantied using a single-pass approximation model with total
generalized variation regularization. A cohort of elderly subjects were recruited and underwent
xx
two MRIs to evaluate the reproducibility of the proposed technique. Correlation analysis was
performed between kw and vascular risk factors, Clinical Dementia Rating scale, neurocognitive
assessments, and white matter hyperintensities.
The capillary/tissue fraction of ASL signal can be reliably dierentiated with the diusion
weighting of b = 50s=mm
2
, given 100-fold dierence between the (pseudo-)diusion coecients of
the 2 compartments. Good reproducibility of kw measurements (intraclass correlation coecient
= 0.75) was achieved. Average kw was 105.0 20.6, 109.6 18.9, and 94.1 19.6 min
1
for whole brain, gray and white matter. kw was increased by 28.2%/19.5% in subjects with
diabetes/hypercholesterolemia. Signicant correlations between kw and vascular risk factors,
Clinical Dementia Rating scale, executive/memory function, and the Fazekas scale of white matter
hyperintensities were observed. Through comparison with DCE-MRI, signicant correlation was
found between kw and Ktrans in caudate, which indicate BBB permeability to both water and
contrast agent could serve as imaging markers for cerebral small vessel disease. Our preliminary
results also suggest regional kw and Ktrans changes are sensitive to dierent aspects of cognitive
function, such as attention, episodic memory or cognitive
exibility.
In conclusion, a diusion prepared 3D pseudo-continuous arterial spin labeling sequence with
total generalized variation regularized single-pass approximation modeling was proposed to mea-
sure BBB water permeability non-invasively with good reproducibility. kw may serve as a surro-
gate imaging marker of cerebral SVD and associated cognitive impairment.
xxi
Chapter 1
Introduction
Advances in medical imaging benets the human beings for better understanding of anatomical
and the physiological processes of the body in both health and disease, and promotes the devel-
opment of diagnostic medicine. Conventional devices usually involve ionizing radiation to form
medical images, such as x-ray and radioactive tracers used by CT and positron emission tomog-
raphy (PET) scanners. With advances in quantum physics, magnetic resonance imaging (MRI)
systems have been developed using strong magnetic elds, gradients, and tuned radio frequency
waves to generate images of the organs in the body without exposing to radiation since 1970s.
With superior soft tissue contrast and cutting-edge pulse sequence designs, MRI has been widely
used in neuroimaging studies and is capable of capturing the subtle structural and functional
changes in brain.
The focus of this dissertation is the blood-brain barrier (BBB), and the question to answer
is when and how BBB permeability changes in neurodegenerative diseases. While BBB leakage
causes accumulation of toxic proteins in brain parenchymal and has been considered as a common
consequence of aging and neurological disorders, the goal of this study is to propose a surrogate
imaging biomarker of the underlying microvascular changes at the early stage of disease pro-
gresses. Water is an endogenous tracer and water exchange across the BBB is facilitated by water
channels. Through development of a novel MRI pulse sequence with innovative reconstruction
1
algorithm, reliable quantication of water permeability across the BBB, kw, can be achieved.
The eectiveness of the proposed technique has been thoroughly evaluated through comparison
with dynamic contrast enhanced (DCE)-MRI and clinical cognitive measurements in aged sub-
jects at risk of cerebral small vessel disease (SVD), and we have demonstrated that kw could be a
surrogate biomarker for vascular risks, white matter lesions (WML) and early cognitive decline.
The rst chapter introduced the physiological backgrounds about the BBB, aquaporins in
central nervous system (CNS), and cerebral SVD. The structure units of the BBB and their roles
in maintaining BBB function were brie
y discussed. Although BBB disruption typically involves
loss of tight junctions, subtle structural and functional changes in pericytes and astrocytes have
also shown high relevance with cognitive impairment, dementia and AD. During the past decades
it has been realized that water exchange across the BBB is mainly facilitated by water channels
instead of diusion through the lipid bilayer of the plasma membrane. Distribution of aquaporins
and their role in maintaining brain water balance and glymphatic functions were discussed in this
chapter. The last portion of this chapter introduced the potential mechanisms of cerebral SVD
and existing imaging biomarkers. Cerebral SVD is the most common vascular cause of dementia,
a major contributor to mixed dementia, and the cause of about 20% of all strokes worldwide.
Identifying surrogate biomarkers could be essentially benecial to guide early interventions to
prevent the onset of SVD or evaluating treatment eciency.
1.1 The blood-brain barrier (BBB)
There are three type of barriers between the blood and brain in CNS: the BBB, the blood-
CSF barrier (BCSFB) and the arachnoid barrier. All the barrier layers play important roles in
regulating substances exchange between blood and brain parenchymal. Among these barriers,
BBB is the most important functional and physical barrier and constitutes the largest interface
for blood-brain exchanges [1].
2
Figure 1.1: The cell associations at the BBB. The cerebral endothelial cells form tight junctions at
their margins which seal the aqueous paracellular diusional pathway between the cells. Pericytes
are distributed discontinuously along the length of the cerebral capillaries and partially surround
the endothelium. Both the cerebral endothelial cells and the pericytes are enclosed by, and
contribute to, the local basement membrane which forms a distinct perivascular extracellular
matrix (basal lamina 1, BL1), dierent in composition from the extracellular matrix of the glial
endfeet bounding the brain parenchyma (BL2). Foot processes from astrocytes form a complex
network surrounding the capillaries and this close cell association is important in induction and
maintenance of the barrier properties. (Figure from Abbott N J, et al. Neurobiology of disease,
2010, 37(1): 13-25.)
Figure 1.1 illustrates the basic structure units of BBB. BBB is created at the level of the
capillary endothelial cells and formed by tight junctions (TJs). Paracellular aqueous diusional
pathways to the macromolecules and polarized solutes (ions) are severely restricted by the TJs at
the margins of endothelial cells. Endothelium is surrounded by the pericytes, and both pericytes
and capillary endothelial cells are enclosed by the basement membrane, which forms the basal
lamina. The outer layer surrounding the capillaries (brain parenchyma) is formed by the astrocytes
endfeet, which is important to maintain the inductive properties of the BBB, which upregulates
the tight junction proteins and modulates the polarity of endothelial cells.
With ion channels and specic transporters, the BBB regulates the ion concentration and
provides optimal ion environment for synaptic signaling. More importantly, BBB separates the
neurotransmitters in the central and peripheral nervous systems, prevents macromolecules such
3
as albumin from entering the brain and shields the CNS from neurotoxic substances circulating
in the blood. Since adult CNS does not have a signicant regenerative capacity, integrity of BBB
is critical to eliminate the neurotoxins within the CNS, prevent any acceleration of neuronal cell
death and maintain normal neural functions [1]. Although the BBB prevents the majority of
large blood-borne molecules from entering the CNS, some large molecules and complexes, such
as proteins and peptides, can still be transported across the BBB via specic endocytotic or
transcytotic mechanisms [2]. BBB also has limited passive permeability to essential water soluble
nutrients. Specic transport channels are expressed in the BBB to ensure an adequate supply of
nutrients required by brain.
In healthy brain, intact BBB normally blocks the entering of molecules larger than 400 Da
(blood-bone neurotoxic products, pathogens and cells) [3, 4]. However, when the BBB is disrupted
in neuropathological diseases, plasma derived neurotoxic substances (enzyme, hemoglobin, free
ion, microbial pathogens, etc) starts passing through the BBB gaps and accumulates in the brain
parenchymal causing neuronal toxic eects including neurodegenerative changes or neuronal loss
[4]. Since complete healing of neurodegenerative diseases is challenging and disease progression
is typically irreversible, it is of paramount importance to develop imaging tools to identify early
BBB leakage and locate the disruption sites, which permits early interventions to reduce long-term
disease progression and to aid recovery from neuronal injury [5].
1.2 Aquaporin (AQP): the water channel
Recent studies have revealed that transport processes of water molecules cross the blood brain
barrier were mainly facilitated by aquaporins. Aquaporins are integral membrane proteins that
provides a more ecient pathway for transfer of water, as compared to water diusion through
the lipid bilayer of plasma membrane. More than 10 isoforms of aquaporins have been identied
4
in mammalian cells, and each of them is dierentially expressed in specic cells or organs in the
body [6]. There are 2 types of aquaporins found in the CNS: AQP1 and AQP4 [7].
1.2.1 AQP1
AQP1 was found in the epithelium of the choroid plexus and also in the pia and ependyma. Lack of
AQP1 expression may cause an important phenotypic dierence among brain endothelial cells [8].
The expression of AQP1 was controlled by two-way regulations. According to [9], AQP1 expression
was up-regulated in astrocytomas and metastatic carcinomas. BBB function was impaired in such
brain tumours leading to formation of edema. Astrocytic factors, which play a role in reduction of
endothelial AQP1 expression [10], are also important in maintaining the tightness of BBB. This
explains the up-regulation of endothelial AQP1 in tumours when astroglial cells are involved.
Another study also demonstrated the down-regulation of the tight-junction proteins claudin and
occludin in microvessels in glioblastoma multiforme [11]. Thus, loss of BBB function and the
expression of AQP1 may both be regarded as down-regulation of BBB phenotype [7].
1.2.2 AQP4: Maintaining Brain Homeostasis
Given the fact that both blood-brain interface and water molecules are polarized, eciency of
water diusion across the plasma membrane is low and AQP4 was identied as the major water
channel modulating the trans-membrane water
ux [12]. AQP4 is localized in astrocytes, which
cover the BBB, and braincerebrospinal
uid interface. Normally, density of AQP4 was 10-fold
higher in the astrocytic end-foot membranes as compared to non-endfoot membranes [13], while
redistribution of AQP4 in astrocytes were observed in diseases [13, 14]. Figure 1.2 shows the
immunolabeling of AQP4 in cortex. Entire nextwork of small vessels, including capillaries, is
covered by astrocytic processes and the AQP4 labeling reveals continuous coverage by astrocytic
end feet [12].
5
Figure 1.2: Double immunolabeling of AQP4 (red) and glial brillary acidic protein (GFAP)
(green) in cortex. AQP4 immunolabeling reveals that the entire network of vessels, including
capillaries, is covered by astrocytic processes, albeit GFAP negative. Smaller vessels and capillaries
are mostly GFAP negative but display intense labeling against the astrocyte-specic channel
AQP4. The AQP4 labeling reveals continuous coverage by astrocytic end feet. (From Simard M.
et al. J Neurosci, 2003, 23: 9254-9262.)
6
Figure 1.3: Schematic of blood water transfer from intra-vascular space to extra-vascular tissue
space. A. Labelled blood water molecules exchange into the extra-vascular space via all water
transport mechanisms including aquaporin-4 (AQP4) water channels in wild-type (WT) mice. B.
In AQP4-decient mice, the water transport mechanisms are restricted and only occur through
cotransport with organic molecules and by diusion through the lipid bilayer of the plasma mem-
brane. (Figure adapted from Ohene Y, et al. NeuroImage, 2019, 188: 515-523.)
A
uorescence study has shown that a 7-fold reduction of water permeability in cells without
AQP4 expression, as compared to the normal cells [15]. In-vivo studies using transgenic mice
(AQP4-decient) have shown the functional importance of AQP4-facilitated water transport in
cerebral water imbalance or edema [16]. AQP4 deletion has been shown to be protective, reducing
edema burden, in cellular edema, while AQP4 deletion in extracellular edema results in reduced
edema clearance rate. BBB disruption was reported to be coincided with strong induction of AQP4
mRNA and proteins in perivascular glial cells in a animal stroke model, which provides ecient
clearance of vasogenic edema [17]. During pathological conditions, AQP4 is a main entrance route
maintaining the brain homeostasis. Disruption of the polarized expression of AQP4 also aects
the water exchange in CNS diseases, such as brain ischemia and water intoxication [18, 19, 20].
A recent non-invasive MRI study using multiple echo arterial spin labeling (ASL) technique
has shown 31% increase in water exchange time in AQP-4 decient mice, as compared to the
wild-type control group, while no signicant dierence was found in cerebral blood
ow, arterial
transit time and apparent diusion coecient (ADC) was found between two groups [21]. Figure
1.3 illustrates water exchange across the BBB in normal and AQP4-decient mice. While AQP4
7
accelerates the water exchange into the extravascular space in normal mice, water transport
mechanisms are restricted to diusion or co-transport with organic molecules in AQP4-decient
mice leading to a longer water exchange time.
1.2.3 AQP4: Glymphatic Function and Clearance of Deleterious
Proteins
AQP4 is important to maintain glymphatic functions and assists clearance of deleterious proteins.
An animal study using two-photon imaging of small
uorescent tracers reported that lacking
AQP4 in astrocytes lead to about 70% reduction in glymphatic function, and about 55% decrease
in Amyloid- clearance rate from the brain parenchyma along the paravascular pathways [22].
Another study investigated the relationship between age-related changes in the expression of the
astrocytic AQP4 and glymphatic pathway function impairment, and they found aging could induce
widespread loss of AQP4 and about 40% decrease in clearance of Amyloid- in mice [23].
Due to the high density of AQP4 in astrocytic end-foot membranes, AQP4 was considered
as an excellent marker of astrocytic polarization [13]. Loss of astrocytic polarization has been
associated with BBB disruption such as in brain tumor [14]. Moreover, Yang and colleagues
reported that the loss of astrocyte polarization is also associated with development of amyloid
deposits, and redistribution of AQP4 was observed from end-foot membranes to non-endfoot
membranes. Restriction of AQP4 to the end-feet membrane of astrocyte might be essentially
important to maintain the BBB integrity and perturbation of water homeostasis could contribute
to cognitive decline or even AD [13]. Figure 1.4 shows the electron micrographs explaining how
Amyloid- deposition is associated with loss of astrocytic polarity and increased density of AQP4
in non-end-foot membranes.
8
Figure 1.4: Electron micrographs show how Amyloid- deposition is associated with loss of as-
trocytic polarity at 8 months. A. Normal linear distribution of AQP4 in wild type, gold particles
indicated by arrows. B. AQP4 (gold particles indicated by arrows) lose their polarity as Amyloid-
(asterisk) accumulates around the vessel (CAA). C. Increase in AQP4-immunoreactivity in the
non-endfeet membranes, indicated by arrows. (Compare with wild type littermate in A) Abbre-
viations: Ast, astrocyte; E, endothelial cell; GFAP, glial brillary acidic protein; L, lumen. Scale
bars, 0.5m. (Figure from Yang J, et al. Journal of Alzheimer's Disease, 2011, 27(4): 711-722.)
9
1.3 Cerebral small vessel disease (SVD)
Cerebral SVD is a heterogeneous group of neurological conditions, including arteriolosclerosis (or
hypertensive SVD), which is strongly associated with aging, diabetes and hypertension, cerebral
amyloid angiopathy (CAA), venous collagenosis, and CADASIL (cerebral autosomal dominant
arteriopathy with subcortical ischemic strokes and leukoencephalopathy) [24]. Alzheimers dis-
ease (AD) and cerebrovascular diseases share common risk factors such as hypertension, obesity,
and diabetes; these conditions coexist in 40-50% of clinically diagnosed AD, making mixed AD-
vascular dementia the most common cause of cognitive impairment in the aged [25]. The clinical
dierentiation of AD from vascular cognitive impairment and dementia (VCID) is not well dier-
entiated [26]. Cerebral small vessel disease (SVD) is the most common vascular cause of dementia,
a major contributor to mixed dementia, and the cause of about one fth of all strokes worldwide
[24, 27]. 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 [28].
However, the underlying mechanisms of small vessel disease are not well understood, resulting
in no specic guidelines for its prevention and treatment. Figure 1.5 shows several mechanisms
leading to vascular cognitive impairment, including vascular causes and brain parenchymal lesions
[29]. Among these mechanisms, multiple infarcts or localized infarcts in strategic brain regions
have been considered as a cause of dementia in the elderly [30], while cerebral SVD also has been
considered as a common cause to vascular cognitive impairment and typically manifests with
white matter lesions (WML). However, the mechanisms that link cerebal SVD with parenchymal
damage and neurological decits are not completely understood. One hypothetical mechanism is
that disrupted blood-brain barrier (BBB) causes tissue injury and demyelination of white matter
bers [31]. More details will be discussed in Chapter 2.
Cerebral small vessels, including arterioles, capillaries, and venules, are beyond the spatial
resolution limit of most of current MRI vessel imaging sequences. Lack of proper imaging tool,
10
Figure 1.5: Major mechanisms underlying vascular cognitive impairment (VCI). A. Vascular
causes. B. Brain parenchymal lesions associated with VCI. (Figure from Dichgans M, et al.
Circulation research, 2017, 120(3): 573-591.)
there is a large knowledge gap between underlying functional and anatomical changes of small
vessels and the development of cerebral SVD. To date, clinical diagnosis of SVD relies on con-
ventional MRI ndings including lacunar infarcts (DWI), WMH (FLAIR), cerebral micro-bleeds
(SWI), prominent perivascular spaces (both T1/FLAIR and T2) and atrophy [32, 33]. Figure 1.6
shows example MRI images used for clinical diagnosis of SVD.
Large epidemiological studies have shown that silent cerebral infarction and WMLs are asso-
ciated with both non-memory-related cognitive decits (e.g. executive function and perceptual
speed) [31, 34], and memory impairment [35, 36, 37]. Recent evidence suggests that WML is a
core feature of AD [38], and the progression of WML is a better predictor of cognitive impairment
than baseline WML burden [39].
11
Figure 1.6: Imaging of Small Vessel Disease: example ndings (upper), schematic representation
(middle) and a summary of imaging characteristics (lower) of MRI features for changes related to
small vessel disease. (Figure from Shi Y, et al. Stroke and vascular neurology, 2016, 1(3): 83-92.)
12
Chapter 2
BBB permeability
The physiological background of the BBB and aquaporins was introduced in Chapter 1, this chap-
ter continued the discussion about BBB permeability changes and MRI techniques assessing the
BBB permeability. Potential mechanisms leading to BBB disruption and resultant parenchymal
damage were discussed. BBB leakage has been considered as the consequence rather than a surro-
gate biomarker of the underlying microvascular changes in neuropathological diseases. Although
the contribution of BBB extracellular matrix, pericytes and astrocytes, remains largely unknown,
recently the direct role of pericyte in maintaining BBB integrity was studied using unique animal
mutants with pericyte deciency. Water transport across the BBB is aected by both AQP4
expression and BBB opening, thus BBB permeability to water could be sensitive to subtle BBB
structure and functional changes. A bi-directional change of BBB permeability was hypothesized:
water permeability increases at the early stage of BBB opening due to loss of pericytes, then
decreased water permeability should be expected after chronic BBB leakage and resultant decou-
pling of astrocytes and loss of AQP4. The last portion of this chapter summarized current MRI
techniques for BBB permeability assessment. Limitations of the DCE-MRI, a common technique
to study BBB permeability, were discussed and emerging techniques assessing the water exchange
across the BBB were brie
y reviewed.
13
2.1 BBB disruption and leakage of plasma constituents
BBB is a specialized physical and functional barrier composed of tight junctions of cerebral en-
dothelial cells. Trans-endothelial permeability is exceedingly low in healthy brain tissue with a
fully functional BBB, and there is minimal passive extravasation of plasma proteins, inorganic
solutes, or even water molecules [40]. The BBB becomes increasingly permeable with normal
advancing age, particularly in patients with vascular dementia and SVD [41, 42]. Although the
pathogenesis of lacunar infarcts, WMH, progressive cognitive impairment, and dementia have
been vigorously debated, opening of the BBB, which causes leakage of serum components and
neurotoxic into the parenchymal and leads to neuronal and glial damage, could be an important
common mechanism in these cerebrovascular diseases [40].
2.1.1 Mechanisms contributing to the BBB leakage
Hypertension and diabetes are two major risk factors for white matter hyperintensities and lacunar
stroke, and may facilitate some endothelial damage processes [40]. An animal study reported
experimental induced hypertension can cause BBB disruption and cognitive decline [43]. Diabetes
mellitus has been associated with glycosylation of endothelial proteins and leads to abnormal
thicker and weaker basement membrane of the small vessels. As a result, the microvessels in
the brain and body of diabetic subjects are susceptible to microbleeds, protein leakage, and
hypoperfusion [44]. Thickening of the small-vessel wall (or narrowing of vascular lumen) is also
a major feature of intracranial atherosclerosis, which is the main pathological lesion of lacunar
infarction.
Figure 2.1 demonstrates potential mechanisms could cause BBB to become more permeable
and damage to surrounding neurons and glial cells [40]. Initially, intact BBB separates intravas-
cular plasma constituents from the brain tissue. Endothelial damage or thickening of the vessel
wall may be initiated by toxic substances in plasma or extremely high blood pressure. The second
14
Figure 2.1: Illustration of how BBB becomes more permeable, causing damage to surrounding
neurons and glial cells. (Figure from Wardlaw J M et al, Stroke, 2003, 34(3): 806-812.)
phase of BBB leakage begins with loss of BBB integrity. Vessel wall starts swelling which leads
to luminal narrowing and shrinked vascular space. Extraversion of plasma substances into the
interstitial space causes neuronal and glial damage. When the BBB becomes totally disrupted,
the brain loses its protection from all blood substances. In the mean time, gross vessel wall
swelling and narrowing of lumen lead to reduction in CBF and ischemia. Since physical barri-
ers no longer exist and blood can freely exchange between intra-/extravascula spaces, 'apparent
vascular volume' might increase despite the luminal narrowing eect.
2.1.2 Leakage of plasma through the BBB causes neuronal damage
Plasma components could leak into brain parenchymal through disrupted BBB. Plasma-carrying
blood constituents such as plasmin (an ative enzyme in plasma), toxins and infectious agents or
15
altered electrolyte balance in interstitial spaces could cause perivascular damage. Animal stud-
ies have indicated that leakage of plasma (i.e. experimental injection of plasmin) into the brain
parenchymal induces the development of brinoid necrosis, perivascular lesions and increased en-
dothelial permeability [45, 46]. These studies suggested that edema
uid could be neurotoxic and
BBB leakage could account for the perivascular edema and neuronal damage. Epidemiological
studies also demonstrated the potential links between BBB leakage and dementia. A study re-
ported high prevalence of neurovascular instability (i.e. orthostatic hypotension and vasovagal
syncope) in patients with neurodegenerative dementia and AD [47]. Drugs (i.e. Statins), which
aects the endothelium of the BBB and ameliorate intracranial atherosclerosis, have shown eects
in reducing the risk of dementia.
During the past decade, the possibility of blood-brain barrier abnormalities occurring with ad-
vancing age, diabetes, and dementia has been considered [40]. Leakage of albumin into the CSF
(increased CSF/albumin ratio) was found in patients with vascular dementia or AD [48, 49, 50].
Plasma proteins have been found in the tissue around perforating arteries in patients with diabet-
ics, WMH, sub-cortical dementia and AD, indicating the leakage of proteins from small perforating
arteries [51, 52, 53]. However, parenchymal lesions or leakage of plasma proteins are both the con-
sequences of SVD rather than the surrogate biomarkers of the underlying microvascular changes,
and cannot guide early interventions to change the course of VCID. It is of paramount importance
to identify and develop imaging markers of early microvascular changes related to SVD for the de-
sign of future clinical trials to prevent and treat VCID. Together with advances in neuropathology,
epidemiology and genetics, developing novel neuroimaging approaches, which specically targets
at small vessels and BBB, could lead to a deeper understanding of how BBB permeability change
is related with cognitive function.
16
Figure 2.2: Changes at the neurovascular unit caused by pericyte deciency. In the absence
of pericytes (P, orange) the endothelium (E, pink) shows convoluted junctions (J) and increased
formation of pinocytotic vesicles (arrowheads). Pericyte-deciency aects polarization of astrocyte
end-feet (AF, blue), shown by mislocalization of aquaporin 4 (AQP4, red) and defective deposition
of astrocyte-derived basement membrane (aBM, dark brown). Deposition of endothelium-derived
basement membrane (eBM, beige) is not aected by pericyte-deciency. (Figure from Armulik A,
et al. Nature, 2010, 468(7323): 557.)
2.2 The role of pericytes in BBB permeability
CNS endothelial cell combined with tight junctions forms the main physical barrier, while the con-
tribution of BBB extracellular matrix, pericytes and astrocytes, remains largely unknown. Recent
studies uses pericyte-decient mouse mutants to demonstrate the direct role of the pericytes at
the BBB [54, 55]. Figure 2.2 is a schematically illustration of pericyte-loss induced changes at the
neurovascular unit. Increased formation of pinocytotic vesicles and disrupted astrocyte-derived
basement membrane are observed in the absence of pericytes, and these observations indicate that
increased macromolecular permeability across the BBB might occur through a endothelial tran-
scytosis process. Pericyte deciency also leads to redistribution of AQP4 and changes in astrocyte
end-foot polarization, which has been associated with BBB breakdown in brain tumor [14] and
development of amyloid deposits in AD [13] (as discussed in section 1.2).
Platelet-derived growth factor (PDGF)-B is a necessary factor for pericyte recruitment during
angiogenesis. Three types of mouse mutants were studied in [54] including PDGF-B retention
motif knockouts Pdgfb
ret=ret
and mutants in which Pdgfb alleles were complemented by a silent
17
PDGF-B transgene targeted to Rosa26 locus (R26P
+=0
and R26P
+=+
). Figure 2.3 (e) shows
cerebral neocortex vasculature reconstructed from 3D confocal image. Pericyte coverage and
basement membrane were depicted by CD13 and collagen IV staining. Reduced pericyte density
was associated with reduced capillary coverage (gure 2.3 (a,b)) and increased vessel diameter
((gure 2.3 (c,d)).
Pericyte deciency also increases the BBB permeability to water [54]. Figure 2.3 (f) shows
increased water content (wet/dry brain weight ratio) in brain and gure 2.3 (g,f) show the quan-
tication of Evans blue in the cerebrum, both observations indicate impairment of the BBB in
pericyte-decient mouse.
Pericytes maintain the physiological environment and neuronal connectivity in the WM. Re-
cently, Montagne and colleagues found pericyte degeneration causes white matter dysfunction in
CNS using pericyte-decient mice and DCE-MRI [55]. Figure 2.4 demonstrates the pathological
progress in the WM with degenerated pericytes. Pericyte loss causes BBB (blood-axon barrier)
disruption and leads to the accumulation of brinogen deposits, which initiates cell death in oligo-
dendrocyte and pericyte, and blood-derived ctotoxic proteins in the perivascular spaces, which
results in enlargement of perivascular spaces and triggers a loss of myelin, axons and oligoden-
drocytes. Disruption of neural circuits causes WM-related functional decits and eventually WM
diseases and neuronal loss.
In conclusion, pericytes play an important role in maintaining WM structure and function.
Since pericyte-loss causes redistribution of AQP4 in astrocyte end-foot and increased BBB perme-
ability to water, investigation of the relationship between altered BBB function, especially water
permeability, and resultant WM lesions has implications for the pathogenesis of vascular cognitive
impairment associated with WMH and SVD.
18
Figure 2.3: Pericyte coverage correlates with BBB integrity. a, b, Quantication of pericyte
coverage of capillaries in the cerebral neocortex of adult Pdgfb
ret=ret
mice (a), and R26P
+=0
and
R26P
+=+
mice (b). c, d, Capillary diameter in the cerebral neocortex of adult Pdgfb
ret=ret
mice
(c), and R26P
+=0
and R26P
+=+
mice (d), with or without pericyte coverage. e, Three-dimensional
reconstructions of confocal image z-stacks of adult cerebral neocortex vasculature depicted by
collagen IV (basement membrane) and CD13 (pericyte) staining in Pdgfb
ret=ret
, R26P
+=0
and
R26P
+=+
mice. f, Wet/dry weight ratios of control and Pdgfbret/ret mice. g, h, Quantication of
Evans blue in Pdgfb
ret=ret
mice (g) and R26P
+=0
and R26P
+=+
mice (h) in the cerebrum after
16 h of circulation. i, Time course of Evans blue accumulation in the cerebrum of Pdgfb
ret=ret
animals. y-axis shows optical density (OD) at 620 nm per gram of tissue. *P<0.03; **P<0.007;
***P<0.0005. All error bars show means.e.m. (Figure from Armulik A, et al. Nature, 2010,
468(7323): 557.)
19
Figure 2.4: A
ow diagram demonstrating the Pathological progress in the WM induced by
pericyte loss.
2.3 Bidirectional water permeability change after BBB
disruption
BBB permeability to water is mainly controlled by AQP4 and thus studying AQP4 expression
may have implications for underlying altered water permeability in response to neuropathological
conditions. Although most of DCE-MRI studies reported increased BBB permeability to contrast
agents when BBB is disrupted (i.e. brain tumor [56] and acute ischemic stroke [57]), water
exchange rate across the BBB could be altered or bi-directional depending on AQP4 expression.
BBB disruption is likely to induce vasogenic edema and extracellular water accumulation, and
clearance of BBB-mediated edema has been associated with increased water permeability and
AQP4 expression after traumatic brain injuries [58]. Expression of AQP4 was studied at the
mRNA level after intraparenchymal injection of lipopolysaccharide (LPS), which is an ecient
approach to induce BBB disruption [17]. Figure 2.5 shows the expression of AQP4 mRNA in
perivascular space in normal brain and after LPS injection. LPS-induced BBB disruption co-
incided with high induction of AQP4 mRNA in parenchymal at the beginning (gure 2.5 (b)).
However, progressively decreasing of AQP4 mRNA expression was observed afterwards and min-
imal level of AQP4 mRNA was found after two weeks of BBB disruption (gure 2.5 (c-e)).
The underlying mechanism about the bi-directional AQP4 expression or water permeability
changes remains largely unknown. The initial phase of AQP4 increased was believed to clear edema
20
and restore the brain homeostasis. While decreased water permeability after chronic BBB leakage
could be caused by accumulation of toxic substances in the perivascular spaces and resultant
disruption of astrocyte-vascular coupling [59].
2.4 MRI techniques assessing BBB permeability
Perfusion and permeability cross the BBB can be measured by tracing the delivery of the contrast
agents across the BBB in MRI studies. Contrast agents in MRI studies can shorten the eective
transverse (T2*) or longitudinal (T1) relaxation time. The amount of shortening is approximately
proportional to the concentration injected. With dynamic measurement of T1 or T2* changes,
concentration-time curves can be obtained given the relaxivity of the contrast agent.
2.4.1 Dynamic susceptibility contrast MRI
Injected contrast agent creates susceptibility gradients in the vascular space, which causes de-
phasing of the spins and signal loss in T2*-weighted MR images. Dynamic Susceptibility Con-
trast MRI (DSC-MRI) measures the T2* changes during the passage of the contrast agents bolus,
which are typically more sensitive than T1-relaxivity when the rst-pass contrast concentration
is low due to the small blood volumes [60]. However, when BBB is disrupted, relaxivity of T2*
drops rapidly when the contrast agent leaks into the extravascular space and it becomes dicult
to interpolate the relation between contrast agent concentration change and the image signal in-
tensities. Although methods have been proposed to compensate the contrast extravasation eects
in DSC-MRI [61], DSC-MRI has been mostly used for perfusion measurements (cerebral blood

ow (CBF) and cerebral blood volume (CBV)) while has limited applications in BBB permeabil-
ity studies [62]. The concentration of the contrast agents C(t) can be calculated according to
equation 2.1:
21
Figure 2.5: Expression of AQP4 mRNA in mesencephalic perivascular in ovariectomized animals.
Bright-eld high-magnication photographs showing AQP4 mRNA expression in perivascular glial
processes in control (A), and LPS-injected animals at 6 hr (B), 24 hr (C), 48 hr (D), and 14
days (E) after endotoxin injection. Scale bar = 50 um. The highest perivascular AQP4 mRNA
expression was found at 6 hr and was progressively decreasing afterwards. (Figure adapted from
Toms-Camardiel M, et al. Journal of neuroscience research, 2005, 80(2): 235-246.)
22
Figure 2.6: Dynamic MRI parametric maps of a patient with medulloblastoma. Circles indicate
region of interest (ROI). The top row (a-e) is preoperative images. The conventional post GBCA
T1-weighted MRI (a) shows a homogenously enhancing tumor in the posterior fossa (arrow). DSC
MRI parametric maps (b-c) demonstrate elevated vascularity in the tumor. DCE MRI parametric
maps (d-e) demonstrate elevated Ktrans and heterogenous Ve. The bottom row (f-j) was obtained
postoperatively and demonstrates complete tumor resection and substantially reduced vascularity
and permeability. (Figure is adapted from Thompson E M, et al. Journal of neuro-oncology, 2012,
109(1): 105-114.)
C(t) =
1
kTE
log(
S
0
S(t)
) (2.1)
where S
0
and S(t) are the baseline signal and signal after contrast agents injection, k is the
relaxivity constant, TE is echo time. CBF is dened as (equation 2.2):
C(t) =CBF (AIF (t)
R(t)) (2.2)
whereAIF (t) is arterial input function measured in major arteries,R(t) is the residue function
describing the fraction of the contrast agent remaining in the target voxel at time t andR(0) = 1
by denition. CBF can be calculated by de-convolution methods and CBV is the time integral of
C(t).
23
2.4.2 Dynamic contrast enhanced MRI
As compared to DSC-MRI, DCE-MRI, which uses gadolinium-based contrast agents (GBCAs), has
been considered as the most suitable MRI technique for permeability measurement in conventional
views. GBCAs, which shorten the longitudinal relaxation (T1) time, have been used for more than
two decades and in a broad spectrum of the CNS disorders accompanied with opening of BBB,
including multiple sclerosis (MS), ischemic strokes and brain tumors [63]. While these disorders
usually occur with disrupted BBB, more and more DCE-MRI researches are being conducted to
evaluate the subtle BBB permeability change in pathologies such as cerebral SVD, diabetes and
AD [64]. In T1-weighted DCE-MRI, the delivery of contrast agents induces brightened signal as
compared to the regions without contrast agents. Similar to equation 2.1, concentration of the
GBCAs can be calculated based on the T1 changes after injection, and the tissue tracer uptake
rate in can be expressed as (equation 2.3):
dC(t)
dt
=K
trans
(C
p

C(t)
V
e
) (2.3)
where C(t) and C
p
are contrast agent concentration in tissue and in arterial blood. V
e
is
the extravascular space volume. Tracker-kinetic (TK) parameters including Ktrans, V
e
and, frac-
tional plasma volumeV
p
can be calculated by solving the dierential equation 2.3 using standard
Tofts or modied Tofts model [65, 66]. The Patlak model, which performs better in low perme-
ability regions as compared to Tofts model [67], ignores back-
ux from the extravascular space
into the plasma compartment and estimates two kinetic parameters Ktrans and V
p
. Ktrans is a
combined parameter to represent several physiological eects. For example, Ktrans re
ects the
permeability-surface area product (PS) when BBB permeability is low. Ktrans could also rep-
resent the blood
ow when intra-/extravascular spaces are no longer distinguishable with totally
disrupted BBB. Ktrans can be used for direct evaluation of the tumor severity, which enables
DCE-MRI to characterize tumor biology and study the treatment response [56, 62]. However,
24
DCE-MRI usually takes longer scanning time to ensure sucient spatial and temporal resolution,
and it has been challenging to interpolate the quantitative DCE-MRI results since TK mod-
elling strategies and reconstruction techniques are yet to be improved especially in low permeable
regions, which makes it dicult for the clinical translation of DCE-MRI [68].
Figure 2.6 shows the preoperative and postoperative dynamic MRI parametric maps of a pa-
tient with medulloblastoma. DSC MRI parametric images (gure 2.6 (b-c)) demonstrate increased
rCBV and rCBF in the tumor. DCE MRI parametric images (gure 2.6 (d-e)) demonstrate in-
creased Ktrans and Ve. Figure 2.6 (f-j) demonstrates substantially reduced vascularity and
permeability with complete tumor resection.
2.4.3 Water: an endogenous tracer
Although no radiation is involved in DCE-MRI, GBCAs can have complications in persons with
compromised kidney function and have been linked to permanent Gd deposition in the brain,
especially in persons undergoing repeated scans. Both the US Food and Drug Administration
and International Society for Magnetic Resonance in Medicine have recently issued statements to
limit the use of GBCAs to clinical circumstances in which the additional information provided
by the contrast is necessary [69]. An alternative to exogenous contrast agents is water, which is
an abundant and endogenous tracer with limited permeability across the BBB. Because GBCAs
have relatively large molecular weights (Gd-DTPA 550 Da), BBB permeability has to reach a
critical level before extravasation occurs [42, 70] and a previous study reported no evidence of
BBB leakage in patients with dementia using GBCAs [71]. Water molecules have much smaller
molecular weight; assessing BBB water permeability could potentially provide a more direct and
sensitive biomarker of BBB function at the early stage of disease progression. A recent study
reported about 2-fold increase in water permeability (gure 2.7 (a)) in AD rats as compared to
wild-type rates, while trans-membrane leakage of contrast agents remains unaltered ((gure 2.7
(b))) [72]. The increased water permeability was hypothesized to be associated with reduced
25
Figure 2.7: MRI and immunostaining results in TgF344-AD and wild-type rats. (a) PSw is
signicantly higher (up to 2-fold) in TgF344-AD rats relative to wild-types (p<0.05; two-way
ANOVA). (b) Trans-BBB leakage of contrast agent (Ktrans) is unaltered between TgF344-AD
rats and wild-types (p=0.477; two-way ANOVA). (c) Occludin is reduced in TgF344-AD relative
to wild-types (p<0.05; two-way ANOVA), corresponding well with genotype dierences in PSw.
(d) No detectable TgF344AD/wild-type dierences were observed for aquaporin-4 (AQP4). (e)
Representative occludin and lectin immuno-stains. Aspecic staining of amyloid- was visually
identied on the lectin snapshots and manually segmented as shown. (Figure adapted from Dickie
B R, et al. NeuroImage, 2019, 184: 349-358.)
expression of a tight junction protein (occludin, gure 2.7 (c,e)), while no signicant dierence
was observed for AQP4 expression between two groups (gure 2.7).
Arterial spin labeling (ASL) perfusion MRI permits noninvasive measurement of cerebral blood

ow (CBF) using magnetically labeled water as an endogenous tracer. Water exchange across the
BBB can be quantied based on ASL signal fractions in the intra- and extravascular compart-
ments. Recently, global water extraction fraction (Ew) and PSw were determined by measuring
arterial labeled blood spins that are drained into cerebral veins [73] with extreme long delay time
26
(PLD > 3000 ms), which generates reliable results in several minutes but cannot reveal BBB
permeability change in local regions.
Due to T1 relation of blood (T1 of arterial blood at 3T is 1664 ms accordting to [74]), ASL
decays fast with longer PLDs. Kinetic models have been proposed to map the whole-brain trans-
capillary water exchange at shorter PLDs, before ASL signals are totally extracted into the ex-
travascular space. The rst step of applying the kinetic models is to estimate the relative fraction
of ASL signal in intravascular and extravascular spaces. Multi-echo ASL has been proposed to
measure water exchange across the BBB based on the T2 and T2* dierences in the two com-
partments [75, 76]. However, reliable and accurate quantication remains challenging because of
the small dierences of T2/T2* of intra-/extravascular blood signals.
A new method for PSw was proposed recently by utilizing the intrinsic diusion weighting
of gradient and spin echo (GRASE) readout (IDEALS) [77]. Given the dierent diusivity of
the intra-/extravascular signals, the intrinsic diusion weighting (from the crusher gradients of
the GRASE) introduce dierent blurring eects for intra-/extravascular ASL signals along the
partition direction. The fraction of intra-/extravascular signals can be determined from multiple
GRASE acquisitions with dierent segmentation patterns, and Ew and PSw were estimated as-
suming that total extraction occurred at PLD > 2000 ms and Ew equals to the fraction of the
extravascular ASL signal. Although the authors have demonstrated that reduced CBF, Ew and
PSw were observed after the caeine challenge in healthy subjects, the sensitivity and accuracy of
the IDEALS method need more evaluations in CNS diseases. Moreover, diusion weighting from
crusher gradients is inherently small (see discussion in section 3.3) and the dierence between
the intra-extravascular blurring eects could be limited. Thus, computing the fractions requires
sophisticated deconvolution algorithms which might fail in low SNR applications, e.g. in aged
subjects.
Diusion-weighted (DW) ASL with 2D EPI readout has been proposed to dierentiate the
fraction of labeled water in capillary and brain tissue based on their distinctive (pseudo-)diusion
27
coecients (high in capillary and low in tissue) [78, 79]. This DW ASL technique has recently
been validated by mannitol administration to open the BBB and using an ischemia-reperfusion
model to disrupt BBB in rats [80, 81]. The large dierence between tissue ADC and capillary
pseudo-diusion coecient has been validated in this animal study (gure 2.8 row I and II).
Increased Ktrans and Evan's Blue staining were observed in the hemisphere with BBB opening
after mannitol injection (gure 2.8 row III). However, signicant decrease in kw was observed in
the aected hemisphere (gure 2.8 row IV). Decreased BBB water permeability was also found
in subjects with obstructive sleep apnea (OSA) using 2D DW-ASL [82]. Although increased
water permeability has been associated with subtle BBB opening [54, 72], decreased kw in regions
with disrupted BBB could be potentially explained by parenchymal damage and loss of AQP4
according to our bidirectional water permeability change hypothesis (section 2.3).
As discussed in section 3.4, the strength of our technique is that there is 2 orders of magnitude
dierence between the (pseudo-)diusion coecients of the intra- and extravascular spaces, which
can be separated by a small diusion gradient without. Figure 2.9 shows average diusion weighted
ASL signals acquired at four PLDs. Intravascular signals decays rapidly with a small b-value and
diminished with longer PLDs due to increased extraction ratio of labeled water into tissue space.
Figure 2.10 shows the water exchange rate (kw (min
1
), PSw divide by vascular space volume,
see Chapter 4) maps estimated from the ratio of ASL signals with and without diusion weighting
from a previous DW-ASL study. However, the clinical utility of the 2D DW-pCASL is limited
due to the following: 1. lengthy scans are required for reliable measurement due to relatively low
signal-to-noise ratio (SNR) of 2D ASL; 2. sequential acquisition of 2D slices results in dierent
post-labeling delays and bolus kinetics complicating the signal modeling and quantication. In
this dissertation, a novel 3D diusion prepared ASL sequence (chapter 3) was developed and a
regularized reconstruction algorithm (chapter 4) was proposed to improve the reliability of the
kw measurement.
28
Figure 2.8: Row I and II: M and M (non-labeled ASL) signals from multi-b data tted well to
the bi-exponential model (n=8,R
2
= 0.997, 0.998 respectively). Using the full model, A1 and kw
were estimated to be 0.094 (95% Condence bounds: 0.053, 0.14) and 70min
1
(95% Condence
bounds: 59 min
1
, 86 min
1
) respectively. Using the simplied two-b model, A1 and kw were
estimated to be 0.140.02 and 586min
1
respectively. Row III shows the A1 and kw map before
and after mannitol, along with the Ktrans map and Evans blue brain slice. Mannitol caused BBB
disruption laregely to one hemisphere. kw in the aected hemisphere decreased to 102 min
1
after mannitol (P <0.01, n=5) (row IV). Ktrans and Evans blue staining showed similar patterns
of BBB disruption. (Figure from Tiwari YV, et al, Proceedings of ISMRM, p0792, 2015.)
29
Figure 2.9: Average M signal (symbols) acquired at four post-labeling delays plotted as a
function of diusion-weighting strength. Data were averaged across all pixels in grey matter
(a), or white matter (b), and across four subjects. Each subjects data were normalized to the
signal without diusion weighting (b=0). The errors bars represent the standard error. The solid
lines represent the best t of a bi-exponential decay model to each M series. (Figure from St.
Lawrence K S, et al. Magnetic Resonance in Medicine, 2012, 67(5): 1275-1284.)
Figure 2.10: DW-ASL data acquired with PLD = 1.5 s. First row: average ASL signal without
diusion weighting DM(b0); Second row: average ASL signal with diusion weighting DM(bDW);
Third row: the ratio images DM(bDW)/DM(b0); Last row: kw images. All images were smoothed
with a gaussian lter with a kernel size of 15 mm for the DW-ASL (kw) images. (Figure from St.
Lawrence K S, et al. Magnetic Resonance in Medicine, 2012, 67(5): 1275-1284.)
30
Chapter 3
When diusion meets perfusion
This chapter introduced the physic basis of separating and quantifying the intra-/extravasular
components using diusion and perfusion weighted MRI. Due to pseudo-randomly orientated
capillary network, capillary blood signal is aected by diusion gradients and its pseudo-diusion
coecients is much bigger than the apparent diusion coecients (ADC) of tissue blood. In other
words, capillary blood signal is more sensitive to diusion gradients and thus intra-/extravasular
components can be separated by a small diusion weighting. Intra-voxel incoherent motion (IVIM)
technique, which acquires images with increasing diusion weightings, has been used to t intra-
/extravasular contributions (or perfusion fraction). However, assessing BBB permeability could be
challenging since IVIM measures all free water in the brain, while other imaging techniques such
as ASL, which traces the transfer of labeled blood across the BBB, would be more desirable. The
emphasis of this chapter is the MRI pulse sequence design. Combining 3D GRASE and diusion
gradients is dicult due to violation of Carr-Purcell-Meiboom-Gill condition. A novel non-CPMG
3D diusion prepared pCASL sequence was proposed and validation studies were performed. An
appropriate diusion gradient was chosen which suciently eliminates the intravascular signal
while has minimal eects on tissue blood signal.
31
3.1 Microcirculation of the capillary network and intra-
voxel incoherent motion (IVIM)
Microcirculation is a term describing the blood circulation in the microvessels within tissues,
which include terminal arterioles, capillaries and venules. Oxygenated blood enters into capillaries
through arterioles and
ows out into the venules then into the veins, while most of microcirculatory
exchanges occurs in the capillaries [83].
Microcirculation of blood in the capillary network has been considered as a specic type of
diusion (macrodiusion) because there are no preferred orientations of capillaries at the level of
spatial resolution of MRI images (about a few mm) [84]. This "macro-diusion" is determined by
the blood velocity and orientation distribution of the capillary network.
In diusion weighted MRI (DW-MRI), signal attenuation in response to diusion weightings
(b values) re
ects tissue diusivity, which is commonly quantied by the ADC. ADC is a measure
of microscopic random movement (Brownian motion) of water molecules within tissue, and is
commonly tted using a mono-exponential decay model (equation 3.1):
S(b)
S(0)
=e
bD
(3.1)
where S(b) and S(0) are the signal intensities with and without diusion weighting. D is the
true molecular diusion coecient.
However, considering the random and isotropic nature of the microvascular network system
and assuming that the capillary vessels are randomly distributed within an isotropic voxel, blood
signal within capillary network would also be attenuated with increasing b-values. Figure 3.1
illustrates the collective water
ow in the capillary segments resulting in pseudo-diusion eect
for capillary blood
ow. As a result, ADC is aected by both tissue diusivity (true diusion)
32
Figure 3.1: Molecular diusion and blood pseudo-diusion. Molecular diusion (left) is a random
process at individual molecular level resulting in a Gaussian distribution of molecular displacement
(free diusion). Pseudo-diusion for blood
ow results from the collective water
ow in the
randomly orientated capillary segments. l is average displacements, v is velocity and D/D* are
diusion coecient and pseudo-diusion coecient. (Figure from Le Bihan D. NeuroImage, 2017
(in press).)
33
and micro-capillary perfusion according to the IVIM model. Equation 3.1 can be adjusted for two
compartments in IVIM model (equation 3.2) [84]:
S(b)
S(0)
= (1f)e
bD
+fe
bD

(3.2)
where S(b) and S(0) are the signal intensities with and without diusion weighting. D is
the true molecular diusion coecient. D

is the pseudo-diusion coecient which indicates the
pseudo-randomly orientated motion of blood within the capillary network. f (perfusion fraction)
is the fraction of the signal from the capillary within one voxel.
D

, which is considered as a "fast decay compartment", is typically much greater than D
[85]. While Brownian motion of water molecules contributes to the most of signal loss when
a relatively strong diusion gradient is applied, the second IVIM mechanism (microcirculation
of blood in the capillary network) contributes more to the signal attenuation when diusion
gradient is weak. IVIM has been used to measure perfusion fraction f
IVIM
, which could be
contributed from dierent vessel segmets (capillary, pre-capillary vessels, etc) [86]. Assessing
BBB permeability using IVIM could be challenging since IVIM re
ects all randomly moving
blood while other imaging techniques such as ASL that traces the blood originated from outside
of the voxels and monitors the exchange of blood across the BBB would be more desirable.
3.2 Aterial spin labling (ASL) and perfusion signals
ASL is a non-invasive technique to measure perfusion using magnetically labeled arterial water as
an endogenous tracer. Compared to other perfusion imaging techniques, such as PET, DSC, and
DCE MRI, ASL oers the similar capability of measuring CBF while introducing no radiation or
contrast agent. Pseudo-continuous ASL (pCASL), which employs a train of discrete RF pulses to
mimic
ow-driven adiabatic inversion, is capable of selectively labeling carotid artery and tracing
the passage of labeled blood in microcirculation. Figure 3.2 illustrates the experiment setup of
34
Figure 3.2: (a) Experiment setup of pCASL scan. Labeling plane (yellow line) is typically placed
perpendicular to the carotid below the cerebellum. Incoming blood is tagged by
ow driven
adiabatic inversion and imaged after entering capillary/tissue space at PLD. Blue box indicates
the imaging volume. (b) Perfusion signal is obtained by subtraction of control (without ASL
tagging) and label (with ASL tagging) images.
pCASL scans. ASL signal is obtained by subtraction of label and control pairs and usually less
than 1% of the background tissue signal and thus extremely sensitive to noise induced by motion
and system instabilities. With background suppression and 3D acquisition, signal
uctuations
has been eectively suppressed and temporal signal-to-noise ratio (t-SNR) increased signicantly
[87]. During recent years, ASL has seen a rapid growth in applications in both clinical and basic
neuroscience studies [88, 89].
In presence of perfusion, Block equation for the longitudinal magnetization can be modied
into equation 3.3 [90]:
dM
dt
=
M
0
M
T
1
+fM
b

f

M (3.3)
35
where f is the blood
ow, is the brain-blood partition coecient. M and M
b
are the
longitudinal magnetization of the tissue water and arterial blood. M
0
is the fully recovered
proton density weighted signal. The single compartment Kety model, which assumes water to
be a freely diusible tracer between capillaries and the surrounding tissue, is commonly used for
quantication of cerebral blood
ow (CBF) given the similar relaxation rates of blood and tissue
[91]:
CBF =
Me
PLD
T
1;blood
2M
0
T
1;blood
(1e

T
1;blood
)
(3.4)
where M is the average ASL signal, PLD is post-labeling delay, is the labeling eciency
and T
1;blood
is the longitudinal relaxation time of blood.
In reality, water diusion across the BBB is limited and water exchange is mainly facilitated
by aquaporins. Thus, equation 3.4 usually leads to a progressive underestimation of CBF, as
conrmed by
15
O-water PET studies [92].
St Lawrence, et al, proposed a one-barrier distributed parameter model to describe the ex-
change of water between the capillary and tissue spaces [91]. As illustrated by gure 3.3, the
capillary and tissue spaces are represented by two concentric cylinders, while the BBB is repre-
sented by an innitely thin barrier in the middle. The ASL signal in these two compartments
(C
b
(x;t) andC
c
(x;t)) were connected by the parameter capillary permeability-surface area prod-
uct (PS). St Lawrence further simplied the model by assuming that labeled blood that exchanges
into the brain tissue compartment relaxes before exchanging back into the capillary. And this
single-pass approximation (SPA) model will be used to quantify the water exchange across BBB
(Details are discussed in section 4.1).
36
Figure 3.3: Capillary-tissue unit dened by the one-barrier distributed parameter (1BDP) model.
The inner cylinder represents the capillary space and the outer cylinder represents surrounding
brain tissue. The capillary space has cross-sectional areaA
c
, volumeV
c
, and tracer concentration
C
c
(x;t), where x is the spatial dimension along the length of the cylinder. The surrounding brain
tissue has cross-sectional area A
b
, volume V
b
, and tracer concentration C
b
(x;t). A permeable
membrane that represents the blood-brain barrier separates the two spaces: exchange of tracer
across the blood-brain barrier is characterized by the PS product. Labeled water
ows into the
capillary-tissue unit via arterial blood at a concentration C
a
(t), and exits via venous blood at
a concentration C
v
(t). The rate of blood
ow into the capillary space is F. (Figure from St.
Lawrence K S, et al. MRM, 2000, 44(3): 440-449.)
3.3 MRI pulse sequence design
Figure 3.4 (a) shows the sequence scheme, which consists of pCASL labeling, background sup-
pression, diusion preparation and GRASE readout. Diusion preparation was implemented by
non-selective pulses and bi-polar gradients along the z-direction, as shown in Figure 3.4 (b), and
eddy current was minimized by optimizing the timing of gradients [93] (Details about gradients
design are discussed in Appendix A). Transverse signal was tipped-up before readout and a spoiler
was added to destroy residual phase. Bulk motion during the diusion encoding induces spatially
varying phase shift, which nulls GRASE signal due to violation of CPMG condition. To elimi-
nate the phase sensitivity, spins were spread out within each voxel by dephasing gradients before
the tip-up and spoiler, and re-phasing gradients were added before acquisition and then rewound
after acquisition, as shown in Figure 3.4 (b, c). This non-CMPG approach requires an ideal slice
prole of the 90
0
pulse, and we used a non-selective GRASE excitation while selective refocusing
pulses determined the slab coverage. Imaging parameters were: FOV = 224 mm, 12 slices (10%
37
Figure 3.4: (a) Sequence scheme of 3D DW-pCASL. (b) Diusion preparation module: Non-
selective pulses were used to compensate eld inhomogeneity, timing of gradients was optimized
to minimize eddy current. De-phasing gradient was added along y-axis (4 dephasing per voxel)
before tip-up to eliminate phase sensitivity of GRASE readout. Strong spoiler along three axes
were added after tip-up to remove residual transverse magnetization. (c) GRASE readout: Non-
selective excitation was used to improve the slab prole, re-phasing and rewound de-phasing
gradients were added at two sides of EPI readout to maintain MG condition.
oversampling), turbo factor = 14, EPI factor = 64, resolution = 3.53.58 mm
3
, TE = 36.5 ms,
TR = 4000 ms, label/control duration = 1500 ms.
3.4 Diusion weighted perfusion signal
Four healthy volunteers (3 male; age = 34 11 years) underwent MRI scans on a Siemens 3T
Prisma system (Siemens, Erlangen, Germany) using a 20-channel head coil after they provided
informed consent according to a protocol approved by the Institutional Review Board of the Uni-
versity of Southern California (Los Angeles, CA) for pulse sequence optimization. Control/label
images were corrected for rigid head motion oine using SPM12 (Wellcome Trust Centre for Neu-
roimaging, UCL, London, UK) and subtracted to obtain perfusion images. Temporal
uctuations
38
in the dierence image series owing to residual motion and physiological noise were minimized
using an algorithm based on principal component analysis (PCA) [94].
To determine the optimal b
DW
, the proposed sequence was performed in 4 healthy subjects
with 3 PLDs (1500, 1800 and 2100 ms) and 6 b-values (b = 0, 10, 25, 50, 100, and 200 s=mm
2
).
Twenty repetitions (2 minutes 40 seconds) were acquired for each b-value. Bi-exponential tting
of ASL signals with 6 diusion weightings was conducted to calculate the diusion coecients for
capillary (Dc) and tissue (Db) compartments and determine the appropriateb
DW
that suppresses
capillary signal with minimal eect on tissue signal (Equation 3.5):
M
b
M
0
=A
1
e
bDc
+ (1A
1
)e
bD
b
(3.5)
Figure 3.5 shows DW pCASL perfusion images of a single slice acquired at 3 PLDs and 6
b-values. The DW pCASL signal intensity decays with increasing PLD or b-values.
Average perfusion signal intensity from 4 subjects (marks) and bi-exponential tting results
(curves) are shown in Figure 3.6 (R
2
= 0.997, 0.988, and 0.996 for the bi-exponential tting
of perfusion signals at PLD = 1500, 1800, and 2100 ms, respectively). On average, 76%, 85%,
and 89% of labeled blood enters brain tissue space at the PLD of 1500, 1800, and 2100 ms,
respectively. Estimated (pseudo-)diusion coecients of capillary/brain tissue (Dc/Db) were
0.08/0.0010, 0.09/0.0009, and 0.05/0.0006 mm
2
=s at the PLD of 1500, 1800, and 2100 ms, re-
spectively. Based on these results, b
DW
= 50 s=mm
2
and PLD = 1800 ms were chosen for
subsequent kw measurements, where perfusion signal in capillary and brain tissue compartments
were 1.1% and 95.6% of its original signal intensity, respectively. In other words, perfusion signal
M1800 50 contains 1.1% and 98.9% of capillary and tissue signal, respectively. The dierentia-
tion between capillary and tissue space is reliable given the large diusion coecient dierence
( 100 fold) between the two compartments. A sensitivity analysis with20% change in b
DW
(50
s=mm
2
) would induce only 1% change in remaining capillary signal according to Equation 3.5.
39
Figure 3.5: Perfusion map with 6 diusion weightings acquired at PLD = 1500, 1800, and 2100
ms, respectively. Gray scale indicates relative perfusion signal intensity compared to average
perfusion signal acquired with b = 0 s=mm
2
at PLD = 1500 ms.
40
Figure 3.6: Average perfusion signals from 4 subjects with 6 diusion weightings acquired at PLD
= 1500, 1800, and 2100 ms. Error bar indicates the standard deviation of kw measurements across
4 subjects. Biexponential tting results are shown in the upper right corner. Capillary/tissue
fraction were 24%/76% when PLD = 1500 ms, 15%/85% when PLD = 1800 ms, and 11%/89%
when PLD = 2100 ms, respectively
41
We utilized the 2-stage approach proposed by St Lawrence et al to measure ATT and kw
[79]. Fifteen repetitions were acquired for each b-value of the FEAST scan at PLD = 900 ms
with a total acquisition time of four minutes. kw was calculated from scans acquired at PLD =
1800 ms, when the labeled blood reaches the microvascular compartment, with b = 0 and b
DW
.
Twenty repetitions were acquired for each b-value of the kw scan, and total acquisition time was
six minutes.
An extra reference M
0
image without background suppression was acquired at the PLD of
2000 ms to generate CBF and R
1b
maps. CBF was calculated from the reference image M
0
and
M
1800
0
according to equation 3.4, using blood-tissue water partition coecient = 0.9 g/mL and
labeling eciency = 77%. R1b map was computed from M
0
, M
900
0
and M
1800
0
according to
[95], details are discuss in Appendix B.
42
Chapter 4
Improved SPA modeling of kw
Mathematical modeling of diusion weighted ASL signal was discussed in this chapter. A single-
pass approximation (SPA) model was proposed by St Lawrence, et al, to calculate kw from
intra-/extravasular ASL signal ratio. However, spurious high kw was observed due to low SNR.
A novel TGV regularized SPA modeling algorithm was proposed to improve the reliability of
kw and ATT quantication. Reproducibility of the kw measurement was evaluated by test and
retest scans. Intra-class correlation coecient (ICC) was calculated from nineteen subjects and
compared with previous 2D DW-pCASL measurements.
4.1 Modeling of water exchange rate kw across BBB
According to the Renkin-Crone equation [96, 97], the permeability surface product of water (PSw)
can be calculated based on the water extraction ratio (Ew) and CBF (Equation 4.1):
PS
w
=ln(1E
w
)CBF (4.1)
To estimated Ew, a long PLD is usually required to allow complete extraction of labeled
water into tissue space [73]. Because of T1 relaxation, the low SNR of remaining ASL signal
makes it impractical to generate reliable voxel-wise water exchange rate map. St Lawrence et
43
al proposed a SPA solution to model ASL signal in the capillary and brain tissue compartments
while incorporating the exchange rate of water from blood to tissue (kw) [91] (Equations 4.2 and
4.3):
M
c
=(
2"CBFM
0
(kw +R
1a
)
)e
(R1a(kw+R1a))ATT
(e
(kw+R1a)(t)
e
(kw+R1a)t
) (4.2)
M
b
= (
2"CBFM
0
(kw +R
1a
)
)
k
w
kw + (R
1a
R
1b
)
[
e
(R1aR
1b
)ATT
R
1b
(e
R
1b
(t)
e
R
1b
t
)

e
(R1a(kw+R1a))ATT
k
w
+R
1a
(e
(kw+R1a)(t)
e
(kw+R1a)t
)]
(4.3)
where M
c
(t) and M
b
(t) are ASL signals from the capillary and tissue space, respectively; "
is labeling eciency, is labeling duration, is the partition coecient of water in the brain, and
R
1a
and R
1b
are the longitudinal relaxation rate of arterial blood and brain tissue, respectively.
R
1a
was assumed to be 0.601 s
1
[74]. The voxel-wise R
1b
map was tted from background
suppressed control images acquired at 2 PLDs according to a previous work [95]. The water
exchange rate, kw, dened as capillary permeability surface-area product of water (PSw) divided
by distribution volume of water tracer in the capillary space (Vc), was calculated based on a
monotonic relationship with the fraction of capillary signal at a given arterial transit time (ATT),
as demonstrated by Figure 2 in St Lawrence et al [79] (Equation 4.4):
k
w
=f(A
1
;ATT )
A
1
=
M
c
(t)
M
c
(t) + M
b
(t)
(4.4)
where f was derived from Equations 4.2 and 4.3. Capillary signal would be suppressed by
a small diusion gradient because of its pseudo-random motion, and A
1
can be calculated by
(Equation 4.5):
A
1
= 1
M
b
DW
M
0
(4.5)
44
where M
PLD
bvalue
is ASL signal with specic PLD (ms) and b-value (s=mm
2
) indicated by
superscript and subscript, respectively. The appropriate diusion gradient with a weighting of
b
DW
, which suppresses capillary signal while imparting minimal eect on tissue signal, can be
determined by bi-exponential tting of the DW pCASL signals acquired at multiple b-values. ATT
was estimated by the
ow-encoding arterial spin tagging (FEAST) method [98], as a function of
the ratio of the vascular suppressed (with diusion weightingb
ATT
= 14s=mm
2
, velocity encoding
[VENC] = 7.5 mm/s) ASL signal to the total signal acquired at a short PLD (900 ms; Equation
4.6):
ATT =g

M
900
b
ATT
M
900
0

(4.6)
4.2 Estimation of kw with total generalized variation
regularized SPA model
According to St Lawrence et al [79], estimated kw is sensitive to noise when tissue fraction is
close to 1. A Gaussian lter was applied to ASL images to improve SNR; however, a predened
threshold of kw was still required to exclude spuriously high values in local regions. Instead of
using a Gaussian lter, we propose a novel total generalized variation (TGV) regularized SPA
modeling algorithm for estimating ATT and kw. TGV is an improved mathematical framework
based on minimizing both rst- and second-order total variation (TV) for MRI denoising or
undersampled reconstruction, which minimizes blotchy (or oil painting like) appearance in MRI
images reconstructed with traditional TV algorithm [99]. ATT and kw can be estimated from
DW pCASL data acquired at the PLD of 900 and 1800 ms with respective b-values (Equations
4.7 and 4.8):
45
arg min
ATT;ATT
0
h
1
2
kATTg

M
900
b
ATT
M
900
0

k
2
2
+
1
jOATTATT
0
j
1
+

0
2
jOATT
0
+OATT
0
T
j
1
i
(4.7)
arg min
kw;k
0
w
h
1
2
kk
w
f

1
M
1800
b
DW
M
1800
0
;ATT

k
2
2
+
1
jOk
w
k
0
w
j
1
+

0
2
jOk
0
w
+Ok
0
T
w
j
1
i
(4.8)
where = 0.05 is the weighting factor balancing data delity and TGV penalty function, O
donates discrete dierentiation,
1
= 1 and
0
= 2, which were recommended by Knoll et al [99],
balances between the rst and second derivative of ATT and kw map.
Figure 4.1 shows the comparison results from direct SPA modeling with a Gaussian lter
(rst row) and the proposed SPA modeling with TGV regularization (second row). Figure 4.1
(a) and (b) shows the perfusion maps acquired at the PLD of 900 ms without and with diusion
weighting for vascular signal suppression (b = 14 s=mm
2
), respectively. Figure 4.1 (d) and (e)
shows the perfusion maps acquired at the PLD of 1800 ms without and with diusion weighting
for suppression of the microvascular/capillary signal (b = 50 s=mm
2
). A 3D Gaussian lter with
a full-width at half maximum (FWHM) of 5 mm was applied to obtain the perfusion images in
the rst row of Figure 4.1 (a,b,d,e). Figure 4.1 (c) shows estimated ATT maps. Prolonged ATT is
observed in the posterior area, which is consistent with previous ndings [98]. Figure 4.1 (f) shows
the kw map estimated from direct SPA modeling (rst row) and the proposed TGV regularized
SPA modeling (second row). Direct SPA modeling with a Gaussian lter generates smoother kw
maps whereas TGV regularized SPA modeling preserved the original image resolution. The local
bright regions (indicated by red arrows, kw> 200min
1
) with spuriously high kw values in direct
SPA modeling were suppressed by TGV regularized SPA modeling.
46
Figure 4.1: Comparison of direct modeling with Gaussian smoothing (rst row) and regularized
SPA modeling (second row). (a) Perfusion map without diusion weighting acquired at PLD
= 900 ms. (b) Perfusion map with b = 14 s=mm
2
(VENC = 7.5 cm/s to suppress vascular
signal) acquired at PLD = 900 ms. (c) ATT map. (d) Perfusion map without diusion weighting
acquired at PLD = 1800 ms. (e) Perfusion map with b = 50s=mm
2
acquired at PLD = 1800 ms.
(f) kw map. Red arrows indicate the local regions with noise induced spuriously high kw values
using direct modeling (rst row). kw map from regularized SPA modeling was relatively smooth
(second row).
4.3 Test and retest reproducibility
Nineteen aged subjects (7 male; age = 68.8 7.6 years, all Latinos) enrolled from the MarkVCID
study (www.markvcid.org) for clinical evaluation of the developed pulse sequences and 5 subjects
from the same cohort (2 male; age = 68 6 years) for comparison with 2D DW pCASL. To
evaluate the reproducibility of the proposed sequence, subjects underwent 2 MRIs approximately
2 weeks apart on a Siemens 3T Prisma system (Siemens, Erlangen, Germany) using a 20-channel
head coil after they provided informed consent according to a protocol approved by the Institu-
tional Review Board of the University of Southern California (Los Angeles, CA). Test-retest MRI
scans were conducted on similar times of day to minimize potential eects of circadian rhythms,
and subjects were abstinent from caeine intake 3 hours before MRI scans. Imaging parameters
of the 2D DW pCASL were: FOV = 224 mm, matrix size = 64 64, 7/8 partial Fourier factor,
12 slices, ascending ordering, slice gap = 1 mm, resolution = 3:5 3:5 8 mm
3
, bandwidth =
3125 Hz/pixel, TE = 48 ms, TR = 4300 ms, label/control duration = 1500 ms. Fifteen pairs were
47
Figure 4.2: kw map of 6 slices from one representative subjects test and retest scans
acquired at PLD = 900 ms with b = 0 and 10 (VENC = 7.5 mm/s) s/ mm
2
, and 20 pairs were
acquired at PLD = 1800 seconds with b = 0 and 50 s=mm
2
, respectively.
The test-retest reproducibility of average kw and CBF in the whole brain was quantied by
intra-class correlation coecient (ICC). The kw maps were then normalized into the Canonical
Montreal Neurological Institute space, and the ICC of kw was also computed in 8 regions of
interests (ROIs) related to AD: frontal lobe, temporal lobe, parietal lobe, hippocampus, parahip-
pocampal gyrus, anterior/posterior cingulum, and precuneus [100].
Figure 4.2 shows six slices of kw maps from test-retest scans (global kw = 95.3 and 96.5min
1
)
acquired by the proposed sequence of one representative subject (female, 64 years). Average kw
values of the whole brain acquired at the second scan are plotted against the kw values acquired
at the rst scan, as shown in Figure 4.3 (a). A good test-retest reproducibility (ICC = 0.75) was
achieved for the proposed DP 3D GRASE pCASL sequence, whereas poor reproducibility was
observed for 2D DW pCASL results (ICC = 0.21; Figure 4.4).
48
Figure 4.3: (a) Average kw values from test-retest experiments using the proposed 3D DP pCASL
sequence. Horizontal and vertical axis indicates the kw measurements from the rst and second
MRI scan, respectively. (b) Average global CBF values from test-retest experiments. Horizontal
and vertical axis indicates the CBF measurements from the rst and second MRI scan, respectively
Figure 4.4: Average kw values from test-retest experiments using the 2D DW-pCASL sequence.
Horizontal and vertical axis indicates the kw measurements from the rst and second MRI scan,
respectively.
49
Average kw (min
1
) ICC
Frontal 98.3 20.8 0.72
Temporal 97.8 17.3 0.54
Parietal 100.6 22.2 0.52
Hippocampus 101.7 22.4 0.30
Para hippocampal gyrus 88.9 21.8 0.17
Anterior cingulum 106.6 21.9 0.74
Posterior cingulum 108.6 22.5 0.57
Precuneus 102.4 19.9 0.63
Table 4.1: Average kw and ICC values of test and retest measurements in 8 ROIs related to AD
Table 4.1 summaries the average kw and ICC values of test and retest measurements from 19
subjects in the 8 ROIs. ICC ranges from 0.17 in parahippocampal gyrus and 0.3 in hippocampus
to 0.63 in precuneus and 0.72 in frontal lobe, with an average of 0.52.
Estimated average kw was 105.020.6, 109.618.9, and 94.119.6 min
1
for the whole brain,
GM, and WM, respectively, which corresponds well with the literature [79]. Average ATT was
1242.1111.1, 1220.6100.2, and 1288.8113.7 ms for the whole brain, GM, and WM, respec-
tively. The measured ATT values fall into the lower end of the literature values [98], which may be
caused by the single excitation of the GRASE readout as compared to the previous 2D sequential
slice acquisitions.
Average global CBF = 45.611.6 mL/100g/min across nineteen aged subjects from both test
and retest scans. CBF values of the whole brain acquired at the second scan are plotted against
the CBF values acquired at the rst scan, as shown in gure 4.3 (b). ICC = 0.85 for CBF acquired
from test and retest scans.
A major innovation of the present study is TGV regularized SPA modeling. In the original
SPA modeling strategy [79], the estimated kw is very sensitive to noise when the tissue fraction is
close to 1 (see Figure 2 of St Lawrence et al, [79]). This challenge is accentuated by the relatively
low SNR of ASL signals. Including spatial regularization in the SPA modeling would improve
the reliability of kw estimation, which typically utilizes the TV metric. The TGV is an improved
mathematical framework based on minimizing both rst- and second-order TV to avoid blotchy
50
appearances commonly observed in TV-constrained image reconstruction [99], which has also been
applied for ASL denoising [101]. In the present study, we were able to preserve the original image
resolution, minimizing spuriously high kw values while improving SNR using TGV regularized SPA
modeling. A good ICC = 0.75 was achieved for test and retest kw measurements in aged subjects.
As compared to the relatively poor ICC from 2D kw measurements, combing 3D acquisition and
TGV signicantly improve the reliability of water exchange measurement. Sensitivity analysis of
kw versus weighting factor was performed by calculating kw in a representative subject with
varying from 0.01 to 0.10 at a step size of 0.01, around5% changes of kw was observed as
compared to the kw calculated with = 0.05. Using the ADMM algorithm, the average calculation
time was within 1 minute on a stand-alone computer (2.3-GHz dual-core processor).
51
Chapter 5
Initial evaluation of water exchange across the BBB in
elder subjects at risk of SVD
There is growing evidence indicating that BBB permeability increases with advancing age, and
these changes are accelerated in microvascular disease and dementia [41, 42, 40]. Loss of BBB
integrity may contribute to progression of SVD by allowing neurotoxin access to the brain and
causing ionic imbalance, an in
ammatory response around vessels, and eventually demyelination of
WM bers [31]. Elevated levels of albumin, which does not cross the intact BBB, in cerebrospinal

uid (CSF) has been reported in patients with vascular dementia [48, 49]. BBB dysfunction
has also been implicated in the pathogenesis of AD [102, 103]. Currently, assessment of BBB
permeability relies on CSF sampling and/or DCE-MRI using GBCAs. Biochemical assays of
CSF require lumbar puncture whereas DCE-MRI requires administration of contrast and long
scan time (>15 minutes). In addition, because albumin (66 kDa) and contrast agents (550 Da)
have relatively large molecular weights, BBB permeability has to reach a critical level before
extravasation occurs.
52
5.1 Clinical assessments
Subjects underwent a physical exam, medical history evaluation (hypertension, diabetes, and
hypercholesterolemia), and blood draw before the rst MRI scan. Presence or absence of hy-
pertension, diabetes, and hypercholesterolemia was dened by a past diagnosis and/or current
treatment for these conditions. Vascular risk factor (from 0 to 3) was calculated as the combi-
nation of presences of hypertension, diabetes, or hypercholesterolemia. Neuropsychological as-
sessment was performed using the Alzheimers Disease Centers Uniform Data Set v3 (UDS3) as
well as the NIH toolbox. Volumes of white matter hyperintensity (WMH) was manually seg-
mented by a clinical fellow from T2-weighted
uid-attenuated inversion recovery (FLAIR) images
(resolution = 1 1 1 mm
3
, inversion time/TE/TR = 1800/388/5000 ms) using ITK-SNAP
(www.itksnap.org) [104]. The Fazekas scale of WMH was rated for each subject according to
[105]. Clinical information and descriptions of all clinical assessments are summarized in Table
5.1.
Correlation between average kw from both test and retest scans and clinical/behavioral as-
sessments were evaluated using a mixed-eects linear regression model implemented in STATA
software (version 13.1; StataCorp LP, College Station, TX), incorporating age and sex as covari-
ates and time (test and retest) as the random variable. Mixed eects linear regression was also
performed to evaluate the correlation between average kw and CBF from test and retest scans.
Two signicant levels were set as P value less than 0.05 and 0.005 (2-sided).
5.2 Correction between kw and vascular risk factors
Table 5.2 summarizes the results of mixed-eects model analysis between kw (whole brain/GM/WM)
and vascular risk factors: hypertension, diabetes and hypercholesterolemia. We found no signif-
icant correlation between average kw and gender ( = -10.3; P = 0.28). A positive trend was
53
Measurement Statistics/description
Medical
history
Hypertension 13 subjects (68.4%)
Diabetes 6 subjects (31.6%)
Hypercholesterolemia 14 subjects (73.7%)
Vascular risk factor Combination of presences of hypertension, diabetes, or hyper-
cholesterolemia (rated from 0 to 3). 3/3/9/4 subjects were
rated as 0/1/2/3
Alzheimers
Disease
Centers
Uniform Data
Set v3 (UDS3)
CDR scale
Sum of
Boxes
(CDR-SB)
Normal (0) 10 subjects (52.6%)
Questionable cognitive impair-
ment (0.5, greater scores indi-
cate more-severe impairment)
9 subjects (47.4%)
Global
score
(CDR-GS)
Normal(0) 10 subjects (52.6%)
Questionable cognitive impair-
ment (0.5)
9 subjects (47.4%)
Montreal Cognitive As-
sessment (MoCA)
A measure of visuospatial construction, executive function,
verbal memory, attention, working memory, language, and ori-
entation; score 26 considered as normal (score range, 0-30)
NIH toolbox
Flanker The Flanker is a measure of attention and inhibitory control;
higher Flanker scores indicate higher level of ability to attend
to relevant stimuli and inhibit attention from irrelevant stim-
uli.
Dimensional Change
Card Sort Test
(DCCS)
The DCCS is a measure of cognitive
exibility; higher DCCS
scores indicate higher level of cognitive
exibility.
Picture Sequence
Memory Test (PSMT,
version a and b)
The PSMT is a measure of episodic memory, which involves
the acquisition, storage, and eortful recall of new informa-
tion; higher PSMT scores indicate better episodic memory.
Pattern Comparison
Processing Speed Test
(PCPS)
The PCPS is a measure of speed of processing for pattern
comparison; higher PCPS scores indicate faster speed of pro-
cessing.
Pegboard Dexterity
Test (dominant hand)
The test records the time (seconds) required for a participant
to place and remove 9 plastic pegs into a plastic pegboard;
faster completion time indicates better manual dexterity.
Grip Strength Test
(dominant hand)
The test records the force (pounds) of a participant squeezing
a digital hand dynamometer; greater force indicates greater
strength.
4-meter Walking Gait
Speed Test
The test records the time (seconds) required for a participant
to walk 4 meters at usual pace; shorter time indicates better
gait speed, as a measure of bipedal motion.
White matter
hyper-intensity
(WMH)
Volume Volume of WMH regions manually measured by clinical fellows
from 3D T2 FLAIR images.
Total Fazekas scale Quantication of WMH lesions. Total Fazekas scale is the
sum of 2 scales rated from 0 (absent) to 3 (large con
uent
areas) in periventricular white matter and deep white matter.
1/1/15/1/1 subjects were rated as 0/1/2/3/4.
Table 5.1: Summary of clinical assessments performed in this study.
54
found between average kw and age which matches previous studies reporting increased BBB per-
meability with aging [41, 42, 105], although no signicant level was achieved ( = 0.49;P = 0.43)
in this study.
We found signicantly increased kw in subjects with type 2 diabetes ( = 25.7; P < 0.001;
Figure 5.1 (a)) and hypercholesterolemia ( = 17.8;P = 0.04; Figure 5.1 (b)), which is consistent
with DCE-MRI [51] and biochemical studies [106]. Both of diabetes and hypocholesterolemia have
emerged as risk factors of SVD and AD. Hypercholesterolemia has been known to be associated
with vascular pathology and dysfunction, including vascular in
ammation and atherosclerosis,
which may lead to early breakdown of the BBB [106]. Diabetes mellitus leads to glycosylation of
endothelial proteins and also causes the basement membrane in the vessel wall to grow abnormally
thicker and weaker. As a result, the microvessels in the brain and body of diabetic subjects are
susceptible to microbleeds, protein leakage, and hypoperfusion [44]. Population based studies have
shown that both diabetes and hypercholesterolemia lead to increased risk of neurodegeneration,
cognitive impairment, and dementia [107, 108].
The ecacy of treating hypertension to restore BBB integrity and prevent cognitive impair-
ment is still controversial. While recent animal study showed that inhibited renin-angiotensin
system (RAS) expression ameliorated the BBB leakage and restored the cognitive decline [109],
studies also pointed out that the eectiveness of prevention of dementia by lowering blood pressure
needs more evaluations for aged people (> 80 yr) [110]. In this study, we didn't nd signicant
change of average kw in subjects with chronic hypertension ( = -2.6,P = 0.78). The correlation
between kw and hypertension could be complicated due to fact that a common characteristic of
hypertension is persistent intravascular volume expansion [111], which leads to increased intrava-
sular space and potentially has a counter eect to the change of kw.
Vascular risk factors are related to the impaired cognitive functions and occurrence of both
vascular dementia and Alzheimer's Diseases [112]. The vascular risk factor was dened as the
combination of presences of hypertension, diabetes and hypercholesterolemia (rated from 0 to 3)
55
Clinical
Age Gender (F-0,M-1) Hyper-tension
kw
Whole brain 0.49 (0.43) -10.3 (0.28) -2.6 (0.78)
GM 0.46 (0.43) -9.2 (0.31) -2.8 (0.75)
WM 0.44 (0.50) -11.5 (0.26) -1.8 (0.85)
Diabetes Hypercholesterolemia Vascular risk
kw
Whole brain 25.7** (<0.001) 17.8* (0.04) 9.4* (0.02)
GM 24.5** (<0.001) 16.7* (0.05) 8.8* (0.02)
WM 26.2** (0.001) 20.2* (0.03) 10.3* (0.01)
Table 5.2: Repeated measures mixed-eects linear regression coecients for kw and clinical
measurements. P values are listed in the parentheses. Signicant correlations with P values
smaller than 0.05 and 0.005 are indicated by asterisks.
in this study. Signicantly increased kw was found in subjects with higher vascular risk factor
( = 9.4; P = 0.02; Figure 5.1 (c)). Our observation of increased kw in subjects with diabetes
and hypercholesterolemia and total vascular risk factors is consistent with existing literature,
suggesting that kw may provide a surrogate imaging biomarker of cerebral eects of common
vascular risk factors and early SVD and/or AD [51].
5.3 Correction between kw and cognitive measurements
Cognitive function of each subject was assessed by two scoring system: Alzheimer's Disease Cen-
ter's Uniform Data Set (UDS) [113] and NIH toolbox. Two scores were generated from UDS:
Clinical Dementia Rating Scale (CDR) and Montreal Cognitive Assessment (MoCA). Table 5.4
summarizes the results of mixed-eects model analysis between kw (whole brain/GM/WM) and
UDS measurements. Table 5.5 summarizes the results of mixed-eects model analysis between
kw (whole brain/GM/WM) and NIH toolbox measurements.
5.3.1 CDR and MoCA
The CDR scoring system is used to characterize cognitive and functional performance applicable
to Alzheimer disease and related dementia from the following six domains: Memory, Orientation,
56
Figure 5.1: (a,b) Bar plot of average kw in normal subjects versus subjects with diabetes (a)
and hypercholesterolemia (b). (c) Bar plot of average kw versus vascular risk factors. Error bars
indicate standard deviation of kw across subjects.
57
CDR score 0 0.5 1 2 3
Level of dementia Normal Very Mild Mild Moderate Severe
Table 5.3: Characterizing level of dementia using CDR scores
Judgment and Problem Solving, Community Aairs, Home and Hobbies, and Personal Care. The
subject was asked to take a semi-structured interview and CDR scoring system yields two scores:
global score (CRD-GS) and Sum of Boxes (CDR-SB) scores, while both scores have been used
to determine dementia severity [113, 114]. The CDR scoring table (Table 5.3) explains how to
characterize a subject's level of dementia using CDR scores.
In this study, 52.6% subjects are considered as normal (CDR-GS = 0) while 47.4% subjects
show very mild dementia (CDR-GS = 0.5). Thus, all subjects recruited in this study are within the
early stage of dementia. Both the global (CDR-GS, = 44.6; P = 0.002) and sum of box scores
(CDR-SB, = 21.0; P = 0.001) were signicant predictors of kw (Figure 5.2), which indicates
that increased BBB permeability is associated with a greater severity of functional impairment.
The MoCA scoring system has also been widely used as a screening assessment for detecting
cognitive impairment, which assesses the following cognitive domains: The short-term memory;
Visuospatial abilities; Executive functions; Attention, concentration, and working memory; Lan-
guage and orientation to time and place. MoCA scores range between 0 and 30 and average
MoCA score of 26 or over is considered to be normal. Average MoCA score was reported to be
27.4, 22.1 and 21.6 for normal subjects, subjects with mild cognitive impairment and subjects
with Alzheimer's disease [115]. In our study, the average MoCA score was 21.64.6 across nine-
teen subjects, which indicates the subjects recruited in this study might experience mild cognitive
impairment. However, We didn't nd signicant correlation between average kw and MoCA (
= -0.86, P = 0.45). more experiments are required to have a better understanding between kw
and MoCA scores.
58
UDS
CDR-SB CDR-GS MoCA
kw
Whole brain 21.0** (0.001) 44.6** (0.002) -0.86 (0.45)
GM 20.3** (0.001) 43.7** (0.002) -0.79 (0.46)
WM 22.2** (0.001) 46.9** (0.003) -0.95 (0.41)
Table 5.4: Repeated measures mixed-eects linear regression coecients for kw and UDS mea-
surements. P values are listed in the parentheses. Signicant correlations with P values smaller
than 0.05 and 0.005 are indicated by asterisks.
5.3.2 NIH toolbox
The NIH Toolbox (http://www.healthmeasures.net/) is a comprehensive set of neuro-behavioral
measurements which assess cognitive, emotional, sensory, and motor functions. In this study, we
have evaluated the correlation between kw and 7 NIH toolbox measurements: Flanker score, Di-
mensional Change Card Sort Test (DCCS), Picture Sequence Memory Test (PSMT, version a and
b), Pattern Comparison Processing Speed Test (PCPS), Pegboard Dexterity Test, Grip Strength
Test and 4-meter Walking Gait Speed Test. Descriptions about NIH toolbox measurements are
summarized in Table 5.1.
In this study, We found signicant correlations between average kw and DCCS ( = -1.10;
P = 0.02, gure 5.3 (c)), PSMTa ( = -0.98; P = 0.03, gure 5.3 (b)) and PSMTb ( = -1.19;
P = 0.001, gure 5.3 (d)), which indicates that increased BBB water permeability is associated
with a lower level of cognitive
exibility and worse episodic memory. We also observed a trend
of negative correlation between average kw and Flanker ( = -0.58; P = 0.08, gure 5.3 (a)),
which indicates that increased BBB water permeability is associated with a trend of decreased
attention/inhibitory control.
Regarding motor functions, we didn't nd signicant correlation between average kw and
results of pegboard dexterity test ( = 1.59,P = 0.15), grip strength test ( = 0.24,P = 0.68) or
the 4-meter walking gait speed test ( = 2.29, P = 0.70). More experiments need to be conduct
for a better understanding of how motor function changes aect kw.
59
Figure 5.2: (a,b) Bar plot of average kw versus clinical dementia rating scales CDR-SB (a) and
CDR-GS (b). Error bars indicate standard deviation of kw across subjects.
NIH toolbox
Flanker DCCS PSMTa PSMTb
kw
Whole brain -0.58 (0.08) -1.10* (0.02) -0.98* (0.03) -1.19** (0.001)
GM -0.57 (0.07) -1.09* (0.01) -0.97* (0.02) -1.15** (<0.001)
WM -0.57 (0.12) -1.09* (0.03) -0.99* (0.04) -1.31** (<0.001)
PCPS Dexterity Strength WGS
kw
Whole brain -0.37 (0.28) 1.59 (0.15) 0.24 (0.68) 2.29 (0.70)
GM -0.35 (0.28) 1.51 (0.15) 0.24 (0.67) 2.33 (0.68)
WM -0.37 (0.31) 1.52 (0.20) 0.30 (0.63) 2.70 (0.68)
Table 5.5: Repeated measures mixed-eects linear regression coecients for kw and NIH toolbox
measurements. P values are listed in the parentheses. Signicant correlations with P values smaller
than 0.05 and 0.005 are indicated by asterisks.
60
Figure 5.3: (a-d) Scatter plots of average kw versus NIH toolbox measurements: Flanker (a),
DCCS (c), PSMTa (b), and PSMTb (d). Slopes and R
2
of linear regressions (without controlling
age/sex, indicated by the black dashed lines) are listed in each scatter plot.
61
WMH
Volumes Fazekas scale
kw
Whole brain 1.68 (0.20) 10.61* (0.04)
GM 1.72 (0.16) 10.53* (0.03)
WM 1.75 (0.21) 11.07* (0.04)
Table 5.6: Repeated measures mixed-eects linear regression coecients for kw and WMH
measurements. P values are listed in the parentheses. Signicant correlations with P values
smaller than 0.05 and 0.005 are indicated by asterisks.
5.4 Correction between kw and WMH measurements
Volume of WMH regions was manually measured by clinical fellows from 3D T2 FLAIR images,
and total Fazekas score was used to quantify the severity of WMH. Total Fazekas scale is the
sum of two scales rated from 0 (absent) to 3 (large con
uent areas) in both periventricular white
matter and deep white matter.
We found that increased kw was signicantly associated with increased total Fazekas scale (
= 10.64, P = 0.04), which indicates that higher kw should be expected in subjects with more
severe WMH lesions. We also observed a trend of positive correlation between kw and WMH
volume. A pathological report has associated WMH with demyelination and axonal loss [116],
and clinical studies have shown associations between WMH and progressive cognitive impairment
and increased risk of dementia [36].
Although previous studies reported globally reduced CBF in cerebral SVD patients with
greater WMHs [117], the kw changes in 19 subjects with potential SVD were not signicantly
associated with CBF changes in this study ( = 0.35, P = 0.22, gure 5.5). ASL labeled blood
serves as an intravascular tracer, which might leak into extravasular space with leaky BBB, thus
absolute CBF could be under-estimated and interpolation of the blood
ow needs to be cautious.
Subjects recruited in this study are in the early stages of WMH development (average WMH
volume is 2.6 cm
3
), and its association with kw will provide important opportunities to prevent
brain damage attributed to SVD in the earliest stages and ameliorate cognitive impairment.
62
Figure 5.4: Bar plot of average kw versus Fazekas scale. Error bars indicate standard deviation
of kw across subjects.
Figure 5.5: Scatter plot of average kw versus CBF. kw and CBF from both test and retest
experiments are displayed in the scattor plot.
63
As an initial validation study, results discussed in chapter 4 and 5 have been published in Shao,
et al, MRM, 2019 [88]. A cohort of elderly Latinos subjects were enrolled from the MarkVCID
study. MarkVCID is a consortium of US academic medical/research centers and the goal is to
identify and validate biomarkers for the cerebral SVD that produce VCID. Among the nineteen
subjects recruited in this study, 52.6% were considered as normal while 47.4% were identied with
mild cognitive impairment (CDR scale = 0.5). This pre-clinical population are in the early stage
of dementia, which might be associated with subtle BBB permeability change. While previous
studies reported decreased kw with disrupted BBB [80, 81, 82], a trend of increasing kw in
subjects with worse cognitive function was observed in this study, which is consistent with the
ndings in early cognitive impairment/AD [54, 72] and shows a good accordance with the proposed
bidirectional water permeability change theory (section 2.3).
64
Chapter 6
Preliminary comparison of water permeability and BBB
permeability to contrast agents
DCE-MRI is a commonly used technique to assess BBB leakage to contrast agents in both human
and animal studies. This chapter presents the preliminary results comparing Ktrans and kw
acquired from the same cohort of aged subjects. With higher spatial resolution, studies have
detected BBB permeability changes associated with neurodegenerative diseases in small WM and
deep GM ROIs [40, 42, 67, 118, 119]. Regional Ktrans and kw were measured and compared
in ROIs of MCA perforator territory, basal ganglia nuclei, medial temporal lobe (MTL) and
MTL subregions such as hippocampus. Correlations between regional Ktrans/kw and cognitive
functions were studied. The goal of this preliminary study was to nd potential correlations
between kw and Ktrans changes at early stage of BBB opening, and the sensitivity of regional
Ktrans and kw changes in cognitive decline. While the safety of Gadolinium MRI has been
debated, the proposed water permeability measurements would open a new path to study BBB
permeability non-invasively.
65
6.1 DCE-MRI acquisition and modeling of Ktrans
Both DCE-MRI and the proposed 3D DP-GRASE pCASL were performed in 16 aged subjects (3
male; age = 67.9 3.0 years, all Latinos) from the MarkVCID cohort. DCE-MRI scan consisted
of a pre-contrast T1-mapping protocol and a dynamic T1-weighted acquisition with contrast agent
injection. T1 map was estimated from 3D fast low angle shot (FLASH) images with 5 dierent

ip angles: 2
0
, 5
0
, 10
0
, 12
0
, and 15
0
. Imaging parameters for T1 mapping were: FOV = 175175
mm
2
, 14 slices, resolution = 1.11.15 mm
3
, TE = 2.18 ms, TR = 5.14 ms. Imaging parameters
for dynamic T1w acquisition were: FOV = 175175 mm
2
, 14 slices, resolution = 1.11.15
mm
3
, TE = 3 ms, TR = 8 ms, temporal resolution = 15 s, 64 frames were acquired with a
total acquisition time of 16 mins. Both scans were acquired along coronal direction and covered
basal ganglia and medial temporal lobe (MTL). 20 mL of contrast agent (Dotarem R
, Gadoterate
meglumine, 0.5 mmol/mL) was injected after 30 s of image acquisition with an average injection
rate of 3 mL/s.
A linear model was used to t T1 values from images with dierent
ip angles, and gure 6.1
(a) shows a representative slice of T1 map. Dynamic T1w images were corrected for rigid head
motion using SPM12. Noise ltering, signal drift correction, and model tting were performed
using ROCKETSHIP [120]. Ktrans andv
p
were tted from dynamic signal intensities using Patlak
model (equation 6.1), which has good performance in low permeability regions [67]:
C
tissue
(t) =K
trans
Z
t
0
C
AIF
()d +v
p
C
AIF
(t) (6.1)
where C
tissue
(t) and C
AIF
(t) are the contrast agent concentration curves for tissue and AIF,
respectively, and individual C
AIF
(t) was manually measured in MCA.
Both Ktrans and kw maps were normalized into the MNI space, and regional analysis was
performed in MCA perforator territory, caudate, MTL and subregions including hippocampus,
parahippocampal gyrus (PHG) and amygdala. Small lenticulostriate arteries (LSAs, diameter of
66
Figure 6.1: A representative slice of T1 (a) and Ktrans map measured from a 63-year-old subject.
Ktrans (10
3
min
1
) kw (min
1
)
MCA perf 0:9 0:3 86:0 13:5
Caudate 0:8 0:3 110:2 24:2
MTL 1:1 0:3 103:0 18:3
Hippocampus 1:3 0:4 106:5 20:5
Parahippocampal gyrus 1:3 0:3 82:5 22:2
Amygdala 1:1 0:5 117:6 25:0
Table 6.1: Regional kw and Ktrans values.
a few hundred microns) are originated directly from MCA with high blood
ow velocities, making
them especially susceptible to damage. Leakage of plasma proteins from small perforating arteries
in the MCA perforator territory has been associated with WML, sub-cortical dementia and AD
[48, 49, 50]. Caudate is a component of the basal ganglia and BBB leakage in caudate has been
associated with WML and SVD [121]. The MTL is a conceptual region and is critically involved
in learning and memory. Hippocampus is the most signicant component of the MTL, which is
crucial for episodic memory and consolidation processes, and BBB breakdown in hippocampus
has been associated with cognitive impairment in the aging brain [42]. Regional kw and Ktrans
measurements are summarized in table 6.1.
67
6.2 Correlation between BBB permeability to water (kw)
and contrast agent (Ktrans)
Linear regressions were performed between regional kw and Ktrans measured from 6 ROIs across
16 subjects. P < 0.05 was considered as signicant. Table 6.2 summarizes the correlation coe-
cients and P values. Average Ktrans in Caudate was signicantly correlated with average kw in
Caudate (r = 0:49;P = 0:05), MTL (r = 0:54;P = 0:03) and Hippocampus (r = 0:53;P = 0:03),
while a trend of positive correlation was found between regional kw measurements and average
Ktrans in MCA perforator territory. Figure 6.2 and 6.3 show the scatter plots between regional
kw and average Ktrans in MCA perforator territory and caudate, respectively.
Considering the dierent size of water molecule and contrast agent, and dierent transport
mechanisms across the BBB, correlation between kw and Ktrans could be complicated and depend
on level of BBB opening. Studying kw and Ktrans could provide comprehensive understanding
about changes in neurovascular unit function and BBB permeability with aging and neurodegen-
eration. We observed a signicant positive correlation between kw and Ktrans in the caudate.
Since BBB leakage in caudate has been associated with WML and cerebral SVD [121], our prelim-
inary results indicate BBB permeability to both water and contrast agent could serve as sensitive
imaging markers for SVD. However, further studies in a larger cohort of subjects and specic
disease models are necessary to reveal the connection between kw and Ktrans at dierent BBB
opening levels.
68
Figure 6.2: Scatter plot between regional kw measurements and average Ktrans in MCA perforator
territory. Dashed lines represent the tted linear regression curve.
69
Figure 6.3: Scatter plot between regional kw measurements and average Ktrans in Caudate.
Dashed lines represent the tted linear regression curve.
70
Ktrans
MCA perf Caudate MTL Hippocampus PHG Amygdala
kw
MCA perf 0.29 (0.28) 0.39 (0.14) 0.11 (0.69) -0.08 (0.78) -0.05 (0.85) -0.27 (0.31)
Caudate 0.41 (0.11) 0.49 (0.05) 0.23 (0.40) 0.09 (0.73) -0.04 (0.89) -0.14 (0.62)
MTL 0.47 (0.07) 0.54 (0.03) 0.18 (0.51) -0.02 (0.94) -0.19 (0.49) -0.19 (0.48)
Hippocampus 0.38 (0.14) 0.53 (0.03) 0.09 (0.73) -0.17 (0.53) -0.30 (0.27) -0.22 (0.40)
PHG 0.50 (0.05) 0.48 (0.06) 0.22 (0.42) -0.00 (0.99) -0.24 (0.36) -0.04 (0.90)
Amygdala 0.42 (0.10) 0.35 (0.18) 0.15 (0.58) -0.17 (0.53) -0.14 (0.61) -0.08 (0.77)
Table 6.2: Correlation coecients between regional kw and Ktrans measurements (P values were
listed in brackets).
6.3 Correlation between regional BBB permeability and
cognitive measurements
To assess correlations between Ktrans/kw and cognitive measurements, linear regressions were
performed between regional kw/Ktrans and Flanker (attention), PSMTa (episodic memory) and,
DCCS (cognitive
exibility). P < 0:5 was considered as signicant.
Figure 6.4 shows the scatter plot of regional Ktrans and kw versus Flanker, a measurement of
attention and inhibitory control. Signicant correlations were found between Flanker and Ktrans
in MCA perforator territory ( =26:3 10
3
;P = 0:01), caudate ( =32:8 10
3
;P = 0:01),
MTL ( =34:6 10
3
;P = 0:02), and kw in MCA perforator territory ( =0:61;P = 0:03),
amygdala ( =0:31;P = 0:04). These results suggest that poor attention and inhibitory control
could be associated with BBB leakage of GBCAs in MCA perf, caudate, MTL and increased water
exchange in MCA perf and amygdala.
Figure 6.5 shows the scatter plot of regional Ktrans and kw versus PSMTa, a measurement of
episodic memory. Signicant correlations were found between PSMTa and kw in MCA perforator
territory ( =0:48;P = 0:02), caudate ( =0:24;P = 0:04), MTL ( =0:37;P = 0:02),
hippocampus ( =0:28;P = 0:05), parahippocampal gyrus ( =0:33;P = 0:02) and amyg-
dala ( =0:32;P = 0:003). However, no signicant correlation was found between PSMTa and
71
regional Ktrans measurements. These results suggest that kw could be a sensitive biomarker for
episodic memory, and poor episodic memory could be associated with increased water permeability
in MCA perforator territory, basal ganglia and MTL.
Figure 6.6 shows the scatter plot of regional Ktrans and kw versus DCCS, a measurement
of cognitive
exibility. Signicant correlations were found between DCCS and Ktrans in MCA
perforator territory ( =17:6 10
3
;P = 0:03), caudate ( =23:3 10
3
;P = 0:02), MTL
( =26:2 10
3
;P = 0:02) and amygdala ( =12:5 10
3
;P = 0:03), and kw in amygdala
( =0:26;P = 0:02). These results suggest that poor cognitive
exibility could be associated
with BBB leakage of GBCAs in MCA perf, caudate, MTL and increased water exchange in
amygdala.
Our preliminary results have shown cognitive function decline might be associate with in-
creased BBB permeability to water or contrast agents in specic regions, while kw change is more
sensitive to episodic memory. Although signicant correlations were not found in some of these
ROIs, the trend was quite clear that kw and Ktrans increase with poor attention and inhibitory
control, episodic memory and cognitive
exibility.
DCE-MRI has been commonly used for decades, while the debts over the potential brain
deposition of Gadolinium became heated in recent years. Both patients and radiologists started
to question the safety of Gadolinium MRI back to 2013 when a study showed deposition and
retention of Gadolinium based contrast agents in the brain [122]. And recently both the US Food
and Drug Administration and ISMRM have issued statements to limit the use of GBCAs to clinical
circumstances [69]. With the proposed technique, we have demonstrated that water permeability
could be a surrogate imaging marker of cerebral SVD and associated cognitive impairment, and
our preliminary results suggest potential correlation between kw and Ktrans. Considering the
facts that limited number of subjects were recruited and size of selected ROIs are small, these
preliminary results are encouraging that kw could be an alternative imaging marker to Ktrans for
72
assessment of cognitive functions in the future, and might be more sensitive due to small water
molecule size and special water transport mechanisms.
73
Figure 6.4: Scatter plots of regional Ktrans (a-f) and kw (g-l) versus Flanker. Dashed lines
represent the tted linear regression curve.
74
Figure 6.5: Scatter plots of regional Ktrans (a-f) and kw (g-l) versus PSMTa. Dashed lines
represent the tted linear regression curve.
75
Figure 6.6: Scatter plots of regional Ktrans (a-f) and kw (g-l) versus DCCS. Dashed lines represent
the tted linear regression curve.
76
Chapter 7
Discussion and ongoing work
A novel 3D diusion prepared ASL technique was developed to map water exchange across the
BBB in aged subjects at risk of SVD. The proposed technique uses water as an endogenous tracer
and generates whole brain map within 10 minutes, which is shorter than the current DCE-MRI
protocols and applicable to both research and clinical studies. This dissertation starts with an
introduction of the structure and function of BBB. Passive diusion was thought to be the major
mechanism for water to exchange across the BBB. However, recent studies discovered a water
channel named aquaporin, which is abundant in endothelial cells and end-foot of astrocytes and
transports water across the BBB more eciently.
Chapter 2 discussed potential mechanisms leading to BBB disruption and resultant parenchy-
mal damage. Recently the direct role of pericyte in maintaining BBB integrity was studied using
unique animal mutants with pericyte deciency, and increased water permeability was observed
with pericyte loss. We proposed to assess early BBB permeability change through measuring wa-
ter exchange across the BBB, however, water transport is aected by both AQP4 expression and
BBB opening. A bi-directional water permeability change was hypothesized: increased water per-
meability at the early stage of BBB opening, then water permeability decreases after chronic BBB
leakage and resultant parenchymal damage. Traditional MRI techniques for BBB permeability
assessment were discussed and compared with emerging techniques assessing the water exchange
77
across the BBB. DCE-MRI has been commonly used to quantify BBB leakage in diseases using
Gadolinium based contrast agents. However, a major limitation of the DCE-MRI is the large
molecule size of contrast agents which prohibits its crossing the BBB with subtle permeability
change. Since water is much smaller than Gadolinium based contrast agents and water transport
across the BBB is mainly facilitated by aquaporins, water permeability (kw) could serve as a
sensitive biomarker to study the BBB dysfunction and subtle BBB permeability change in aging
and early stage of neurological disease such as SVD.
Chapter 3 introduced the biophysical basis of separating and quantifying the intra-/extravasular
components using diusion and perfusion weighted MRI. Diusion weighted MRI is a technique
to characterize the microscopic Brownian motion of water and local diusivity can be quantied
by diusion coecients. Since capillary vessels are pseudo-randomly orientated, capillary blood
signal is also sensitive to diusion gradients and its pseudo-diusion coecients is usually bigger as
compared to tissue blood. ASL is a non-invasive perfusion technique, which tags incoming blood
at carotid and traces the blood passage through arteries, arterioles, capillaries and exchange with
tissue. We found a 100-fold dierence between diusion coecients of tissue blood and capillary
blood, and a small diusion weighting could be sucient to suppress the intravasular component
of the ASL signal while has minimal eect on tissue blood signal. Details about the pulse sequence
design were also discussed in this chapter.
Chapter 4 introduced the mathematical models for kw and CBF quantication. The SPA
model was proposed to calculate kw from diusion weighted ASL signals. However, it was chal-
lenging to generate reliable kw map due to low SNR of previously used 2D-EPI readout. SNR
of diusion weighted ASL signals increased signicantly with the proposed 3D DP-GRASE read-
out, and we also proposed a novel TGV regularized SPA modeling algorithm to further improve
the reliability of kw quantication. TGV regularization has shown promising eects minimizing
spuriously high kw values while improving SNR. Test and retest reproducibility was evaluated
by scanning the same subject twice (two weeks apart) and ICC was calculated from nineteen
78
subjects' repeated kw measurements. Good ICC (0.75) was achieved for whole brain average
kw while using the proposed 3D DP GRASE sequence and TGV regularized SPA modeling, and
the ICC is comparable to current ASL studies. However, ICC was still low especially in smaller
regions (i.e. hippocampus). In order to perform regional analysis of kw in diseases, improving
spatial resolution and SNR will be the major tasks in the future.
Clinical evaluations of the proposed kw measurements were discussed in chapter 5. Recruited
subjects are at risk of SVD and early stage of cognitive impairment. Vascular risk factors were di-
agnosed from blood draw and neuropsychological assessment was performed using the Alzheimers
Disease Centers Uniform Data Set and the NIH toolbox. Volume and severity of WMH lesions
were also measured from T2-FLAIR images. Since all participants were aged subjects, no signi-
cant correlation was found between age and kw in this study. Increased kw has been associated
with type 2 diabetes, hypercholesterolemia and total vascular risk factors, which is consistent
with DCE-MRI and biochemical studies. Increased kw has also been associated with cognitive
impairment and mild dementia, as conrmed by correlation with CDR scores and NIH toolbox
measurements. Increased kw was also correlated with increased severity of WMH lesions. Assess-
ing WMH lesions has been used for clinical diagnosis of SVD, and its association with kw will
provide important opportunities to prevent brain damage attributed to SVD at the earliest stages
and ameliorate cognitive impairment.
Chapter 6 discussed preliminary results of comparison between kw and Ktrans. Regional
kw and Ktrans were measured in several small ROIs that are associated with WML, SVD and
cognitive decline. Signicant correlations were found between Ktrans and kw in caudate, and a
trend of positive correlation was observed between regional kw measurements and Ktrans in MCA
perforator territory, which indicate BBB permeability to both water and contrast agent could serve
as sensitive imaging markers for cerebral SVD. Our preliminary results also suggest regional kw
and Ktrans changes are sensitive to dierent aspects of cognitive function, such as attention,
episodic memory or cognitive
exibility. Since the safety of Gadolinium MRI are debated, water
79
permeability measurements could be a promising imaging marker for assessment of cognitive
functions considering the small water molecule size and special water transport mechanisms.
In the rest of this chapter, limitations of the proposed DP-GRASE sequence and experiment
design were discussed. Potential approaches to further improve the SNR, spatial resolution, and
reliability of kw quantication were explored. And animal studies, which could help better under-
stand the underlying mechanisms of vasculature and permeability changes in dierent diseases,
will be conducted in the future.
7.1 Limitations of the current study
There are limitations of this study. Because segmented acquisition introduces inter-segment phase
inconsistency and shading artifacts, single-shot acquisition is required for the proposed DP 3D
pCASL sequence. Resolution of the kw/ATT map is relatively low as compared to standard ASL
studies (also to compensate for half signal loss). To improve spatial resolution, fast imaging, such
as 2D controlled aliasing in volumetric parallel imaging [95], and reconstruction algorithm with
spatial and temporal constraints will be utilized [101] (Details are discussed in section 7.2).
For comparison of 2D and 3D kw measurements, the sample size of the 2D experiment was
small. Presence of arterial and venous compartments, which were considered as nonexchangeable
compartments, may bias the capillary/tissue fraction estimation. The PLD of 1800 ms was chosen
to exclude/ minimize the arterial and venous compartments, because ATT was estimated to be
1200 to 1300 ms in this study and Lin et al [73] reported detectable venous signal at PLD >2500
ms. Recent studies also reported water exchange in periarterial and -venous spaces through
aquaporin [123, 124]. This study has demonstrated the potential of kw as a sensitive marker of
BBB permeability. However, Vc may alter in diseases (e.g., decreased Vc in diabetes attributed
to thicken vessel wall and increased perivascular space) and complicates the understanding of
the relation between kw and PSw. With the proposed sequence, total extraction ratio Ew and
80
PSw can be computed with DP 3D pCASL signals acquired at longer PLD (>2.5 seconds) [73],
which remains to be explored in future studies. Studies also reported reduced vascular volume
and increased BBB permeability, both of which leads to increased kw, in hippocampus of MCI
subjects [119]. While CBF and PSw were reported to be signicantly correlated [77, 92], kw
may serve as a more special and sensitive surrogate biomarker for brain functional change in the
process of cognitive decline.
7.2 Temporal regularized TGV
One major limitation of ASL perfusion weighted imaging is the low SNR due to the fact that ASL
signal is usually less than 1% of background tissue signal. To achieve reliable image quality, ASL
signal is usually averaged from numerous repetitions and lengthy scan is required for higher spatial
resolutions (e.g. < 2 mm for in-plane resolution). Various de-noising methods have been proposed
by applying lters to the perfusion image or in the wavelet domain [125, 126, 127]. Spann and
colleagues systematically evaluated seven commonly used de-noisng approaches and proposed a
total generalized variation (TGV) based method for ASL [101]. Since ASL scan requires multiple
averages, the TGV method proposed by Spann et al incorporated temporal information, which
helps detect and eliminate outlier signals, to improve the SNR of the ASL images. The TGV
method has shown better image quality and robustness to noise [101].
In this study, we proposed a novel TGV regularized SPA model for ATT and kw quanti-
cation. Potentially TGV could also be benecial for improveing the SNR of ASL signal prior
to the modeling process. Since the proposed diusion prepared 3D GRASE sequence requires
single-shot acquisition, we have developed a single-shot 3D GRASE pCASL technique using 2D
CAIPI sampling strategy, and reconstruct the under-sampled data using ESPIRiT [95]. Figure 7.1
demonstrates the 2D CAIPI pattern with 2-fold under-sampling in both phase (ky) and partition
(kz) encoding directions results in a total acceleration factor of 4. This technique was successfully
81
Figure 7.1: Two under-sampling patterns and corresponding aliasing appearances are shown in
(a) and (b). Dashed circles indicate un-acquired k-space lines. 2-fold under-sampling in both
phase (ky) and partition (kz) encoding directions results in a total acceleration factor of 4.
applied for high resolution multi-delay ASL imaging. Figure 7.2 shows perfusion images from fully
sampled data acquired in 4 segments and single-shot acquisition with 4-fold under-sampling.
Total acquisition time was kept the same (4 min 33 sec) for both acquisitions. Temporal-SNR
(t-SNR) was 1.83 and 3.47 for single-shot and segmented acquisition respectively. Interestingly,
SNR of under-sampled signal could be slightly higher than that of fully sampled data if considered
the 4-fold time dierence acquiring one perfusion image, and potential reason could be due to the
de-nosing eect of parallel imaging reconstruction. Moreover, single-shot acquisition is more
robust to motion and provides higher temporal resolution, which allows acquiring ASL signal and
multiple PLDs. The single-shot GRASE pCASL sequence was also implemented for a multi-delay
ASL protocol: 4, 4, 4, 8 and 8 pairs of label/control images were acquired for PLD of 500, 1000,
1500, 2000 and 2500 ms respectively (Total scan time is 4 min 17 sec). Voxel-wise cerebral blood

ow (CBF) and arterial transit time (ATT) were quantied from the ve delay ASL signals using
a weighted-delay approach [128]. The T1 map was derived from background suppressed control
signals at ve PLDs (Similar to Appendix A). Figure 7.3 shows the simultaneously calculated
CBF, ATT and T1 map from the single-shot ve delay acquisition.
82
Figure 7.2: Perfusion map from (a) fully sampled data acquired in 4 segments and (b) single-
shot acquisition with 4-fold under-sampling. Total acquisition time was kept the same for both
protocols. Resolution = 333 mm
3
.
Figure 7.3: Simultaneously calculated CBF, ATT and T1 map from single-shot multi-delay
pCASL.
83
To further improve the acquisition eciency, we recently developed a 6-fold under-sampling
pattern, which is also shifted between averages to maximize the spatial and temporal incoherence
between repetitions. Figure 7.4 demonstrates the time-dependent under-sampling pattern for 6-
fold acquisition acceleration. The TGV algorithm could be adapted to reconstruct control and
label image series by incorporating both spatial and temporal regularization according to [129]
(equation 7.1):
arg min
c;l

c
2
kK
c
dck
2
2
+

l
2
kK
l
dlk
2
2
+
1
(s)TGV
1;0
(l) +
1
(s)TGV
1;0
(c)
+
2
(s)TGV
1;0
(cl)

(7.1)
where c and l are the target control and label image series to be reconstructed. d
c
and d
l
are acquired 4-D k-space signal of control and label series.
c
and
l
are the regularization
parameters for the control and label images. The parameter s controls the balance between three
TGV regularization terms.
0
and
1
are model parameters as described in [99]. The temporal
regularization is incorporated by the operator K which contains Fourier transform, time-domain
padding and under-sampling operators.
Combing TGV with temporal regularization has been proven to be a powerful tool to improve
the robustness of ASL signals [129], which is benecial for kw quantication. However, the
performance of spatial-temporal regularized TGV in diusion weighted ASL signals needs more
evaluations. In the future, more subjects will be recruited and scanned using the newly developed
time dependent 2D CAIPI sequence. And the possibility of combing temporal regularization with
TGV SPA modeling will be studied.
84
Figure 7.4: Time dependent 2D CAIPI under-sampling patterns for six consecutive repetitions.
Dashed circles indicate un-acquired k-space lines. 3-fold under-sampling in phase (ky) and 2-
fold under-sampling in partition (kz) encoding directions results in a total acceleration factor of
6. Spatial and temporal incoherence is maximized by shifting under-sampling patterns between
repetitions
85
7.3 Alternative methods combing diusion weighting and
TSE based sequences
3D GRASE was recommended by the ASL white paper [130] for clinical implementations of pCASL
perfusion MRI [87]. However, it has been challenging to combine diusion weightings with 3D
turbo-spin echo (TSE)-based sequences [131]. Diusion gradients induce extra phase attributed
to bulk motion (e.g., head movement or respiration). Violation of the CPMG condition causes
rapid signal decrease in regions where induced phase is not along Meiboom-Gill (MG) phase
direction, leading to dark bands or shades in images [131]. Ensuring the refocusing pulse to be
exactly 180 is the most straightforward approach to avoid phase sensitivity, which is not commonly
used because of specic absorption rate limitations, and a small deviation from 180
0
is sucient
to introduce artifacts. Figure 7.5 (a) shows 18 perfusion images after single label and control
subtraction acquired at PLD = 1800 ms with GRASE readout and a small diusion weighting
(b = 50 s=mm
2
). Traditional bipolar diusion gradients were directly added at the beginning of
GRASE readout and
ip angle of refocusing pulses was increased to 180
0
to maintain the CPMG
condition. However, severe signal
uctuations are still observable between perfusion maps.
Motion-compensated diusion preparation has been proposed to reduce the sensitivity of TSE
to bulk motion [132]. However, it is not suitable for the FEAST scheme to measure ATT given that
vascular signal is compensated. And eectiveness of motion compensation is highly dependent
on the motion patterns (order) during the diusion encoding. Figure 7.5 (b) shows 18 perfusion
images after a single label and control subtraction (PLD = 1800 and b = 50 s=mm
2
) ms with
GRASE readout and M1 (rst order, motion with a steady velocity) compensated diusion gra-
dients. Although motion compensation reduced signal
uctuations, artifacts (negative perfusion)
are still observable and the average perfusion map is not reliable for quantitative purpose. Higher
order motion compensation may be benecial, however, complicated gradient form will induce
eddy current artifacts and also leads to lower SNR due to increased TE.
86
Figure 7.5: A single slice of ASL perfusion images from 18 label/control pairs acquired at PLD
= 1800 ms with diusion weighting (b = 50 s=mm
2
) and GRASE readout. Diusion weighting
was induced by (a) traditional bipolar diusion gradients and 120
0
refocusing pulses in GRASE,
(b) diusion gradients with M1 motion compensation, and (c) the proposed diusion preparation
module. All images were acquired from the same subjects who was asked to avoid head motion
during the scan. Color scale indicates relate perfusion signal intensity to average signal intensity
of M0.
87
Other methods, including echo splitting with doubled receiver bandwidth [133], which doubles
the echo spacing, or quadratic phase modulation of refocusing phases [134], which requires long
echo train and sophisticated signal modeling, have been proposed. However, these methods are not
suitable for the current experimental setup because long GRASE readout causes image blurring
attributed to T2 relaxation. The non-CPMG diusion preparation adopted in this study has
been proven to be robust to motion, however, at the cost of half signal loss [135]. Figure 7.5 (c)
shows ASL perfusion images from 18 label/control pairs acquired with the proposed DP-GRASE
sequence. No artifacts were observed and signal stability was improved signicantly. In the
present study, we used a relatively thick slice (8 mm) to compensate for SNR loss.
Alternative non-CPMG approaches could be applied for GRASE in the future when the 2D
CAIPI accelerated GRASE readout and time-constrained TGV reconstruction are successfully
combined. For example, phase errors induced during the diusion encoding could be corrected
from signals acquired with quadratic modulations of excitation and receiver phases. [134]. As
discussed in [136], system equation becomes stationary in a suitable rotating frame (quadratic
phase modulations of excitation/receiver). With suciently long echo train, 'in-phase' and 'out-
of-phase' signals can be represented by two eigenstates of the system corresponding to eigenvalues
of1, and signal magnitude of both components can be directly computed from dierent quadratic
phase modulations without the half signal loss. For the current experiment setup, only 12 echoes
were acquired for each volume which is not sucient to establish the stationary status [136].
With 2D CAIPI accelerated GRASE readout, echo time will be shorter and 20-30 echoes could be
acquired within single acquisition, which potentially can be combined with the quadratic phase
modulation method. However, more researches are needed to investigate how to reconstruct the
under-sampled k-space with phase modulations.
88
7.4 Deep learning for ASL de-noising
Deep learning (DL) is an emerging technique which captures the 'deep' information and nds
hidden connection between signals acquired from multiple modalities without prior knowledge or
modeling. As the in
uence of DL continues to grow, researchers are investigating more and more
new approaches utilizing the DL to improve the MRI image quality to make it more applicable
for clinical use. Due to inherently low SNR, multiple averages are acquired to improve reliability
for ASL. A single-delay ASL scan usually takes more than ve minutes, while multi-delay ASL
scan, which generates more accurate ATT and CBF maps, would typically double the scan time.
Recently, several groups have proposed DL algorithms to improve the quality of ASL images
with signicantly shortened scan time. Convolutional neural network (CNN), the most commonly
used algorithm for ASL denoising, was inspired by biologic processes of the visual cortex, which
receive and respond to signals from overlapped small receptive elds. As compared to statistical
modeling algorithms, CNN is capable of automatically capturing hierarchical information without
specically dened image features.
Figure 7.6 is a schematic illustration of a CNN frame, which was trained to produce high-
quality perfusion images from noisy ASL images acquired from a smaller number of averages
according to [137]. The image features extracted from each CNN layer was passed to the next
layer as an input, which allows the whole network to extract a hierarchy of more complex image
features. Two pathways (local and global) were concatenated after 4 CNN layers to combine the
multiple-scale information and perfusion images were produced after the last convolutional layer.
Kim KH et al, has demonstrated that superior ASL quality can be achieved with a short scan
time using CNN for both healthy subjects and stroke patients. [137]. To further improve the DL
algorithm, CBF quantication model [138] and structural information (gray matter probability)
[139] were incorporated with the CNN frame. All these DL-based algorithms have reported to
89
Figure 7.6: Schematic illustration of a CNN framework for ASL denoising. The rst convolutional
layer (conv) received two or three subtraction images as input, followed by four convolutional
layers in the global (orange arrow) and local (large purple arrow) pathways. Dilation factors
of four convolutional layers in the global pathway were 2, 4, 8 and 16. Parallel pathways were
concatenated at a later step (green arrows). Residual image was acquired by subtracting the
ground truth (GT) from average of input images. Number of lters is shown in italics. The
residual learning approach was used to speed the learning process and improve the performance
of CNNs by simplifying image generation (Figure from Kim K H, et al. Radiology, 2017, 287(2):
658-666.)
achieve promising performance for ASL MRI as compared to simple average in terms of higher
SNR and more accurate CBF quantication with shorter scan time.
DL based denoising algorithms could benet our study by improving the ASL SNR. However,
it is still challenging to apply in our study due to two reasons. First of all, these DL algorithms
haven't been evaluated systematically for aged subjects. ASL signal from aged subjects has even
lower SNR and more severe motion artifacts as compared to younger subjects. Besides, CNN
models need to be trained for specic populations and large training data set is usually required.
90
Figure 7.7: Schematic illustration of a CNN framework for SPA modeling. Inputs are perfusion
images with and without diusion weightings and a structural image. And output are kw and
ATT maps. The CNN works as a 'black box' which computes kw and ATT based on pre-trained
convolutional networks.
The recruiting process in our study is more complicated and thus it's not ecient to recruit
subjects specically for the CNN training purpose.
Instead of denoising, DL could be potentially helpful to improve the SPA modeling reliability.
A novel TGV regularized SPA modeling was proposed in this study to compute ATT and kw from
perfusion signal with and without diusion weightings. Likewise, DL can play a similar role to
nd the hierarchy correlation between ATT/kw and diusion weighted perfusion signals. Figure
7.7 illustrates the basic idea using CNN to compute ATT and kw maps from diusion weighted
ASL signals. Structural image (e.g. T1w/T2w images) or GM/WM masks could also be included
as an additional input to incorporate tissue probability as a prior regularization. As compared to
denoising process, well-dened mathematical model (SPA modeling) could provide a more reliable
basis for training CNN kernels. However, the whole network needs to be carefully designed and
specic parameters (i.e. kernel size, number of layers) need to be optimized in the future.
91
7.5 Animal studies to evaluate kw changes in dierent
disease models
Animal studies could provide crucial insights about the underlying mechanisms related to the BBB
leakage in neurodegeneration conditions. DCE MRI has been commonly used for studying BBB
permeability, and according to a review study published in 2014 [64], 20% of BBB permeability
studies using DCE MRI involves animal experiments with diseases which are known to aect the
BBB function. Although majority of the animal studies were conducted to assess the BBB leakage
in tumors, a few research groups studies the subtle BBB permeability change in aging, cognitive
impairments and AD [64]. In this section, potential experimental disease models, which could
help understand the water permeability change in disease, will be discussed.
7.5.1 BBB disruption in neurological disease
Aryal M.P. and colleagues studied the correlations between Ktrans, Ve and V
D
measured by
DCE-MRI and tumor cellularity. U251 tumor cells were inducted to athymic rats, and sectioned
brain tissues were stained for cell counting and compared with DCE-MRI results [140]. Although
tumor is known to cause BBB disruption, how tumor aects the water transport and vascular
space volume remains poorly understood. With totally disrupted BBB, water can travel across
the BBB freely, which means apparently increased vascular space or 'free-water' volume. As a
result, kw (PS/Vc) may decrease although water exchange freely, which is consistent with previous
study [78].
BBB disruption has also been associated with stroke. Previous studies reported bi-phasic
opening of the BBB following ischemia-reperfusion, however, it remains controversial regarding
the second opening after reperfusion [141]. A longitudinal DCE-MRI study was performed in rats
with intraluminal suture occlusion of the MCA at multiple time-points after reperfusion, and has
conrmed that BBB leakage to contrast agents is continuous following transient focal cerebral
92
Figure 7.8: Typical T2, ADC, CBF, T1, A2, DCE subtraction,Ktrans maps and Evans blue (EB)
slices 2 days after stroke. The ROI denitions for normal (green) and infarcted tissue (red) are
drawn on T2 maps. Scale bar are: T2 (40-120 ms), T1 (1250-2500 ms), ADC (0-0.0001 s/mm
2
),
CBF (0-1.5 ml/g/min), A2 (0.4-1.4), DCE subtraction (0-25,000 signal unit), KTrans (0-0.003
min
1
). (Figure from Tiwari Y V, et al. JCBFM, 2017, 37(8): 2706-2715.)
ischemia [142]. Tiwari YV and colleagus have recently cross-validated the water exchange rate
and Evans blue histology in the rats with experimental ischemic stroke using mannitol, which
induces disruption of the BBB [80, 81]. Figure 7.8 shows comparison results acquired two days
after stroke: A2 (tissue fraction of DW-ASL signal),Ktrans and Evans blue histology. Signicant
lower kw (or A2) was found in the ischemic core as compared to the contralesional normal tissue,
which is also consistent with the Evans blue histology. With experimental induction of tumor
or stroke, animal studies could help us have a clear view of vasculature changes and how water
transport mechanisms are aected when BBB is total disrupted.
7.5.2 Transient BBB opening using focused ultrasound (FUS) and
microbubbles
FUS in the presence of microbubbles has been experimentally established as an eective non-
invasive method for transient and localized BBB opening. Since BBB blocks large molecules from
93
entering the brain under normal physiological conditions, BBB opening after the application of
FUS has shown promising prospect for drug delivery researches [3]. FUS induced BBB opening
was quantitatively evaluated by Ktrans mesured from a DCE-MRI study [143]. FUS dierent
acoustic pressures and microbubble sizes was applied in the murine hippocampus. The volume
of the BBB opening was assess by Ktrans, and was found to be proportional to both acoustic
pressures and diameter of microbubbles before a plateau was reached [143].
FUS induces localized, transient and controllable BBB opening by adjusting the acoustic
pressures or the microbubble sizes. Since water has much smaller molecule weights as compared to
commonly used contrast agents, FUS would be an indispensable tool to study the water exchange
mechanisms at dierent level of the BBB opening and help us have a better understanding about
the water permeability change with subtle BBB opening in aging or neurological diseases.
7.5.3 Pericyte-decient and AQP4-decient model
Pericyte controls microvascular functions including BBB permeability and CBF. Pericytes loss
usually causes BBB breakdown, microvascular reductions and WM disease. Although loss of
pericytes has been associated with cognitive impairment [42, 144], the underlying mechanism
about how pericytes maintain the BBB permeability in CNS diseases is not well understood.
Montagne A and colleagues have studied microcirculatory changes in a pericyte-decient mice
model (with genetic mutations in PDGFR) [55]. By conducting MRI experiments in pericyte-
decient mice, they found that pericyte degeneration triggers disruption of WM microcirculation
and leads to accumulation of toxic deposits and reduced blood
ow. Another approach to directly
assess water permeability across the BBB is using AQP4-decient animal models. Ohene Y and
colleagues proposed to measure water exchange time across the BBB in AQP4-decient mice
using a multi-echo ASL sequence [21]. They found that water exchange time is 31% longer in
AQP4-decient mice as compared to the wild-type control groups. Both pericyte-decient and
AQP4-decient models could be valuable tools to evaluate the sensitivity of our technique to the
94
water permeability change and have a better understanding about the role of water transport in
SVD or other neurodegenerative conditions.
7.6 Conclusion
A DP 3D GRASE pCASL sequence with TGV regularized SPA modeling was proposed to measure
BBB water permeability noninvasively with good reproducibility in a cohort of aged subjects at
risk of SVD. This study demonstrated the capability of kw being a surrogate imaging biomarker
for SVD and early dementia. Its clinical use for detection of BBB dysfunction before leakage of
large-molecule contrast agents awaits further evaluation.
Future studies will focus on improvement of the spatial resolution and SNR of the ASL signals
by developing a new 3D sequence with 6 fold acceleration and time dependent 2D CAIPI under-
sampling pattern. A new temporal regularized TGV reconstruction algorithm could be applied for
this accelerated acquisition, which maximize the temporal incoherence between label and control
pairs. Reliability of SPA modeling could also be improved by utilizing DL based algorithms, but
CNN networks and specic parameters need to be carefully designed. We also hypothesized a
bi-directional change of water permeability at dierent stages of BBB opening. However, animal
studies need to be conducted in the future to have a better understanding about both permeability
and underlying vasculature changes in neupathological diseases.
95
Reference List
[1] N Joan Abbott, Adjanie AK Patabendige, Diana EM Dolman, Siti R Yusof, and David J
Begley. Structure and function of the blood{brain barrier. Neurobiology of disease, 37(1):13{
25, 2010.
[2] BV Zlokovic, DS Skundric, MB Segal, MN Lipovac, JB Mackic, and H Davson. A saturable
mechanism for transport of immunoglobulin g across the blood-brain barrier of the guinea
pig. Experimental neurology, 107(3):263{270, 1990.
[3] William M Pardridge. The blood-brain barrier: bottleneck in brain drug development.
NeuroRx, 2(1):3{14, 2005.
[4] Berislav V Zlokovic. Neurovascular pathways to neurodegeneration in alzheimer's disease
and other disorders. Nature Reviews Neuroscience, 12(12):723, 2011.
[5] N Joan Abbott, Lars R onnb ack, and Elisabeth Hansson. Astrocyte{endothelial interactions
at the blood{brain barrier. Nature reviews neuroscience, 7(1):41, 2006.
[6] Kuniaki Takata, Toshiyuki Matsuzaki, and Yuki Tajika. Aquaporins: water channel proteins
of the cell membrane. Progress in histochemistry and cytochemistry, 39(1):1{83, 2004.
[7] Bonomini Francesca and Rita Rezzani. Aquaporin and blood brain barrier. Current Neu-
ropharmacology, 8(2):92, 2010.
[8] Soren Nielsen, Barbara L Smith, Erik Ils Christensen, and Peter Agre. Distribution of the
aquaporin chip in secretory and resorptive epithelia and capillary endothelia. Proceedings
of the National Academy of Sciences, 90(15):7275{7279, 1993.
[9] S Saadoun, MC Papadopoulos, DC Davies, S Krishna, and BA Bell. Aquaporin-4 expres-
sion is increased in oedematous human brain tumours. J Neurol Neurosurg Psychiatry,
72(2):262{265, 2002.
[10] N Joan Abbott. Astrocyte{endothelial interactions and blood{brain barrier permeability.
Journal of anatomy, 200(6):629{638, 2002.
[11] Stefan Liebner, Arne Fischmann, Gesa Rascher, Frank Duner, E-H Grote, Hubert
Kalbacher, and Hartwig Wolburg. Claudin-1 and claudin-5 expression and tight junction
morphology are altered in blood vessels of human glioblastoma multiforme. Acta neu-
ropathologica, 100(3):323{331, 2000.
[12] M Simard. Arcuino g, takano t, liu qs, and nedergaard m. Signaling at the gliovascular
interface. J Neurosci, 23:9254{9262, 2003.
[13] Jing Yang, Lisa K Lunde, Paworn Nuntagij, Tomohiro Oguchi, Laura Camassa, Lars NG
Nilsson, Lars Lannfelt, Yuming Xu, Mahmood Amiry-Moghaddam, Ole Petter Ottersen,
et al. Loss of astrocyte polarization in the tg-arcswe mouse model of alzheimer's disease.
Journal of Alzheimer's Disease, 27(4):711{722, 2011.
96
[14] Arne Warth, Stephan Kr oger, and Hartwig Wolburg. Redistribution of aquaporin-4 in
human glioblastoma correlates with loss of agrin immunoreactivity from brain capillary
basal laminae. Acta neuropathologica, 107(4):311{318, 2004.
[15] Eugen Solenov, Hiroyuki Watanabe, Georey T Manley, and AS Verkman. Sevenfold-
reduced osmotic water permeability in primary astrocyte cultures from aqp-4-decient mice,
measured by a
uorescence quenching method. American Journal of Physiology-Cell Phys-
iology, 286(2):C426{C432, 2004.
[16] Orin Bloch and Georey T Manley. The role of aquaporin-4 in cerebral water transport
and edema. Neurosurgical focus, 22(5):1{7, 2007.
[17] M Tom as-Camardiel, JL Venero, AJ Herrera, RM De Pablos, Jos e Antonio Pintor-Toro,
A Machado, and J Cano. Blood{brain barrier disruption highly induces aquaporin-4 mrna
and protein in perivascular and parenchymal astrocytes: protective eect by estradiol treat-
ment in ovariectomized animals. Journal of neuroscience research, 80(2):235{246, 2005.
[18] M Amiry-Moghaddam, DS Frydenlund, and OP Ottersen. Anchoring of aquaporin-4 in
brain: molecular mechanisms and implications for the physiology and pathophysiology of
water transport. Neuroscience, 129(4):997{1008, 2004.
[19] Mahmood Amiry-Moghaddam, Anne Williamson, Maria Palomba, Tore Eid, Nihal C
De Lanerolle, Erlend A Nagelhus, Marvin E Adams, Stanley C Froehner, Peter Agre, and
Ole P Ottersen. Delayed k+ clearance associated with aquaporin-4 mislocalization: pheno-
typic defects in brains of -syntrophin-null mice. Proceedings of the National Academy of
Sciences, 100(23):13615{13620, 2003.
[20] Georey T Manley, Miki Fujimura, Tonghui Ma, Nobuo Noshita, Ferda Filiz, Andrew W
Bollen, Pak Chan, and AS Verkman. Aquaporin-4 deletion in mice reduces brain edema
after acute water intoxication and ischemic stroke. Nature medicine, 6(2):159, 2000.
[21] Yolanda Ohene, Ian F Harrison, Payam Nahavandi, Ozama Ismail, Eleanor V Bird, Ole P
Ottersen, Erlend A Nagelhus, David L Thomas, Mark F Lythgoe, and Jack A Wells. Non-
invasive mri of brain clearance pathways using multiple echo time arterial spin labelling: an
aquaporin-4 study. NeuroImage, 188:515{523, 2019.
[22] Jerey J Ili, Minghuan Wang, Yonghong Liao, Benjamin A Plogg, Weiguo Peng, Georg A
Gundersen, Helene Benveniste, G Edward Vates, Rashid Deane, Steven A Goldman, et al.
A paravascular pathway facilitates csf
ow through the brain parenchyma and the clearance
of interstitial solutes, including amyloid. Science translational medicine, 4(147):147ra111{
147ra111, 2012.
[23] Benjamin T Kress, Jerey J Ili, Maosheng Xia, Minghuan Wang, Helen S Wei, Douglas
Zeppenfeld, Lulu Xie, Hongyi Kang, Qiwu Xu, Jason A Liew, et al. Impairment of paravas-
cular clearance pathways in the aging brain. Annals of neurology, 76(6):845{861, 2014.
[24] Leonardo Pantoni. Cerebral small vessel disease: from pathogenesis and clinical character-
istics to therapeutic challenges. The Lancet Neurology, 9(7):689{701, 2010.
[25] Costantino Iadecola. Vascular and metabolic factors in alzheimers disease and related de-
mentias: introduction, 2016.
[26] Julie A Schneider, Zoe Arvanitakis, Woojeong Bang, and David A Bennett. Mixed brain
pathologies account for most dementia cases in community-dwelling older persons. Neurol-
ogy, 69(24):2197{2204, 2007.
97
[27] Bo Norrving. Lacunar infarcts: no black holes in the brain are benign. Practical neurology,
8(4):222{228, 2008.
[28] Philip B Gorelick, Angelo Scuteri, Sandra E Black, Charles DeCarli, Steven M Greenberg,
Costantino Iadecola, Lenore J Launer, Stephane Laurent, Oscar L Lopez, David Nyenhuis,
et al. Vascular contributions to cognitive impairment and dementia: a statement for health-
care professionals from the american heart association/american stroke association. Stroke,
42(9):2672{2713, 2011.
[29] Martin Dichgans and Didier Leys. Vascular cognitive impairment. Circulation research,
120(3):573{591, 2017.
[30] Vladimir C Hachinski, Niels A Lassen, and John Marshall. Multi-infarct dementia: a cause
of mental deterioration in the elderly. The Lancet, 304(7874):207{209, 1974.
[31] Gary A Rosenberg, Anders Wallin, Joanna M Wardlaw, Hugh S Markus, Joan Montaner,
Leslie Wolfson, Costantino Iadecola, Berislav V Zlokovic, Anne Joutel, Martin Dichgans,
et al. Consensus statement for diagnosis of subcortical small vessel disease. Journal of
Cerebral Blood Flow & Metabolism, 36(1):6{25, 2016.
[32] Joanna M Wardlaw, Fergus Doubal, Paul Armitage, Francesca Chappell, Trevor Carpen-
ter, Susana Mu~ noz Maniega, Andrew Farrall, Cathie Sudlow, Martin Dennis, and Baljean
Dhillon. Lacunar stroke is associated with diuse blood{brain barrier dysfunction. An-
nals of Neurology: Ocial Journal of the American Neurological Association and the Child
Neurology Society, 65(2):194{202, 2009.
[33] Yulu Shi and Joanna M Wardlaw. Update on cerebral small vessel disease: a dynamic
whole-brain disease. Stroke and vascular neurology, 1(3):83{92, 2016.
[34] Adriane Mayda and Charles DeCarli. Vascular cognitive impairment: prodrome to vad.
Vascular cognitive impairment in clinical practice, page 11, 2009.
[35] Monique MB Breteler, Nel M van Amerongen, John C van Swieten, Jules J Claus, Dieder-
ick E Grobbee, Jan van Gijn, Albert Hofman, and Frans van Harskamp. Cognitive correlates
of ventricular enlargement and cerebral white matter lesions on magnetic resonance imaging.
the rotterdam study. Stroke, 25(6):1109{1115, 1994.
[36] St ephanie Debette, Alexa Beiser, Charles DeCarli, Rhoda Au, Jayandra J Himali, Margaret
Kelly-Hayes, Jose R Romero, Carlos S Kase, Philip A Wolf, and Sudha Seshadri. Association
of mri markers of vascular brain injury with incident stroke, mild cognitive impairment,
dementia, and mortality: the framingham ospring study. Stroke, 41(4):600{606, 2010.
[37] Charles DeCarli, DGM Murphy, M Tranh, CL Grady, JV Haxby, JA Gillette, JA Salerno,
A Gonzales-Aviles, B Honvitz, SI Rapoport, et al. The eect of white matter hyperintensity
volume on brain structure, cognitive performance, and cerebral metabolism of glucose in 51
healthy adults. Neurology, 45(11):2077{2084, 1995.
[38] Seonjoo Lee, Fawad Viqar, Molly E Zimmerman, Atul Narkhede, Giuseppe Tosto, Tam-
mie LS Benzinger, Daniel S Marcus, Anne M Fagan, Alison Goate, Nick C Fox, et al.
White matter hyperintensities are a core feature of alzheimer's disease: evidence from the
dominantly inherited alzheimer network. Annals of neurology, 79(6):929{939, 2016.
[39] Lisa C Silbert, Diane B Howieson, Hiroko Dodge, and Jerey A Kaye. Cognitive impairment
risk white matter hyperintensity progression matters. Neurology, 73(2):120{125, 2009.
98
[40] JM Wardlaw, PAG Sandercock, MS Dennis, and J Starr. Is breakdown of the blood-brain
barrier responsible for lacunar stroke, leukoaraiosis, and dementia? Stroke, 34(3):806{812,
2003.
[41] Andrew J Farrall and Joanna M Wardlaw. Blood{brain barrier: ageing and microvascular
disease{systematic review and meta-analysis. Neurobiology of aging, 30(3):337{352, 2009.
[42] Axel Montagne, Samuel R Barnes, Melanie D Sweeney, Matthew R Halliday, Abhay P
Sagare, Zhen Zhao, Arthur W Toga, Russell E Jacobs, Collin Y Liu, Lilyana Amezcua, et al.
Blood-brain barrier breakdown in the aging human hippocampus. Neuron, 85(2):296{302,
2015.
[43] TL Kemper, GJ Blatt, RJ Killiany, and MB Moss. Neuropathology of progressive cognitive
decline in chronically hypertensive rhesus monkeys. Acta neuropathologica, 101(2):145{153,
2001.
[44] Geert Jan Biessels and Yael D Reijmer. Brain changes underlying cognitive dysfunction in
diabetes: what can we learn from mri? Diabetes, 63(7):2244{2252, 2014.
[45] Barbro B Johansson. Hypertension mechanisms causing stroke. Clinical and experimental
pharmacology and physiology, 26(7):563{565, 1999.
[46] Diane Armao, Mario Kornfeld, Edward Y Estrada, Mark Grossetete, and Gary A Rosen-
berg. Neutral proteases and disruption of the blood{brain barrier in rat. Brain research,
767(2):259{264, 1997.
[47] Clive Ballard, Fiona Shaw, Ian McKeith, and Roseanne Kenny. High prevalence of neu-
rovascular instability in neurodegenerative dementias. Neurology, 51(6):1760{1762, 1998.
[48] Hidekazu Tomimoto. Subcortical vascular dementia. Neuroscience research, 71(3):193{199,
2011.
[49] Anders Wallin, Kaj Blennow, P Fredman, CG Gottfries, I Karlsson, and L Svenner-
holm. Blood brain barrier function in vascular dementia. Acta neurologica Scandinavica,
81(4):318{322, 1990.
[50] I Skoog, A Wallin, P Fredman, C Hesse, O Aevarsson, I Karlsson, CG Gottfries, and
K Blennow. A population study on blood-brain barrier function in 85-year-olds: relation to
alzheimer's disease and vascular dementia. Neurology, 50(4):966{971, 1998.
[51] JM Starr, J Wardlaw, K Ferguson, A MacLullich, IJ Deary, and I Marshall. Increased
blood{brain barrier permeability in type ii diabetes demonstrated by gadolinium magnetic
resonance imaging. Journal of Neurology, Neurosurgery & Psychiatry, 74(1):70{76, 2003.
[52] Ichiro Akiguchi, Hidekazu Tomimoto, Toshihiko Suenaga, Hideaki Wakita, and Herbert
Budka. Blood-brain barrier dysfunction in binswangers disease; an immunohistochemical
study. Acta neuropathologica, 95(1):78{84, 1997.
[53] Min-Ying Su, Elizabeth Head, William M Brooks, Zhiheng Wang, Bruce A Muggenburg,
Gina E Adam, Robert Sutherland, Carl W Cotman, and Orhan Nalcioglu. Magnetic reso-
nance imaging of anatomic and vascular characteristics in a canine model of human aging.
Neurobiology of aging, 19(5):479{485, 1998.
[54] Annika Armulik, Guillem Genov e, Maarja M ae, Maya H Nisancioglu, Elisabet Wallgard,
Colin Niaudet, Liqun He, Jenny Norlin, Per Lindblom, Karin Strittmatter, et al. Pericytes
regulate the blood{brain barrier. Nature, 468(7323):557, 2010.
99
[55] Axel Montagne, Angeliki M Nikolakopoulou, Zhen Zhao, Abhay P Sagare, Gabriel Si, Divna
Lazic, Samuel R Barnes, Madelaine Daianu, Anita Ramanathan, Ariel Go, et al. Pericyte
degeneration causes white matter dysfunction in the mouse central nervous system. Nature
medicine, 24(3):326, 2018.
[56] Yi Guo, R Marc Lebel, Yinghua Zhu, Sajan Goud Lingala, Mark S Shiroishi, Meng Law,
and Krishna Nayak. High-resolution whole-brain dce-mri using constrained reconstruction:
Prospective clinical evaluation in brain tumor patients. Medical physics, 43(5):2013{2023,
2016.
[57] Zamir Merali, Kun Huang, David Mikulis, Frank Silver, and Andrea Kassner. Evolution
of blood-brain-barrier permeability after acute ischemic stroke. PloS one, 12(2):e0171558,
2017.
[58] Nicole E Lopez, Michael J Krzyzaniak, Chelsea Blow, James Putnam, Yan Ortiz-Pomales,
Ann-Marie Hageny, Brian Eliceiri, Raul Coimbra, and Vishal Bansal. Ghrelin prevents
disruption of the blood{brain barrier after traumatic brain injury. Journal of neurotrauma,
29(2):385{393, 2012.
[59] Stacey Watkins, Stefanie Robel, Ian F Kimbrough, Stephanie M Robert, Graham Ellis-
Davies, and Harald Sontheimer. Disruption of astrocyte{vascular coupling and the blood{
brain barrier by invading glioma cells. Nature communications, 5:4196, 2014.
[60] Birgitte Fuglsang Kjlby, Leif stergaard, and Valerij G Kiselev. Theoretical model of
intravascular paramagnetic tracers eect on tissue relaxation. Magnetic Resonance in
Medicine: An Ocial Journal of the International Society for Magnetic Resonance in
Medicine, 56(1):187{197, 2006.
[61] JL Boxerman, KM Schmainda, and RM Weissko. Relative cerebral blood volume maps
corrected for contrast agent extravasation signicantly correlate with glioma tumor grade,
whereas uncorrected maps do not. American Journal of Neuroradiology, 27(4):859{867,
2006.
[62] Eric M Thompson, Daniel J Guillaume, Edit D osa, Xin Li, Kellie J Nazemi, Seymur Gahra-
manov, Bronwyn E Hamilton, and Edward A Neuwelt. Dual contrast perfusion mri in a
single imaging session for assessment of pediatric brain tumors. Journal of neuro-oncology,
109(1):105{114, 2012.
[63] Axel Montagne, Arthur W Toga, and Berislav V Zlokovic. Blood-brain barrier permeability
and gadolinium: benets and potential pitfalls in research. JAMA neurology, 73(1):13{14,
2016.
[64] Anna K Heye, Ross D Culling, Maria del C Vald es Hern andez, Michael J Thrippleton, and
Joanna M Wardlaw. Assessment of blood{brain barrier disruption using dynamic contrast-
enhanced mri. a systematic review. NeuroImage: Clinical, 6:262{274, 2014.
[65] Yi Guo, Sajan Goud Lingala, Yinghua Zhu, R Marc Lebel, and Krishna S Nayak. Di-
rect estimation of tracer-kinetic parameter maps from highly undersampled brain dynamic
contrast enhanced mri. Magnetic resonance in medicine, 78(4):1566{1578, 2017.
[66] Paul S Tofts, Gunnar Brix, David L Buckley, Jerey L Evelhoch, Elizabeth Henderson,
Michael V Knopp, Henrik BW Larsson, Ting-Yim Lee, Nina A Mayr, Georey JM Parker,
et al. Estimating kinetic parameters from dynamic contrast-enhanced t1-weighted mri
of a diusable tracer: standardized quantities and symbols. Journal of Magnetic Reso-
nance Imaging: An Ocial Journal of the International Society for Magnetic Resonance in
Medicine, 10(3):223{232, 1999.
100
[67] Anna K Heye, Michael J Thrippleton, Paul A Armitage, Maria del C Vald es Hern andez,
Stephen D Makin, Andreas Glatz, Eleni Sakka, and Joanna M Wardlaw. Tracer kinetic
modelling for dce-mri quantication of subtle blood{brain barrier permeability. Neuroimage,
125:446{455, 2016.
[68] Yi Guo, Sajan Goud Lingala, Yannick Bliesener, R Marc Lebel, Yinghua Zhu, and Krishna S
Nayak. Joint arterial input function and tracer kinetic parameter estimation from under-
sampled dynamic contrast-enhanced mri using a model consistency constraint. Magnetic
resonance in medicine, 79(5):2804{2815, 2018.
[69] Vikas Gulani, Fernando Calamante, Frank G Shellock, Emanuel Kanal, Scott B Reeder,
et al. Gadolinium deposition in the brain: summary of evidence and recommendations.
The Lancet Neurology, 16(7):564{570, 2017.
[70] Samuel R Barnes, Thomas SC Ng, Axel Montagne, Meng Law, Berislav V Zlokovic, and
Russell E Jacobs. Optimal acquisition and modeling parameters for accurate assessment of
low ktrans blood{brain barrier permeability using dynamic contrast-enhanced mri. Mag-
netic resonance in medicine, 75(5):1967{1977, 2016.
[71] LARS-OLOF WAHLUND and Lena Bronge. Contrast-enhanced mri of white matter lesions
in patients with blood-brain barrier dysfunction. Annals of the New York Academy of
Sciences, 903(1):477{481, 2000.
[72] Ben R Dickie, Matthias Vandesquille, Jos e Ulloa, Herv e Boutin, Laura M Parkes, and
Geo JM Parker. Water-exchange mri detects subtle blood-brain barrier breakdown in
alzheimer's disease rats. NeuroImage, 184:349{358, 2019.
[73] Zixuan Lin, Yang Li, Pan Su, Deng Mao, Zhiliang Wei, Jay J Pillai, Abhay Moghekar,
Matthias van Osch, Yulin Ge, and Hanzhang Lu. Non-contrast mr imaging of blood-brain
barrier permeability to water. Magnetic resonance in medicine, 2018.
[74] Hanzhang Lu, Chekesha Clingman, Xavier Golay, and Peter CM Van Zijl. Determining the
longitudinal relaxation time (t1) of blood at 3.0 tesla. Magnetic Resonance in Medicine: An
Ocial Journal of the International Society for Magnetic Resonance in Medicine, 52(3):679{
682, 2004.
[75] Johannes Gregori, Norbert Schu, Rolf Kern, and Matthias G unther. T2-based arterial
spin labeling measurements of blood to tissue water transfer in human brain. Journal of
Magnetic Resonance Imaging, 37(2):332{342, 2013.
[76] Xiang He, Marcus E Raichle, and Dmitriy A Yablonskiy. Transmembrane dynamics of water
exchange in human brain. Magnetic resonance in medicine, 67(2):562{571, 2012.
[77] Kenneth Wengler, Lev Bangiyev, Turhan Canli, Tim Q Duong, Mark E Schweitzer, and
Xiang He. 3d mri of whole-brain water permeability with intrinsic diusivity encoding of
arterial labeled spin (ideals). NeuroImage, 2019.
[78] Jiongjiong Wang, Mar a A Fern andez-Seara, Sumei Wang, and Keith S St Lawrence. When
perfusion meets diusion: in vivo measurement of water permeability in human brain.
Journal of Cerebral Blood Flow & Metabolism, 27(4):839{849, 2007.
[79] Keith S St. Lawrence, Daron Owen, and Danny JJ Wang. A two-stage approach for measur-
ing vascular water exchange and arterial transit time by diusion-weighted perfusion mri.
Magnetic Resonance in Medicine, 67(5):1275{1284, 2012.
101
[80] YV Tiwar, Q Shen, Jiang Wang, Li Z, Long W, Fang J, Duong C, and T. Measuring blood-
brain-barrier permeability using diusion-weighted arterial spin labeling (dw-asl): Corrob-
oration with ktrans and evans blue measurements. In Proc Int Soc Magn Reson Med, page
0792, 2015.
[81] Yash V Tiwari, Jianfei Lu, Qiang Shen, Bianca Cerqueira, and Timothy Q Duong. Magnetic
resonance imaging of blood{brain barrier permeability in ischemic stroke using diusion-
weighted arterial spin labeling in rats. Journal of Cerebral Blood Flow & Metabolism,
37(8):2706{2715, 2017.
[82] Jose A Palomares, Sudhakar Tummala, Danny JJ Wang, Bumhee Park, Mary A Woo,
Daniel W Kang, Keith S St Lawrence, Ronald M Harper, and Rajesh Kumar. Water ex-
change across the blood-brain barrier in obstructive sleep apnea: An mri diusion-weighted
pseudo-continuous arterial spin labeling study. Journal of Neuroimaging, 25(6):900{905,
2015.
[83] Wikipedia contributors. Microcirculation | Wikipedia, the free encyclopedia, 2018. [On-
line; accessed 22-February-2019].
[84] Denis Le Bihan, Eric Breton, Denis Lallemand, Philippe Grenier, Emmanuel Cabanis, and
Maurice Laval-Jeantet. Mr imaging of intravoxel incoherent motions: application to diu-
sion and perfusion in neurologic disorders. Radiology, 161(2):401{407, 1986.
[85] J Conklin, C Heyn, M Roux, M Cerny, M Wintermark, and C Federau. A simplied
model for intravoxel incoherent motion perfusion imaging of the brain. American Journal
of Neuroradiology, 37(12):2251{2257, 2016.
[86] Denis Le Bihan. What can we see with ivim mri? NeuroImage, 2017.
[87] Xingfeng Shao, Yi Wang, Steen Moeller, and Danny JJ Wang. A constrained slice-dependent
background suppression scheme for simultaneous multislice pseudo-continuous arterial spin
labeling. Magnetic resonance in medicine, 79(1):394{400, 2018.
[88] Xingfeng Shao, Dapeng Liu, Thomas Martin, Teresa Chanlaw, Sherin U Devaskar, Carla
Janzen, Aisling M Murphy, Daniel Margolis, Kyunghyun Sung, and Danny JJ Wang. Mea-
suring human placental blood
ow with multidelay 3d grase pseudocontinuous arterial spin
labeling at 3t. Journal of Magnetic Resonance Imaging, 47(6):1667{1676, 2018.
[89] Jiongjiong Wang, Hengyi Rao, Gabriel S Wetmore, Patricia M Furlan, Marc Korczykowski,
David F Dinges, and John A Detre. Perfusion functional mri reveals cerebral blood
ow
pattern under psychological stress. Proceedings of the National Academy of Sciences,
102(49):17804{17809, 2005.
[90] John A Detre, John S Leigh, Donald S Williams, and Alan P Koretsky. Perfusion imaging.
Magnetic resonance in medicine, 23(1):37{45, 1992.
[91] KS St. Lawrence, JA Frank, and AC McLaughlin. Eect of restricted water exchange on
cerebral blood
ow values calculated with arterial spin tagging: a theoretical investiga-
tion. Magnetic Resonance in Medicine: An Ocial Journal of the International Society for
Magnetic Resonance in Medicine, 44(3):440{449, 2000.
[92] Peter Herscovitch, Marcus E Raichle, Michael R Kilbourn, and Michael J Welch. Positron
emission tomographic measurement of cerebral blood
ow and permeabilitysurface area
product of water using [15o] water and [11c] butanol, 1987.
102
[93] Timothy G Reese, O Heid, RM Weissko, and VJ Wedeen. Reduction of eddy-current-
induced distortion in diusion mri using a twice-refocused spin echo. Magnetic Resonance
in Medicine: An Ocial Journal of the International Society for Magnetic Resonance in
Medicine, 49(1):177{182, 2003.
[94] Xingfeng Shao, M Dylan Tisdall, Danny JJ Wang, and Andre Jan Willem van der Kouwe.
Prospective motion correction for 3d grase pcasl with volumetric navigators. In Proceedings
of the International Society for Magnetic Resonance in Medicine... Scientic Meeting and
Exhibition. International Society for Magnetic Resonance in Medicine. Scientic Meeting
and Exhibition, volume 25, page 0680. NIH Public Access, 2017.
[95] X Shao and DJ Wang. Single shot high resolution 3d arterial spin labeling using 2d caipi
and espirit reconstruction. In Proc Int Soc Magn Reson Med, volume 25, page 3629, 2017.
[96] Eugene M Renkin. Transport of potassium-42 from blood to tissue in isolated mammalian
skeletal muscles. American Journal of Physiology-Legacy Content, 197(6):1205{1210, 1959.
[97] Christian Crone. The permeability of capillaries in various organs as determined by use of
the indicator diusionmethod. Acta physiologica scandinavica, 58(4):292{305, 1963.
[98] Jiongjiong Wang, Georey K Aguirre, Daniel Y Kimberg, Anne C Roc, Lin Li, and John A
Detre. Arterial spin labeling perfusion fmri with very low task frequency. Magnetic Reso-
nance in Medicine: An Ocial Journal of the International Society for Magnetic Resonance
in Medicine, 49(5):796{802, 2003.
[99] Florian Knoll, Kristian Bredies, Thomas Pock, and Rudolf Stollberger. Second order total
generalized variation (tgv) for mri. Magnetic resonance in medicine, 65(2):480{491, 2011.
[100] Lirong Yan, Collin Y Liu, Koon-Pong Wong, Sung-Cheng Huang, Wendy J Mack, Kay
Jann, Giovanni Coppola, John M Ringman, and Danny JJ Wang. Regional association of
pcasl-mri with fdg-pet and pib-pet in people at risk for autosomal dominant alzheimer's
disease. NeuroImage: Clinical, 17:751{760, 2018.
[101] Stefan M Spann, Kamil S Kazimierski, Christoph S Aigner, Markus Kraiger, Kristian
Bredies, and Rudolf Stollberger. Spatio-temporal tgv denoising for asl perfusion imaging.
Neuroimage, 157:81{96, 2017.
[102] Berislav V Zlokovic. Neurovascular mechanisms of alzheimer's neurodegeneration. Trends
in neurosciences, 28(4):202{208, 2005.
[103] GL Bowman, JA Kaye, M Moore, D Waichunas, NE Carlson, and JF Quinn. Blood{brain
barrier impairment in alzheimer disease stability and functional signicance. Neurology,
68(21):1809{1814, 2007.
[104] Paul A Yushkevich, Joseph Piven, Heather Cody Hazlett, Rachel Gimpel Smith, Sean Ho,
James C Gee, and Guido Gerig. User-guided 3d active contour segmentation of anatomical
structures: signicantly improved eciency and reliability. Neuroimage, 31(3):1116{1128,
2006.
[105] Franz Fazekas, John B Chawluk, Abass Alavi, Howard I Hurtig, and Robert A Zimmerman.
Mr signal abnormalities at 1.5 t in alzheimer's dementia and normal aging. American journal
of roentgenology, 149(2):351{356, 1987.
[106] Nimish K Acharya, Eli C Levin, Peter M Cliord, Min Han, Ryan Tourtellotte, Dean Cham-
berlain, Michael Pollaro, Nicholas J Coretti, Mary C Kosciuk, Eric P Nagele, et al. Diabetes
103
and hypercholesterolemia increase blood-brain barrier permeability and brain amyloid depo-
sition: benecial eects of the lppla2 inhibitor darapladib. Journal of Alzheimer's Disease,
35(1):179{198, 2013.
[107] S Ahtiluoto, T Polvikoski, M Peltonen, A Solomon, J Tuomilehto, B Winblad, R Sulkava,
and M Kivipelto. Diabetes, alzheimer disease, and vascular dementia a population-based
neuropathologic study. Neurology, 75(13):1195{1202, 2010.
[108] A Solomon, I K areholt, T Ngandu, B Winblad, A Nissinen, J Tuomilehto, H Soininen, and
M Kivipelto. Serum cholesterol changes after midlife and late-life cognition: twenty-one-year
follow-up study. Neurology, 68(10):751{756, 2007.
[109] Nicolas Pelisch, Naohisa Hosomi, Hirohito Mori, Tsutomu Masaki, and Akira Nishiyama.
Ras inhibition attenuates cognitive impairment by reducing blood-brain barrier permeability
in hypertensive subjects. Current hypertension reviews, 9(2):93{98, 2013.
[110] Martin Dichgans and Vera Zietemann. Prevention of vascular cognitive impairment. Stroke,
43(11):3137{3146, 2012.
[111] Krishna K Gaddam, Mari K Nishizaka, Monique N Pratt-Ubunama, Eduardo Pimenta,
Inmaculada Aban, Suzanne Oparil, and David A Calhoun. Characterization of resistant
hypertension: association between resistant hypertension, aldosterone, and persistent in-
travascular volume expansion. Archives of internal medicine, 168(11):1159{1164, 2008.
[112] E Duron and Olivier Hanon. Vascular risk factors, cognitve decline, and dementia. Vascular
health and risk management, 4(2):363, 2008.
[113] Charles P Hughes, Leonard Berg, Warren Danziger, Lawrence A Coben, and Ronald L
Martin. A new clinical scale for the staging of dementia. The British journal of psychiatry,
140(6):566{572, 1982.
[114] Sid E OBryant, Stephen C Waring, C Munro Cullum, James Hall, Laura Lacritz, Paul J
Massman, Philip J Lupo, Joan S Reisch, and Rachelle Doody. Staging dementia using
clinical dementia rating scale sum of boxes scores: a texas alzheimer's research consortium
study. Archives of neurology, 65(8):1091{1095, 2008.
[115] Wikipedia contributors. Montreal cognitive assessment | Wikipedia, the free encyclopedia,
2019. [Online; accessed 22-February-2019].
[116] F Fazekas, R Kleinert, H Oenbacher, Rea Schmidt, G Kleinert, F Payer, H Radner, and
H Lechner. Pathologic correlates of incidental mri white matter signal hyperintensities.
Neurology, 43(9):1683{1683, 1993.
[117] Yulu Shi, Michael J Thrippleton, Stephen D Makin, Ian Marshall, Mirjam I Geerlings,
Anton JM de Craen, Mark A van Buchem, and Joanna M Wardlaw. Cerebral blood
ow
in small vessel disease: a systematic review and meta-analysis. Journal of Cerebral Blood
Flow & Metabolism, 36(10):1653{1667, 2016.
[118] Berislav V Zlokovic. The blood-brain barrier in health and chronic neurodegenerative dis-
orders. Neuron, 57(2):178{201, 2008.
[119] Huali Wang, Edward J Golob, and Min-Ying Su. Vascular volume and blood-brain barrier
permeability measured by dynamic contrast enhanced mri in hippocampus and cerebellum of
patients with mci and normal controls. Journal of Magnetic Resonance Imaging: An Ocial
Journal of the International Society for Magnetic Resonance in Medicine, 24(3):695{700,
2006.
104
[120] Samuel R Barnes, Thomas SC Ng, Naomi Santa-Maria, Axel Montagne, Berislav V Zlokovic,
and Russell E Jacobs. Rocketship: a
exible and modular software tool for the planning,
processing and analysis of dynamic mri studies. BMC medical imaging, 15(1):19, 2015.
[121] Leslie R Bridges, Joycelyn Andoh, Andrew J Lawrence, Cheryl HL Khoong, Wayne W
Poon, Margaret M Esiri, Hugh S Markus, and Atticus H Hainsworth. Blood-brain barrier
dysfunction and cerebral small vessel disease (arteriolosclerosis) in brains of older people.
Journal of Neuropathology & Experimental Neurology, 73(11):1026{1033, 2014.
[122] Tomonori Kanda, Kazunari Ishii, Hiroki Kawaguchi, Kazuhiro Kitajima, and Daisuke Tak-
enaka. High signal intensity in the dentate nucleus and globus pallidus on unenhanced
t1-weighted mr images: relationship with increasing cumulative dose of a gadolinium-based
contrast material. Radiology, 270(3):834{841, 2013.
[123] Benjamin A Plog and Maiken Nedergaard. The glymphatic system in central nervous system
health and disease: past, present, and future. Annual Review of Pathology: Mechanisms of
Disease, 13:379{394, 2018.
[124] Marios C Papadopoulos and Alan S Verkman. Aquaporin water channels in the nervous
system. Nature Reviews Neuroscience, 14(4):265, 2013.
[125] Jack A Wells, David L Thomas, Martin D King, Alan Connelly, Mark F Lythgoe, and
Fernando Calamante. Reduction of errors in asl cerebral perfusion and arterial transit time
maps using image de-noising. Magnetic resonance in medicine, 64(3):715{724, 2010.
[126] Xiaoyun Liang, Alan Connelly, and Fernando Calamante. Voxel-wise functional connec-
tomics using arterial spin labeling functional magnetic resonance imaging: the role of de-
noising. Brain connectivity, 5(9):543{553, 2015.
[127] Adnan Bibic, Linda Knutsson, Freddy St ahlberg, and Ronnie Wirestam. Denoising of
arterial spin labeling data: wavelet-domain ltering compared with gaussian smoothing.
Magnetic Resonance Materials in Physics, Biology and Medicine, 23(3):125{137, 2010.
[128] Danny JJ Wang, Jery R Alger, Joe X Qiao, Matthias Gunther, Whitney B Pope, Jerey L
Saver, Noriko Salamon, David S Liebeskind, UCLA Stroke Investigators, et al. Multi-delay
multi-parametric arterial spin-labeled perfusion mri in acute ischemic strokecomparison with
dynamic susceptibility contrast enhanced perfusion imaging. NeuroImage: Clinical, 3:1{7,
2013.
[129] SM Spann, X Shao, DJ Wang, CS Aigner, M Schloegl, K Bredies, and R Stollberger.
Improving temporal resolution of 3d arterial spin labeling perfusion imaging by combining
caipirinha encoding and spatio-temporal tgv reconstruction. In Proc Int Soc Magn Reson
Med, 2019.
[130] David C Alsop, John A Detre, Xavier Golay, Matthias G unther, Jeroen Hendrikse, Luis
Hernandez-Garcia, Hanzhang Lu, Bradley J MacIntosh, Laura M Parkes, Marion Smits,
et al. Recommended implementation of arterial spin-labeled perfusion mri for clinical ap-
plications: A consensus of the ismrm perfusion study group and the european consortium
for asl in dementia. Magnetic resonance in medicine, 73(1):102{116, 2015.
[131] Rita G Nunes, Shaihan J Malik, and Joseph V Hajnal. Single shot fast spin echo diu-
sion imaging with correction for non-linear phase errors using tailored rf pulses. Magnetic
resonance in medicine, 71(2):691{701, 2014.
105
[132] Christopher Nguyen, Zhaoyang Fan, Behzad Sharif, Yi He, Rohan Dharmakumar, Daniel S
Berman, and Debiao Li. In vivo three-dimensional high resolution cardiac diusion-weighted
mri: A motion compensated diusion-prepared balanced steady-state free precession ap-
proach. Magnetic resonance in medicine, 72(5):1257{1267, 2014.
[133] Fritz Schick. Splice: sub-second diusion-sensitive mr imaging using a modied fast spin-
echo acquisition mode. Magnetic resonance in medicine, 38(4):638{644, 1997.
[134] Mark E Bastin and Patrick Le Roux. On the application of a non-cpmg single-shot fast
spin-echo sequence to diusion tensor mri of the human brain. Magnetic Resonance in
Medicine: An Ocial Journal of the International Society for Magnetic Resonance in
Medicine, 48(1):6{14, 2002.
[135] David C Alsop. Phase insensitive preparation of single-shot rare: application to diusion
imaging in humans. Magnetic resonance in medicine, 38(4):527{533, 1997.
[136] Patrick Le Roux. Non-cpmg fast spin echo with full signal. Journal of Magnetic Resonance,
155(2):278{292, 2002.
[137] Ki Hwan Kim, Seung Hong Choi, and Sung-Hong Park. Improving arterial spin labeling by
using deep learning. Radiology, 287(2):658{666, 2017.
[138] Cagdas Ulas, Giles Tetteh, Stephan Kaczmarz, Christine Preibisch, and Bjoern H Menze.
Deepasl: Kinetic model incorporated loss for denoising arterial spin labeled mri via deep
residual learning. In International Conference on Medical Image Computing and Computer-
Assisted Intervention, pages 30{38. Springer, 2018.
[139] Danfeng Xie, Li Bai, and Ze Wang. Denoising arterial spin labeling cerebral blood
ow
images using deep learning. arXiv preprint arXiv:1801.09672, 2018.
[140] Madhava P Aryal, Tavarekere N Nagaraja, Kelly A Keenan, Hassan Bagher-Ebadian,
Swayamprava Panda, Stephen L Brown, Glauber Cabral, Joseph D Fenstermacher, and
James R Ewing. Dynamic contrast enhanced mri parameters and tumor cellularity in a rat
model of cerebral glioma at 7 t. Magnetic resonance in medicine, 71(6):2206{2214, 2014.
[141] Z Gao Huang, Dong Xue, Hasneen Karbalai, Alastair M Buchan, and Edward Preston.
Biphasic opening of the blood-brain barrier following transient focal ischemia: eects of
hypothermia. Canadian journal of neurological sciences, 26(4):298{304, 1999.
[142] Aysan Durukan, Ivan Marinkovic, Daniel Strbian, Miia Pitkonen, Eric Pedrono, Lauri
Soinne, Usama Abo-Ramadan, and Turgut Tatlisumak. Post-ischemic blood{brain barrier
leakage in rats: one-week follow-up by mri. Brain research, 1280:158{165, 2009.
[143] Fotios Vlachos, Yao-Sheng Tung, and Elisa Konofagou. Permeability dependence study
of the focused ultrasound-induced blood{brain barrier opening at distinct pressures and
microbubble diameters using dce-mri. Magnetic resonance in medicine, 66(3):821{830, 2011.
[144] Mitrajit Ghosh, Matilde Balbi, Farida Hellal, Martin Dichgans, Ute Lindauer, and Nikolaus
Plesnila. Pericytes are involved in the pathogenesis of cerebral autosomal dominant arteri-
opathy with subcortical infarcts and leukoencephalopathy. Annals of neurology, 78(6):887{
900, 2015.
106
Appendix A
Optimization of the timing of diusion gradients to
minimize eddy current artifacts
For a given time constant of eddy current decay
EC
, residual gradient led B is induced by
the on and o transition of diusion gradients [93]:
B(
ramp
;
EC
)/
1

ramp

EC

1e

ramp

EC

(A.1)
where
ramp
is the duration of diusion gradient ramping up/down. Assuming the dominant
mono-exponential decay, residual led of gradient waveform at the time t is:
B(t;
EC
)/ B(
ramp
;
EC
)e

t

EC
(A.2)
Residual gradient elds prior to image acquisition is the composition of B(t;
EC
) induced by
four gradient ramp up and four gradeient ramp down of the twice refocused bi-polar gradients:
B
EC
=
8
X
i=1
s(i)B(t
i
;
EC
) (A.3)
whereS(i) = 1 and -1 for gradient ramp up and ramp down respectively,t
i
is the time between
the end of each gradient switch and the tip-up pulse at the end of diusion preparation, which can
be expressed by the duration of four gradient lobes (T
i
), duration of the inversion pulses (T
inv
)
and
ramp
. According to Equation. (1) in [93], T
i
can be expressed by functions of pre-dened
preparation time (T
pr
), duration of diusion preparation (T
diff
) and one free-parameterT
4
. Thus,
equation A.3 can be rephrased as:
B
EC
=
8
X
i=1
s(i)B(t
i
(T
4
);
EC
) (A.4)
Eddy current was minimized by solving the following object function for a range of time
constants of delay (10-50 ms):
arg min
T4
k
X

EC
2[10;50]

8
X
i=1
s(i)B(t
i
(T
4
);
EC
)

k
2
2
(A.5)
For this study,
ramp
= 0.3 ms, T
pr
= 3.1 ms, T
inv
= 3 ms and T
diff
= 20 ms. Optimized
[T
1
, T
2
, T
3
, T
4
] = [3.4, 5.1, 5.5, 3.0] ms, which leads to a maximal of 4.4% (normalized to B)
residual gradient led for
EC
= 10 ms, 1.7% residual gradient eld for
EC
= 50 ms and no
more than 1.7% residual gradient eld for
EC
longer than 50 ms.
107
Appendix B
Estimation of voxel-wise R
1b
Voxel-wiseR
1b
was estimated based on the signal evolution with background suppression scheme
at two PLDs. The background suppression scheme consists of one saturation pulse and two
inversion pulses, which modulate the background suppressed tissue signal M according to T1
relaxation:
M(PLD;R
1b
) =M
0
(1e
sat(PLD)R
1b
2e
inv2(PLD)R
1b
+ 2e
inv1(PLD)R
1b
) (B.1)
M
0
is the reference signal acquired at the PLD of 2000 ms without background suppression or
diusion gradients,
sat
= +PLD is the total saturation time,
inv1
and
inv2
are the interval
between the rst and second inversion pulses and GRASE readout. [
inv1
(900),
inv2
(900)] =
[1166, 351] ms and [
inv1
(1800),
inv2
(1800)] = [1800, 519] ms according to [87]. Voxel-wise R
1b
can be computed by minimizing the object function:
arg min
R
1b
h
kM
900
0
M(900;R
1b
)k
2
2
+kM
1800
0
M(1800;R
1b
)k
2
2
i
(B.2)
where M
900
0
and M
1800
0
are the averaged control images acquired without diusion weighting
at PLD of 900 and 1800 ms respectively.
108

Abstract (if available)

Abstract Blood-brain barrier (BBB) maintains the homeostasis within the brain and the dysfunction of BBB has been linked to multiple central nervous system diseases and psychiatric disorders. In this dissertation, we first performed a thorough literature review on the biological mechanism of water exchange across the BBB. Previous animal studies have shown that increased water exchange occurs with loss of pericytes and before BBB leakage to contrast agent in Alzheimer's Disease (AD), while other studies demonstrated decreased water exchange in ischemic stroke and experiment induced BBB disruption. A hypothesis was proposed that changes of water permeability are bidirectional when BBB is prone to leakage: water permeability increases at early stage of BBB opening and decreases after chronic BBB leakage due to accumulated toxic substances in extravascular space and brain parenchymal damage. ❧ The purpose of this work is to present a novel MR pulse sequence and regularized modeling algorithm to quantify the water exchange rate, kw, across the BBB without contrast, and to evaluate its clinical utility in a cohort of elderly subjects at risk of cerebral small vessel disease (SVD). Ongoing studies have shown preliminary results about correlations between regional kw and Ktrans, a measurement of BBB leakage to contrast agent. ❧ A diffusion preparation module with spoiling of non-Carr-Purcell-Meiboom-Gill signals was integrated with pseudo-continuous arterial spin labeling and 3D gradient and spin echo readout. The tissue/capillary fraction of the arterial spin labeling signal was separated by appropriate diffusion weighting (b = 50 s/mm²). kw was quantifiedfied using a single-pass approximation model with total generalized variation regularization. A cohort of elderly subjects were recruited and underwent two MRIs to evaluate the reproducibility of the proposed technique. Correlation analysis was performed between kw and vascular risk factors, Clinical Dementia Rating scale, neurocognitive assessments, and white matter hyperintensities. ❧ The capillary/tissue fraction of ASL signal can be reliably differentiated with the diffusion weighting of b = 50 s/mm², given 100-fold difference between the (pseudo-)diffusion coefficients of the 2 compartments. Good reproducibility of kw measurements (intraclass correlation coefficient = 0.75) was achieved. Average kw was 105.0 ± 20.6, 109.6 ± 18.9, and 94.1 ± 19.6 min⁻¹ for whole brain, gray and white matter. kw was increased by 28.2%/19.5% in subjects with diabetes/hypercholesterolemia. Significant correlations between kw and vascular risk factors, Clinical Dementia Rating scale, executive/memory function, and the Fazekas scale of white matter hyperintensities were observed. Through comparison with DCE-MRI, significant correlation was found between kw and Ktrans in caudate, which indicate BBB permeability to both water and contrast agent could serve as imaging markers for cerebral small vessel disease. Our preliminary results also suggest regional kw and Ktrans changes are sensitive to different aspects of cognitive function, such as attention, episodic memory or cognitive flexibility. ❧ In conclusion, a diffusion prepared 3D pseudo-continuous arterial spin labeling sequence with total generalized variation regularized single-pass approximation modeling was proposed to measure BBB water permeability non-invasively with good reproducibility. kw may serve as a surrogate imaging marker of cerebral SVD and associated cognitive impairment.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

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

PDF

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

PDF

Radiofrequency pulse performance for myocardial ASL

PDF

Measurement of human brain perfusion with arterial spin labeling magnetic resonance imaging at ultra-high field

PDF

Improving sensitivity and spatial coverage of myocardial arterial spin labeling

PDF

Diffusion MRI of the human brain: signal modeling and quantitative analysis

PDF

Improved myocardial arterial spin labeled perfusion imaging

PDF

Blood-brain barrier pathophysiology in cognitive impairment and injury

PDF

Renin-angiotensin system antihypertensive medicines with and without blood-brain-barrier crossing potential: effects on cognitive outcomes in the elderly

PDF

Susceptibility-weighted MRI for the evaluation of brain oxygenation and brain iron in sickle cell disease

PDF

Improving arterial spin labeling in clinical application with deep learning

PDF

A virtual reality exergaming system to enhance brain health in older adults at risk for Alzheimer’s disease

PDF

Perimenopausal transition increases blood brain permeability: implications for neurodegenerative diseases

PDF

Unveiling the white matter microstructure in 22q11.2 deletion syndrome with diffusion magnetic resonance imaging

PDF

Applications of graph theory to brain connectivity analysis

PDF

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

Asset Metadata

Creator Shao, Xingfeng (author)

Core Title Mapping water exchange rate across the blood-brain barrier

School Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Biomedical Engineering

Publication Date 04/22/2019

Defense Date 03/21/2019

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

Tag arterial spin labeling,blood-brain barrier,magnetic resonance imaging,OAI-PMH Harvest,water permeability

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Wang, Danny JJ (committee chair), Marmarelis, Vasilis (committee member), Nayak, Krishna S. (committee member), Yan, Lirong (committee member)

Creator Email evanshaoxf@gmail.com,xingfens@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-139677

Unique identifier UC11675541

Identifier etd-ShaoXingfe-7211.pdf (filename),usctheses-c89-139677 (legacy record id)

Legacy Identifier etd-ShaoXingfe-7211.pdf

Dmrecord 139677

Document Type Dissertation

Format application/pdf (imt)

Rights Shao, Xingfeng

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