University of Southern California Dissertations and Theses

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

(USC Thesis Other)

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

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Improving the Sensitivity and Spatial
Coverage of Cardiac Arterial Spin Labeling for
Assessment of Coronary Artery Disease
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the Requirements for the Degree
DOCTOR OF PHILOSOPHY
(Electrical Engineering)
Ahsan Javed
December 2019

Dedicated to Ama, Abu, and Vanessa for their constant loving support and prayers ….
Acknowledgements
ii
Acknowledgements
I would like to express my gratitude toward the many people who helped with this
dissertation and supported me over the past 7 years. I am forever in their debt.
Firstly, I am grateful to my advisor Professor Krishna Nayak. I have deep admiration and
respect for him, not just as a PhD advisor but as a role model as well. He fostered a lab culture that
was conducive to scientific research and allowed me to pursue my own ideas and carve out a
meaningful path. Consistently mentoring me during my years of research and providing much
needed professional development, he encouraged me to follow my interests and supported my
career goals and aspirations. I am truly grateful for his advice over the years on topics that include
time management, presentation skills, work life balance, academia, professional relationships and
mentoring, to name a few.
I am also grateful to Professor Justin Haldar for serving on my qualification and thesis
committees. Two of my favorite classes at USC taught by Professor Haldar improved my research
effectiveness and taught me how to approach MRI physics from a signals and systems perspective.
I appreciate his advice which gave me thoughtful insights and aided in my professional
development.
I would also like to thank Eric Wong for serving on my thesis committee and advising me
over the years. It was an honor to work with someone who is a Guru in ASL. I always admire his
humbleness and advice on how to approach research problems. I am grateful to him for making
himself available for skype calls and for driving up to USC for research meetings. He helped
brainstorm ideas for overcoming major roadblocks in my research and provided guidance and
direction for moving my research forward.
I would also like to thank Dr. Danny Wang for serving on both my qualification and
defense committees. Dr. Wang generously offered his time and knowledge to help guide my
research project. He also offered innovative ideas from his expertise in brain ASL to help improve
and guide future research on my project.
I would also like to thank Dr. John Woods and Dr. Richard Leahy for serving on my
qualification committee. They both provided valuable input that shaped my dissertation.
I am thankful to Dr. Parveen Garg for providing clinical support and mentoring me on the
development of a research plan for clinical translational research. His advice and guidance was
Acknowledgements
iii
instrumental in my professional development and his ideas brought one of the major components
of this dissertation to fruition. I am also grateful to Dr. Hossein Bahrami for his valuable insights
on professional and career development. He also taught me how to identify clinical research
problems, design and set up clinical protocol and execute experiments effectively.
I am also grateful to Dr. Graham Wright for providing his input on designing experiments.
I would also like to thank Dr. Nilesh Ghugre for being accommodating and facilitating the animal
studies that were a major component of my work. Both Dr. Wright and Nilesh made themselves
available to discuss research ideas and to provide mentorship over skype and in person during my
visits to Sunny Brook Research Institute. I am also indebted to Jennifer Barry, Jill Weyers, and
Venkat Swaminathan for providing their support for animal studies.
I am also grateful to have been a part of the Magnetic Resonance Engineering Laboratory.
Providing an environment to learn and grow, and colleagues who are now my friends, I am truly
grateful for my current and past lab mates. I have learned from each one of them and hold
wonderful memories. Invaluable to my USC experience, I wouldn’t have finished this dissertation
without them. I would like to especially thank Terrence Jao and Hung Do for being my first
mentors at USC, for sharing with me many meals, for teaching me how to scan, and for the
countless scan parties. I am honored to call them friends who have been there and given comfort
and advice through some very difficult times. I am eternally indebted to Vanessa Landes, who
more than anyone shared the ups and downs of my PhD, and agreed to share her life with me.
Wayne Chen has been an awesome office mate sharing insightful conversations about life and
professional development. I am also thankful for Yannick Bliesener for being the realist and
inspiring me to do better and think differently and to Yongwan Lim for always being there to listen
and share his thoughts and perspectives. I also owe a debt to Sajan Lingala for being a good mentor
and providing valuable insights on research ideas especially as it pertained to making career
choices after my PhD. I admire Johannes Toger who encouraged me to maintain a good work-life
balance.
I would also like to express my gratitude to my other teachers, friends and mentors in Los
Angeles without whom this journey would be incomplete. They provided me with spiritual and
moral support through thick and through thin. They encouraged me to be better than I thought I
could be and inspired me to serve not just professionally but personally and work to better the
Acknowledgements
iv
community around me. For this I believe I am a better person which has significantly influenced
my professional life.
Lastly, I am eternally grateful for my family who has supported me through every
phase in my life, has encouraged me to pursue my dreams and career, and has sacrificed personally
for me. My Aunt and Uncle were like parents away from home, providing for me financially when
needed, cooking and shipping food for me when I was homesick, and always available to listen
and provide moral support. I am also grateful for my siblings Hammad, Hassan, and Zareen for
their moral support and love which made this journey a little more pleasant. Amongst them
Hammad was always there to help me get through all the tough times and to celebrate all the
milestones. I am also grateful for my in-laws, who shared with us all the milestones of our PhDs
including conference travel and supported us in every way possible. Finally, I literally cannot
express enough gratitude to my parents; without their sacrifices, prayers, and support this dream
would never have been possible. To them I owe everything.

Table of Contents
v
Table of Contents
ACKNOWLEDGEMENTS ............................................................................................. ii
LIST OF PUBLICATIONS .......................................................................................... viii
JOURNAL PAPERS .................................................................................................... viii
CONFERENCE PAPERS ............................................................................................ viii
1. INTRODUCTION......................................................................................................... 1
1.1 MOTIVATION: CORONARY ARTERY DISEASE .............................................. 1
1.2 OUTLINE OF CONTRIBUTIONS .......................................................................... 3
1.3 DISSERTATION ORGANIZATION ...................................................................... 4
2. BACKGROUND ........................................................................................................... 5
2.1 MRI PHYSICS.......................................................................................................... 5
2.2 RF PULSE DESIGN ............................................................................................... 13
2.3 VELOCITY BASED DEPHASING ....................................................................... 23
2.4 RAPID IMAGING SEQUENCES .......................................................................... 26

Table of Contents
vi
2.5 ARTERIAL SPIN LABELING .............................................................................. 32
3. SATURATION STEADY PULSED LABELING .................................................... 38
3.1 INTRODUCTION .................................................................................................. 38
3.2 METHODS ............................................................................................................. 38
3.3 RESULTS ............................................................................................................... 42
3.4 DISCUSSION ......................................................................................................... 45
3.5 CONCLUSION ....................................................................................................... 49
4. SINGLE SHOT ECHO PLANAR IMAGING ......................................................... 50
4.1 INTRODUCTION .................................................................................................. 50
4.2 METHODS ............................................................................................................. 51
4.3 RESULTS ............................................................................................................... 56
Table of Contents
vii
4.4 DISCUSSION ......................................................................................................... 61
4.5 CONCLUSIONS..................................................................................................... 65
4.6 SUPPORTING INFORMATION ........................................................................... 66
5. CORONARY ENDOTHELIAL FUNCTION ASSESSMENT .............................. 69
5.1 INTRODUCTION .................................................................................................. 69
5.2 METHODS ............................................................................................................. 70
5.3 RESULTS ............................................................................................................... 75
5.4 DISCUSSION ......................................................................................................... 76
5.5 CONCLUSION ....................................................................................................... 81
6. CONCLUDING REMARKS ..................................................................................... 82
7. BIBLIOGRAPHY ....................................................................................................... 85

List of Publications
viii
List of Publications
Journal Papers
1. A Javed, NG Lee, N Ghugre, GA Wright, KS Nayak. Saturation Steady Pulsed ASL for
Myocardial Perfusion Imaging. Journal of Cardiac Magnetic Resonance in Medicine, 2019.
(in preparation)
2. A Javed, KS Nayak. Single-Shot EPI for ASL-CMR. Magnetic Resonance in Medicine, 2019.
(minor revision)
3. A Javed, A Yoon, S Cen, KS Nayak, P Garg. Feasibility of Coronary Endothelial Function
Assessment using Arterial Spin Labeled CMR. NMR in Biomedicine, 2019. (in press)
4. A Javed, YC Kim, MCK Khoo, SLD Ward, KS Nayak, Dynamic 3D MR Visualization and
Detection of Upper Airway Obstruction during Sleep using Region Growing Segmentation,
IEEE Transaction on Biomedical Engineering, 2016; (2) 431-437.
5. DJ Park, NK Bangerter, A Javed, J Kaggie, MM Khalighi, GR Morrell, A Statistical Analysis
of the Bloch-Siegert B1 Mapping Technique, Phys Med Biol 2013; (16) 5673-5691.
Conference Papers
1. A Javed, NG Lee, HP Do, N Ghugre, G. Wright, EC Wong, KS Nayak. Optimization of
Steady-Pulsed Arterial Spin Labeling for Myocardial Perfusion Imaging. Proc. ISMRM 27th
Scientific Session, Montreal, May 2019. P2210.
2. NG Lee, A Javed, TR Jao, KS Nayak. Improved Quantification for Steady-Pulsed ASL
Perfusion Imaging. Proc. ISMRM 27th Scientific Session, Montreal, May 2019. p2191.
3. VL Landes, TR Jao, A Javed, KS Nayak. Improved Velocity-Selective Labeling Pulses for
Myocardial ASL. Proc. ISMRM 27th Scientific Session, Montreal, May 2019, p. 4967.
4. A Javed, T Jao, N Ghugre, GA Wright, EC Wong, KS Nayak. Saturation-based Steady-
Pulsed Myocardial ASL Perfusion Imaging. Proc. SCMR 22nd Scientific Sessions, Seattle,
February 2019. P012.
5. A Javed, T Jao, KS Nayak. Velocity Sensitivity of Inner-Volume Cardiac Echo Planar
Imaging. Joint Annual Meeting ISMRM-ESMRMB, Paris, 2018.
6. RA Lobos, A Javed, KS Nayak, WS Hoge, J P Haldar. Robust Autocalibrated LORAKS for
Improved EPI Ghost Correction with Structured Low-Rank Matrix Models. Joint Annual
Meeting ISMRM-ESMRMB, Paris, 2018.
7. RA Lobos, A Javed, KS Nayak, WS Hoge, JP Haldar. Robust Auto-calibrated LORAKS for
EPI Ghost Correction. IEEE International Symposium on Biomedical Imaging, Washington
DC, April 2018.
List of Publications
ix
8. A Javed, T Jao, KS Nayak. Velocity Sensitivity of Inner-volume Echo-planar Imaging. Proc.
SCMR 21st Scientific Session, Barcelona, February 2018.
9. A Javed, HP Do, AJ Yoon, KS Nayak, P Garg. Coronary Endothelial Function Testing using
Continuous Cardiac ASL-CMR. Proc. SCMR/ISMRM Workshop on CMR in Ischemic Heart
Disease, Barcelona, February 2018.
10. A Javed, KS Nayak "Cardiac ASL using Single-Shot EPI at 3T." Proc. ISMRM 25th
Scientific Sessions, Honolulu, April 2017.
11. A. Javed, T. Jao , K.S. Nayak, Motion correction facilitates the automation of cardiac ASL
perfusion imaging, 18th SCMR meeting, Feb 2015.

Introduction Motivation: Coronary Artery Disease
1
1. Introduction
1.1 Motivation: Coronary Artery Disease
Coronary artery disease (CAD) is one of the leading causes of death in the United States.
It is responsible for approximately one third of all deaths in individuals over the age of 35 (1–3).
Clinical manifestations of CAD including chest pain (angina) and myocardial infarction are a result
of narrowing of coronary arteries, which reduces blood flow to the heart. This occurs due to
atherosclerosis as shown in Figure 1.1.

Figure 1.1: Progression of coronary artery disease. Coronary arteries (circled, left) provide blood, oxygen and,
nutrients to the heart. A healthy cross section is shown (top, right). Atherosclerosis is a disease in which coronary
arteries become narrowed due to the deposition of plaque. This can lead to CAD (middle, right). If a blood clot forms
and blood flow is blocked, a heart attack (aka, myocardial infarction) can occur (bottom, right). Image source:
mayoclinic.org.
The gold standard for diagnosis of CAD is via direct visualization of the epicardial vessels
using coronary angiography. However, coronary angiography is unsuitable for screening and
routine monitoring because it is invasive, involves ionizing radiation, and is expensive. In general,
it is reserved for the most high-risk patients. An alternative is to use myocardial perfusion imaging
which can assess CAD by comparing myocardial perfusion (MP) under rest and stress (4).
Clinicians have many different options to choose from, each with its own advantages and
disadvantages. Stress echo-cardiography uses regional wall abnormalities as a surrogate biomarker
for perfusion defects but has low sensitivity (4). SPECT and PET are radionuclide-based perfusion
imaging techniques which are more sensitive but expose patients to ionizing radiation and have
poor spatial resolution. They are most widely used but are not well suited for repeated use or
Introduction Motivation: Coronary Artery Disease
2
routing monitoring. Cardiac magnetic Resonance Imaging (CMR) first pass perfusion offers good
spatial resolution and is radiation free. First-pass perfusion has recently been shown to have non-
inferior outcomes to leading invasive assessments such as FFR, when used to guide treatment in
patients with CAD (5). However, first pass perfusion requires injection of a gadolinium-based
contrast agents which are contra-indicated in patients with kidney disease. Gadolinium based
contrast agents can cause nephrogenic systemic fibrosis (NSF) in patients with renal failure (6).
All existing techniques to evaluate CAD are limited by either low sensitivity, use of
ionizing radiation and/or the need for contrast injection which are not well tolerated in patients
with kidney disease. Nevertheless, patients with kidney disease would benefit immensely from
routine monitoring of CAD. Both reduced glomerular filtration rate (GFR) and proteinuria in
chronic kidney disease (CKD) patients increases the risk of CAD. Yet these patients avoid routine
CAD exams because of the associated risk of NSF or cancer. Poorly managed CKD almost always
progresses to end stage renal disease (ESRD). There are approximately 600 K American with
ESRD and 17.6 million with CKD with an annual growth rate of 3-4% (7). Almost half the death
in ESRD patients are cardiovascular of which 20% can be associated with CAD (8). Put simply
patients with kidney diseases have a 10x higher likelihood of developing CAD and 3x higher
likelihood of dying from CAD. The only curative treatment is kidney transplant but that requires
patients to have a yearly CAD assessment. This assessment is usually done with stress echo-
cardiography but we believe that there is a need for a more sensitive MP imaging technique that is
contrast free and poses no incremental risk to the patients.
Arterial spin labeling (ASL) is an alternative contrast free MR technique which uses blood
as an endogenous contrast agent. The technique works by magnetically labeling arterial blood and
imaging the tissue after the labeled blood has perfused the tissue. Pocelete et al. (9) first
demonstrated the feasibility of ASL in the heart in both healthy swine and humans. Several papers
were published over the next few years to establish feasibility of ASL in the heart (10–13). Zun et
al. performed a small clinical study to show that ASL can be used to diagnose coronary artery
disease in humans (14). Their work also exposed several technical challenges that will need to be
resolved before ASL could become clinically viable.
Firstly, perfusion measurements made with ASL-CMR are noisy because the observed
signal is only 1-4% of the total myocardial signal. Even though the heart is one of the most highly
perfused organs, the high signal advantage is lost due to fluctuations from cardiac and respiratory
Introduction Outline of Contributions
3
motion. These fluctuations are referred to as physiological noise and are the dominant source of
noise in ASL measurements. Secondly, to make ASL clinically viable at least 3-slice coverage in
needed based on AHA’s recommendations. However, in humans, pharmacologic stress is limited
to 3-4 min due to patient comfort and safety concerns which is the time currently required for
single slice ASL exam. In this thesis, we try to address the above two limitations by developing
ASL techniques that are compatible with multi-slice imaging and improve sensitivity of ASL-
CMR. We also explore new applications of ASL-CMR for early diagnosis of CAD.
1.2 Outline of Contributions
Sensitivity of ASL
We improved the steady pulsed labeling (SPASL) technique developed by Capron et al
(15) to improve the sensitivity of ASL in the heart. The original technique was not compatible with
multi-slice imaging and was sensitive to spin history. In our work, we eliminated the sensitivity of
the technique to spin history without sacrificing improvements in sensitivity and we demonstrated
SPASL that would be compatible with multi-slice imaging. We further optimized the labeling
scheme to maximize signal strength and signal efficiency.
Spatial Coverage in ASL
Improving spatial coverage in ASL is very challenging because the duration of
pharmacologic stress limits the acquisition time to 3-4 minutes. Spatial coverage can be improved
using either sequential multi-slice or simultaneous multi-slice imaging. With current technology
balanced steady state precession cannot be used with either techniques. In our work, we developed
an echo planar imaging (EPI) sequence that is compatible with sequential multi-slice imaging and
optimized it for cardiac imaging. We then demonstrated the feasibility of using this EPI sequence
to improve spatial coverage of ASL in the heart.
Applications of ASL
ASL is a contrast free technique that poses no incremental risk to the patient which makes
it ideal for screening high-risk but asymptomatic patients for early diagnosis of coronary artery
disease. We explored the application of ASL for diagnosis of coronary endothelial function (CED)
and demonstrated the feasibility of ASL to diagnose CED.
Introduction Dissertation Organization
4
1.3 Dissertation Organization
Chapter 2 contains background information on MRI, including basic MRI physics, basic
RF pulse design, basic principles of two rapid imaging sequences, overview of ASL in the brain
and the heart along with the unmet needs in cardiac ASL. Chapter 3 describes development of
saturation steady pulsed labeling. Chapter 4 presents the implementation of a EPI sequence that is
compatible with sequential multi-slice cardiac imaging and its use to measure myocardial
perfusion using ASL. Chapter 5 presents the application of ASL-CMR to detect coronary
endothelial dysfunction. Finally, Chapter 6 gives concluding remarks.

Background MRI Physics
5
2. Background
This chapter provides the fundamental concepts needed to understand later chapters.
Section 2.1 provides an overview of MRI physics, 2.2 discusses basics of RF pulse design, 2.3
provides basics of balanced steady state free precession and echo planar imaging, and 2.4 gives an
overview of cardiac arterial spin labeling techniques.
2.1 MRI Physics
Magnetic Resonance Imaging (MRI) is a powerful non-invasive technique that is widely
used for diagnostic imaging. This section aims to present a brief overview of the MR phenomenon
using the a classical macroscopic explanation. An in-depth study would require an understanding
of quantum mechanics. However, our explanation should suffice for the contents of this thesis.
Nuclear Magnetic Resonance
In 1946, Felix Bloch and Edward Purcell theorized that atoms with charged nuclei create
an electromagnetic field and atoms with an odd number of protons and/or neutrons possess an
angular momentum or spin. These spins can be visualized as tiny spinning charged spheres which
act like microscopic bar magnets, as shown in Figure 2.1. Many biologically important specimens
possess nuclei with an odd number of protons exhibiting the NMR phenomenon, which explains
why MR is a powerful tool for biological research.
In the absence of an external magnetic field, the spins in a specimen are randomly oriented
with a zero net magnetic moment. In the presence of a strong magnetic field these spins align
themselves either parallel or anti-parallel to the external magnetic field. A small majority of spins
align in parallel to the main magnetic field yielding a non-zero net magnetic moment denoted as
𝑴 . Conventionally, the direction of the main magnetic field (non-zero magnetic moment at
equilibrium) is called the z-direction or the longitudinal direction.
In the presence of an external magnetic field these spins also exhibit nuclear resonance at
a well-defined frequency called the Larmor frequency, 𝜔 .The spins rotate about the main magnetic
field 𝐵 (16). The direction of rotation can be found using the left hand rule: when the thumb points
in the direction of the magnetic field, the fingers point in the direction of rotation.
𝜔 = 𝛾𝐵 , (1)
or
Background MRI Physics
6
𝑓 =
𝛾 2𝜋 𝐵 , (2)
where 𝛾 is the gyromagnetic ratio, a known constant for each type of atom and B is the strength of
the magnetic field. Hydrogen with its single proton is the most abundant element in the human
body, which generates the highest equilibrium polarization. Its gyromagnetic ratio is
𝛾 2𝜋 = 42.58
MHz/Tesla. Since hydrogen is the most studied atom in the human body and the only relevant one
to this work, we will assume hydrogen imaging in this thesis, unless otherwise noted.
It is important to note, the resonant frequency is not uniform inside a specimen and varies
spatially. Sources of variation include in-homogeneity in 𝐵 (known as off-resonance), changes in
susceptibility between tissues, and chemical shift. This plays a crucial role in MR imaging and is
an important consideration when designing an MR experiment.

Figure 2.1: Polarization of atomic spins. (left) Spins are randomly oriented in the absence of an external magnetic
field which results in zero net magnetic moment. (right) In the presence of an external magnetic field B0, spins align
parallel or anti-parallel to the magnetic field. A small majority of the spins aligns with the magnetic field yielding net
magnetic moment 𝑀 ⃗⃗
. Spheres represent atomic nuclei, straight line arrow running through the circle represent
direction of the atomic magnetic moments, and curved arrows show direction of precession about the magnetic
moment, based on the left hand rule.
Excitation
In the presence of B0, bulk magnetization 𝑴 will stay in a state of equilibrium. We can
manipulate this magnetization by carefully applying an additional magnetic field 𝐵 1
(𝑡 ). This
magnetic field is typically tuned to the resonant frequency of the atomic spins and is applied
perpendicular to the main magnetic field. 𝐵 1
(𝑡 ) is generally designed to move the bulk
magnetization 𝑴 into the transverse plane perpendicular to the main magnetic field. This process
of moving 𝑴 into the transverse plane for signal generation is known as excitation.
To simplify the explanation let us consider a frame that rotates at the same frequency as
𝐵 1
(𝑡 ) (generally the Larmor frequency). Such a frame is known as the rotating frame (16). In this
Background MRI Physics
7
frame B0 does not exist and 𝐵 1
(𝑡 ) is present in a static vector direction. 𝐵 1
(𝑡 ) causes the
rotation/precession of the bulk magnetization about its axis based on the left hand rule. The
magnitude of 𝐵 1
(𝑡 ) at any given time determines the instantaneous speed of the rotation and the
integral of 𝐵 1
(𝑡 ) is proportional to the amount of total rotation that is achieved. A simple example
of a 90° rotation is shown Figure 2.2, in the rotating frame.

Figure 2.2: Illustration of a 90° excitation in the rotating frame. The rotation of the bulk magnetization follows the
left hand rule about B1 which causes 𝑀 ⃗⃗
to rotate from its equilibrium position to the transverse plane.
Relaxation
After excitation, once the B1 field is turned off 𝑴 returns to its equilibrium state. The return
to equilibrium involves recovery of the longitudinal component, that is parallel to B0 and decay of
the transverse component, that is perpendicular to B0. Time constants T1 and T2 dictate the
recovery and decay of the longitudinal and transverse components, respectively. This process is
governed by equations 3 and 4:
𝑀 𝑧 (𝑡 ) = 𝑀 0
− (1 − 𝑀 𝑧 (0))𝑒 −
𝑡 𝑇 1
(3)

𝑀 𝑥𝑦
(𝑡 ) = 𝑀 𝑥𝑦
(0)𝑒 −
𝑡 𝑇 2
(4)

where 𝑀 0
is the equilibrium magnetization, 𝑀 𝑧 (0) is the longitudinal magnetization immediately
after excitation, and 𝑀 𝑥𝑦
(0) is the transverse magnetization after excitation. Figure 2.3 shows
longitudinal and transverse relaxation behavior for several tissues at 3T, following a 90° excitation.
In general, the overall behavior of 𝑴 can be described using the using the Bloch Equation:
𝑑 𝑴 𝑑𝑡
= 𝑴 × 𝛾 𝑩 −
𝑀 𝑥 𝒊 + 𝑀 𝑦 𝒋 𝑇 2
−
(𝑀 𝑧 − 𝑀 0
)𝒌 𝑇 1
(5)
Background MRI Physics
8

Detection and Image Formation
After B1(t) is switched off, the bulk magnetization 𝑴 will precess clockwise in the
transverse plane (using the left hand rule), about B0 as described by equation 1.5. As a result of
the Faraday’s law of induction (16), precession of 𝑴 will cause a change in magnetic flux that
would induce a small electro motive force (EMF) in a nearby receiver coil. This is the observed
MR signal, which is oscillating at the Larmor frequency and is decaying exponential at a rate 𝑇 2
∗
.
𝑇 2
∗
is a combination of tissue specific rate of decay 𝑇 2
, and signal decay from magnetic field
inhomogeneity. The MR signal in this case is described as the free induction decay (FID):
𝑠 (𝑡 ) = ∫ 𝑴 (𝒓 )𝑒 −𝑗𝜔 (𝒓 )𝑡 𝑒 −
𝑡 𝑇 2
∗
𝑑 𝒓 𝑣𝑜𝑙 (6)
The detected FID signal so far contains no spatial information and cannot be used to
distinguish tissues/specimens within the sample. To encode spatial position within the FID signal
we use additional coils known as gradient coils which introduce changes in magnetic field as a
function of spatial position. In traditional MR scanners, we have gradient coils that can manipulate
magnetic field along the three spatial dimensions (𝑥 ,𝑦 ,𝑎𝑛𝑑 𝑧 ). The gradient fields always point
along z-axis i.e. the same direction as B0. Therefore, changes in the magnetic field modulate the
precession frequency which is used to encode position in the FID signal. Generally, position along
one axis can be encoded in frequency and position along other axes is encoded in the phase of the
FID signal. A series of these FID’s can be acquired using a set of gradient waveforms to form a
Figure 2.3: Relaxation of magnetization for blood, myocardium, fat, and liver at 3T. A) T1 recovery is shown with
T1 values shown in parenthesis B) T2 decay is shown with T2 values shown in parentheses. Larger T1 values
correspond to slower recovery, and larger T2 values correspond to slower decay.
Background MRI Physics
9
set of k-space signals 𝑀 (𝒌 ), where 𝒌 is spatial frequency. FID signal for an acquisition can then
be defined as:
𝑠 (𝑡 ) = 𝑀 (𝑘 𝑥 ,𝑘 𝑦 ,𝑘 𝑧 ) = ∫
𝑧 ∫
𝑦 ∫𝑚 (𝑥 ,𝑦 ,𝑧 )𝑒 −𝑗 2𝜋 (𝑘 𝑥 (𝑡 )𝑥 +𝑘 𝑦 (𝑡 )𝑦 +𝑘 𝑧 (𝑡 )𝑧 )
𝑥 𝑑𝑥𝑑𝑦𝑑𝑧 (7)

where,
𝑘 𝑥 (𝑡 ) =
𝛾 2𝜋 ∫ 𝐺 𝑥 (𝜏 )𝑑𝜏
𝑡 0

𝑘 𝑦 (𝑡 ) =
𝛾 2𝜋 ∫ 𝐺 𝑦 (𝜏 )𝑑𝜏
𝑡 0
(8)
𝑘 𝑧 (𝑡 ) =
𝛾 2𝜋 ∫ 𝐺 𝑧 (𝜏 )𝑑𝜏
𝑡 0

Equation 1.7 relates 𝑠 (𝑡 ) to the Fourier transformation of the image 𝑚 (𝑥 ,𝑦 ,𝑧 ). In MR data
is acquired in k-space which is the Fourier transform space for images. Additionally, in practice
discrete data is acquired and we use discrete Fourier transform to reconstruct images from k-space
using the equation below (for data sampled on a 𝑀 × 𝑀 × 𝑀 grid):
𝑚 (𝑥 ,𝑦 ,𝑧 ) = ∑
𝑀 2
−1
𝑞 =−
𝑀 2
∑
𝑀 2
−1
𝑙 =−
𝑀 2
∑ 𝑆 (𝑙 Δ𝑘 𝑥 ,𝑝 Δ𝑘 𝑦 ,𝑞 Δ𝑘 𝑧 )𝑒 𝑗 2𝜋 (𝑙 Δ𝑘 𝑥 𝑥 +𝑝 Δ𝑘 𝑦 𝑦 +𝑞 Δ𝑘 𝑧 𝑧 )
𝑀 2
−1
𝑝 =−
𝑀 2
(9)
Please note the above equation 9 defines the reconstruction of an image using a 3DFT but
it is an imperfect approximation that assumes a discrete object. In most MR experiments k-space
is acquired on a cartesian grid in either 2D or 3D. Since k-space represents the spatial frequency
domain for images, simple analytical relationships can be derived to determine the correct
coverage of k-space for particular imaging tasks. The following analysis is relevant to uniform
cartesian sampling. The desired imaging field of view (𝐹𝑂𝑉 ) and spatial resolution (δ) can be used
to determine the sampling interval Δ𝑘 and highest spatial frequency 𝑘 𝑚𝑎𝑥 in each direction using
equations 10 and 11 (16). These can then be used to design gradients for the imaging pulse
sequence. Figure 2.4 shows the relationship between sampling in k-space, image resolution and
FOV.
𝐹𝑂𝑉 =
1
Δ𝑘 (10)
δ ≈
1
2𝑘 𝑚𝑎𝑥 (11)
Background MRI Physics
10

Figure 2.4:Relationship between sampling in k-space, image resolution and FOV. (left) sampling in k-space and
(right) image with an overlaid grid showing resolution and FOV. This image was adapted from
https://www.mriquestions.com/field-of-view-fov.html.
Noise
Thermal Noise
Signal-to-noise (SNR) ratio, and contrast-to-noise (CNR) ratio are often used to evaluate
Image quality in MRI. SNR is defined as signal amplitude (S) divided by standard deviation of
thermal noise (𝜎 𝑇𝑁
) and CNR is defined as the signal difference between two structures
(Δ𝑆 ) divided by 𝜎 𝑇𝑁
.
𝑆𝑁𝑅 ≜
𝑆 𝜎 𝑇𝑁
(12)
𝐶𝑁𝑅 ≜
Δ𝑆 𝜎 𝑇𝑁
(13)
Analysis of SNR and CNR in absolute terms is very complicated because it is dependent
on a multitude of factors that include hardware and imaging sequence parameters. Here we will
discuss the three major factors that affect SNR 1) B0 field strength, 2) acquisition time, and 3)
spatial resolution. Please refer to Nishimura, Haacke et al. and Macovski et al. (16–18) for a more
detailed analysis on how SNR and CNR in MRI are affected by various factors.
B0 Field Strength
Signal and noise in MRI are both dependent on field strength. Signal follows square-law
growth with 𝐵 0
. Variance of noise associated with receiver coils increases with 𝐵 0
1/2
and variance
Background MRI Physics
11
of noise from the inductive coupling between coils and body increases with 𝐵 0
2
. The relation
between SNR and field strength can, therefore, be expressed using the equation:
𝑆𝑁𝑅 ∝
𝐵 0
2
√
𝛼 𝐵 0
1
2
+ 𝛽 𝐵 0
2
(14)

where 𝛼 and 𝛽 are constants that determine the relative contribution of noise from receiver
electronics and body, respectively. It is important to note that noise from both sources is additive,
white gaussian. In the receiver electronics noise originates from the resistance of the circuit
components and body noise comes from the inductive coupling between the body and the receiver
coils. At higher field strengths (those used in most clinical scanners) 𝛼 ≪ 𝛽 i.e. body noise is the
dominant source of noise. In this regime, eq. 14 can be simplified to 𝑆𝑁𝑅 ∝ 𝐵 0
, which implies
that SNR varies linearly with field strength. However, if the noise from the receiver electronics is
dominant (i.e. 𝛼 ≫ 𝛽 ) then S𝑁𝑅 ∝ 𝐵 0
7
4
, which implies that SNR decreases almost as a square of
the decrease in field strength.
The relation between SNR and B0 is a bit more complicated because field strength also
affects other tissue properties such as T1, T2, B1 variation etc. All of them can have a significant
effect on both SNR and CNR. The relations presented in equation 14 can still serve as a good rule
of thumb.
Acquisition Time
Acquisition time can be increased by increasing the number of averages, readout time, or
number of phase encodes. Increasing the number of averages leads to and increase in SNR by
√𝑁 𝑎𝑣𝑒 . Because if 𝑁 𝑎𝑣𝑒 are performed signal and variance of noise both increase by 𝑁 𝑎𝑣𝑒 .
Variance of noise increases linearly with 𝑁 𝑎𝑣𝑒 because variances add for identical and independent
noise then using equation 12 we get:
𝑆𝑁𝑅 =
𝑁 𝑎𝑣𝑒 𝑆 √𝑁 𝑎𝑣𝑒 𝜎 𝑇𝑁
2
= √𝑁 𝑎𝑣𝑒
𝑆 𝜎 𝑇𝑁
(15)

Readout time can be increased in several ways. But to increase readout duration while
keeping the FOV and resolution constant requires the lowering of the readout gradients. Lowering
the gradients lowers the pertinent signal bandwidth and the required sampling rate which matches
the signal bandwidth. Noise variance per sample is subsequently lowered because:
𝜎 𝑇𝑁
2
∝ Δ𝑓 (16)
Background MRI Physics
12
where Δ𝑓 is the sampling frequency, noise samples are still independent, and the signal (S) is
unaffected by lowering the gradients. To increase the readout time by N the gradients need to be
lowered by N which leads to a reduction in Δ𝑓 by N which leads on an increase in SNR by a factor
of √𝑁 , using equation 12. This implies that SNR is proportional to the square root of the readout
duration (Tread).
𝑆𝑁𝑅 ∝ √𝑇 𝑟𝑒𝑎𝑑 (17)
A similar relationship holds when number of phase-encodes are increased without
changing the FOV or resolutions. In summary, SNR is proportional to the square root of number
of averages, readout duration, and number of phase encodes. All of these factors lead to an increase
in scan time so SNR is directly proportional to the square root of total scan time (Tscan).
𝑆𝑁𝑅 ∝ √𝑇 𝑠𝑐𝑎𝑛 (18)
Spatial Resolution
SNR is directly proportional to voxel size. Assuming all other factors remain constant for
a given voxel with dimensions 𝛿 𝑥 ,𝛿 𝑦 ,𝑎𝑛𝑑 𝛿 𝑧 :
𝑆𝑁𝑅 ∝ 𝛿 𝑥 × 𝛿 𝑦 × 𝛿 𝑧 (19)

It is important to note that this relationship is a “rule of thumb” and the exact relationship
depends on the spatial frequency content of the images and other properties.
When choosing parameters from MR experiment it is important to also consider; SNR is
proportional to the square-root of total scan time whereas it is proportional to spatial resolution. In
general, if higher SNR is needed it will be more efficient to acquire lower resolution images instead
of performing averages to improve SNR.
Physiological Noise
Physiological noise is an important consideration in quantitative MRI. It is often used to
assess the variation in measurements over time and when combined with thermal noise defined in
the previous section it gives a more complete picture of noise in an image.
𝜎 = √𝜎 𝑇𝑁
2
+ 𝜎 𝑃𝑁
2
(20)
Using this total noise temporal SNR can be defined to assess the temporal stability of signal
over time:
𝑇𝑆𝑁𝑅 =
𝑆 𝜎 (21)
Background RF Pulse Design
13
Physiological noise was originally defined by Kruger and Glover model (19). In the KG model
physiological noise was assumed to scale with MR signal as 𝜎 𝑃𝑁
= 𝜆𝑆 , where 𝜆 was the
proportionality constant. However, more recently Triantafyllou et al. proposed a new model that
relaxes the assumption that PN must scale with signal (S) by incorporating the physiological noise
in the coil covariances (20,21). They capture the fluctuations of the non-signal dependent
components of physiological noise using an extra term 𝛼 , to give the model:
𝑇𝑆𝑁𝑅 =
𝑆𝑁𝑅 √1 + 𝛼 + (𝜆 𝑆𝑁𝑅 )
2
(22)

This model reduces to the KG model when all components of PN scale with signal i.e. 𝛼 =
0. Please note that in practice both 𝛼 and 𝜆 are unknowns and TSNR is calculated using the mean
signal divided by standard deviation of signal over time. The most important outcome of this model
is that improvements in SNR may not necessarily translate to improvements in TSNR, which may
reach an asymptotic limit of 1/𝜆 . Additionally, from this model we can also estimate the ration of
thermal and physiological noise that allows us to determine if we are operating in a thermal of
physiological noise regime. This can help guide experiment design to optimize TSNR for
quantitative MR experiments. Please refer to the following works for a more complete study of the
relation between TSNR and SNR and the factors that affect the tradeoff between thermal and
physiological noise for both brain (21), and cardiac imaging (22).
2.2 RF Pulse Design
In section 2.1.2, we described how a B1 field tuned to the Larmor frequency can induce the
rotation of the bulk magnetization vector, about the axis of the B1(t) field. In the absence of any
gradients this would excite the entire volume, sensitive to the RF coil. If a large volume is excited,
we need to encode all spatial positions in the volume which can make the MR experiment very
slow. To speed up acquisition we generally excite a limited plane in 2D (referred to as a slice) or
volume in 3D imaging. This is known as a selective excitation.
In a basic slice selective excitation, an RF pulse (B1(t)) is played in the presence of a 1D
gradient (in the direction of selectivity). Intuitively, B1(t) played during the gradient excites spins
that reside in z-location with resonant frequencies within the bandwidth (BW) of the RF pulse.
The resonance frequencies in presence of the gradient (assume Gz) are changed according to the
following equation:
Background RF Pulse Design
14
𝜔 = 𝛾 (𝐵 0
+ 𝐺 𝑧 𝑧 ) = 𝜔 0
+ 𝛾 𝐺 𝑧 𝑧 (23)
The thickness of the excited slice can therefore be controlled using the amplitude of the
slice selective gradient and BW of the RF pulse. This is shown in Figure 2.5.

Figure 2.5: Slice selective excitation. Slice thickness of the excited slice 𝛥𝑧 is a function of Gz and BW of the RF pulse.
The slice or frequency profile is proportional to the Fourier transform of the RF pulse. Slice thickness 𝛥𝑧 is inversely
proportional to gradient amplitude and directly proportional to the BW of the RF pulse.

Small Tip-Angle Approximation
The Bloch equation presented as eq. 5 can be used to study the rotation of the bulk
magnetization in the presence of a B1 field and a slice selective gradient. Assume that B1(t) is
played for a short duration so we can ignore relaxation effects. The Bloch equation for an RF pulse
played at the Larmor frequency with a selective gradient can be written as:
𝑑 𝑴 𝑑𝑡
= 𝑴 × 𝛾 𝑩 𝑒𝑓𝑓
(24)
with
𝑩 𝑒𝑓𝑓
= 𝐵 1
(𝑡 )𝒊 + γG
z
𝑧 𝒌 (25)

The Bloch equation can then be expressed in matrix form by substituting eq. 25 into eq. 24:
𝑑 𝑴 𝑑𝑡
= [
0 𝛾 𝐺 𝑧 𝑧 0
−𝛾 𝐺 𝑧 𝑧 0 𝛾 𝑩 𝟏 (𝒕 )
0 𝛾 𝑩 𝟏 (𝒕 ) 0
]𝑴 (26)
The equation above cannot be solved for the general case unless we assume that 𝑴 =
[0 0 𝑀 0
]
𝑇 at equilibrium and that B1(t) is a weak pulse leading to a small rotation of <30°. These
Background RF Pulse Design
15
two assumptions known as the small tip-angle approximations lead to the following two
approximations after a small tip excitation:
𝑀 𝑧 ≈ 𝑀 0

𝑎𝑛𝑑 ,
𝑑 𝑀 𝑧 𝑑𝑡
≈ 0 (27)
Assuming Mz is approximately the same as M0 after a small tip excitation and there is no
change in longitudinal magnetization, allows us to decouple the longitudinal (z-component) and
transverse components (x, and y- components) of Bloch equation in eq 26.
[

𝑑 𝑀 𝑥 𝑑𝑡
𝑑 𝑀 𝑦 𝑑𝑡
𝑑 𝑀 𝑧 𝑑𝑡
]

= [
0 𝛾 𝐺 𝑧 𝑧 0
−𝛾 𝐺 𝑧 𝑧 0 𝛾 𝑩 𝟏 (𝒕 )
0 0 0
][
𝑀 𝑥 𝑀 𝑦 𝑀 0
] (28)
Defining,
𝑀 𝑥𝑦
= 𝑀 𝑥 + 𝑖 𝑀 𝑦
𝜔 1
(𝑡 ) = 𝛾 𝑩 𝟏 (𝑡 ) (29)
𝜔 (𝑧 ) = 𝛾 𝐺 𝑧 𝑧
we can simplify the Bloch equation in 28 as:
𝑑 𝑀 𝑥𝑦
𝑑𝑡
= −𝑖𝜔 (𝑧 )𝑀 𝑥𝑦
+ 𝑖 𝜔 1
(𝑡 )𝑀 0
(30)
The solution of this first order differential equation is:
𝑀 𝑥𝑦
(𝑡 ,𝑧 ) = 𝑖 𝑀 0
𝑒 −𝑗𝜔 (𝑧 )𝑡 ∫ 𝜔 1
(𝑠 )𝑒 𝑗𝜔 (𝑧 )𝑠 𝑡 0
𝑑𝑠 (31)
The transverse component of excitation is imaginary i.e. it is only in the y-axis because we
assume that B1(t) is along x-axis. Rotation happens about B1 based on the left-hand rule and should
results in magnetization being along y-axis if we start at equilibrium. Equation 31 describes the
relation between transverse magnetization and B1(t). It shows that in the small tip regime, the
Fourier transform of B1(t) predicts the slice profile, which is the spatial extent of excitation as a
function of spatial position (z). Setting z=0 and Gz=0 i.e. excitation is exactly on resonance,
equation 31 simplifies to:
𝑀 𝑥𝑦
(𝜏 ,𝑧 = 0) = 𝑖 𝑀 0
∫ 𝜔 1
(𝑠 )𝑑𝑠
𝜏 0
= 𝑖 𝑀 0
𝜃 (32)
Background RF Pulse Design
16
where 𝜃 is the flip angle and in the small tip regime 𝜃 ≈ sin𝜃 . Another important outcome
of equation 32 is that 𝜃 = ∫ 𝜔 1
(𝑠 )𝑑𝑠
𝜏 0
, can be used to calculate the flip angle for any arbitrary RF
pulse. Please refer to appendix II in Nishimura (16) for the detailed solution of the differential
equation in 1.30 and to references (16,23,24) for more details about the small tip-angle
approximation.
Excitation k-space
As seen in the previous section, there is a Fourier relationship between the slice profile and
B1(t) when RF pulses are played in the presence of a slice select gradient. The relation is similar
to the one which exists for the acquired signal s(t) and k-space (23). Hence, we can think about the
slice select gradients as traversing through a space referred to as excitation k-space while the RF
pulse applies a B1 weighting in this domain. This framework of excitation k-space presented by
Pauly et al. can thus be used to generalize the solution (Equation 31) of the small tip-angle
approximation to time varying gradients G(t).
𝑀 𝑥𝑦
(𝜏 ,𝒓 ) = 𝑖 𝑀 0
∫ 𝜔 1
(𝑠 )𝑒 −𝑗 2𝜋 𝒌 (𝑠 )∙𝒓 𝑑𝑠
𝜏 0
(33)
where,
𝒌 (𝑠 ) =
𝛾 2𝜋 ∫ 𝑮 (𝑡 ′
)𝑑 𝑡 ′
𝜏 𝑠 (34)
where 𝜏 is the duration of RF pulse. Equation 34 describes the trajectory mapped in
excitation k-space and 𝜔 1
(𝑠 ) in equation 33 represents the RF pulse weights at each k-space
location. This is similar to equation 7 for signal acquisition with one important distinction. In
equation 34, s goes from 0 to 𝜏 so excitation k-space trajectory always ends at the origin.
Background RF Pulse Design
17
The RF pulse weighting is not centered about the origin of k-space which implies that the
slice profiles will have a linear phase across the slice from the Fourier shift theorem. The phase
results in signal loss which in practice is easily resolved by playing a refocusing gradient lobe after
the RF pulse in turned off. The area of the refocusing gradient is half the area of the rectangular
gradient used during RF excitation with a symmetric RF pulse, and its amplitude is opposite to the
excitation gradient. In the excitation k-space interpretation it can be thought off as moving the
origin of k-space. In the small tip-angle regime this refocusing lobe is essentially canceling the
linear phase from the Fourier transform not being centered about origin. This is shown in Figure
2.6.
Figure 2.6: One-dimensional slice selective example with k-space interpretation. (top) RF pulse, (center) slice
selective gradient, and (bottom) k-space interpretation of RF. (left) A sinc RF pulse is played with a constant gradient
and the k-space interpretation shows the k-space origin at the termination of the gradient at location market 2. (right)
we have the same pulse with a refocusing gradient and this shows termination of the gradient puts the origin of k-
space at the center of the RF pulse making the RF weight symmetric about excitation k-space origin. This image was
adapted from Nishimura (16).
Background RF Pulse Design
18
Excitation k-space analysis provides a powerful framework to design more elaborate RF
pulses with time varying gradients including multi-dimensional RF pulses which are discussed in
the next section. Design of RF pulses to achieve a desired profile can be done using Shinnar Le-
Roux algorithm. This is a solved problem and interested readers can learn more about this in work
by Le-Roux et al. (25).
Multi-Dimensional RF pulses
Multi-dimensional RF pulses are selective in more than one dimension (26). These include
2D and 3D spatially selective RF pulses and spectral spatial pulses. 2D spatial RF pulses are
selective in two dimensions and select a long strip or cylinder. 3D spatial RF pulses are selective
in three dimensions and select a voxel. 3D spatial RF pulses are very long and are not used very
often. Spectral spatial pulses are spectrally selective in one spatial dimension and are very
frequently used for echo planar imaging (section 2.3.2). In this section we will focus our attention
on 2D spatial RF pulses. Interested reader can find more details on 3D RF pulses in work by
Stenger et al. (27).
Two-dimensional RF pulses require gradients to be played on two axes to achieve 2D
spatial selectivity. To maximize efficiency, the gradient axes are generally chosen to be orthogonal
to one another. The two commonly used k-space trajectories are echo planar and spiral. Here we
will discuss the echo planar trajectory which has similar design consideration to single shot echo
planar imaging discussed in section 2.3.2. For more details on design of spiral 2D pulses the
interested reader can consult work by Hardy et al. (26).
Two-dimensional echo planar (2DEP) pulses are ideal for 2D excitation because they allow
independent control of the slice profile in both dimensions. 2DEP pulses employ a fast oscillating
gradient that is played with B1(t) and it rapidly traverses back and fourth through k-space. The
second gradient, known as the blipped or slow gradient, is applied on the orthogonal axis and it
slowly blips through excitation k-space. This gradient is usually played at the zero crossings of the
fast gradient when the B1(t) is close to zero. Figure 2.7 shows and example of a 2D RF pulse with
23 fast lobes, 22 blips (of same polarity) and the corresponding excitation k-space trajectory.
2DEP pulses can be designed by specifying the oscillating and blipped gradients that
determine the k-space trajectory and two associated 1D RF pulses. One pulse is the fast pulse that
is played along with the fast gradient and determines the selectivity in the fast dimension. Whereas
the slow pulse provides the envelope that modulates the fast pulses throughout the duration of the
Background RF Pulse Design
19
2DEP pulse and provides the spatial profile along the slow dimension. Both of these pulses can be
designed with SINC, SLR, or any other desired pulse shape.
The amplitude of B1 for 2DEP pulses for a given set of gradients can then be determined
by inverting equation 33 as described by Le-Roux et al. (25):
𝐵 1
(𝑡 ) ∝ |𝑮 (𝑡 )||Δ𝒌 (𝑡 )|∫ ∫ 𝑀 𝑥𝑦
(𝒓 )𝑒 𝑗 2𝜋 𝒌 (𝑡 )∙𝒓 𝑑 𝒓 𝑠𝑝𝑎𝑡𝑖𝑎𝑙 𝑝𝑟𝑜𝑓𝑖𝑙𝑒 (35)

where |𝑮 (𝑡 )| provides uniform excitation k-space weighting when RF is played during
gradient ramps and |Δ𝒌 (𝑡 )| is the variable sampling density compensation. The former is related
to rate of traversal through excitation k-space and the later is related to spacing between excitation
k-space lines. In our analysis of 2DEP pulses, we ignore |Δ𝒌 (𝑡 )| because we have uniform spacing
between k-space lines and we account for |𝑮 (𝑡 )| because RF is played during gradient ramps to
reduce pulse duration. More details can be found in variable rate selective excitation (VERSE)
literature (28,29). Using equation 35 we can derive B1(t) for 2DEP pulse to be (30,31):
𝐵 1
(𝑡 ) = 𝐶 (𝜃 )𝐴 𝑠𝑙𝑜𝑤 (𝑘 𝑏𝑙𝑖𝑝 (𝑡 )) 𝐴 𝑓𝑎𝑠𝑡 (𝑘 𝑓𝑎𝑠𝑡 (𝑡 )) |𝑮 (𝑡 )| (36)
where 𝐶 (𝜃 ) is a normalization factor that is used to scale the pulse for flip angle 𝜃 .
Based on equation 36 the thickness of the profile along the fast direction (𝛿 𝑓𝑎𝑠𝑡 ) and slow
direction (𝛿 𝑠𝑙𝑜𝑤 ) can be determined using the dimension-less time bandwidth (TBW) of the fast
and slow RF along with extent of k-space in both dimensions.
𝛿 𝑠𝑙𝑜𝑤 =
𝑇 𝑠𝑙𝑜𝑤 Δ𝑓 𝑠𝑙𝑜𝑤 𝐾 𝑏𝑙𝑖𝑝 , (37)

𝛿 𝑓𝑎𝑠𝑡 =
𝑇 𝑓𝑎𝑠𝑡 Δ𝑓 𝑓𝑎𝑠𝑡 𝐾 𝑓𝑎𝑠𝑡 , (38)
where Tfast is the duration of the fast pulse, Tslow is the total duration of 2DRF, Δ𝑓 𝑓𝑎𝑠𝑡 is the
BW of the fast pulse, Δ𝑓 𝑠𝑙𝑜𝑤 is the BW of the slow pulse, 𝐾 𝑏𝑙𝑖𝑝 is the total extent of k-space
traversed in the blipped direction, and 𝐾 𝑓𝑎𝑠𝑡 is the total extent of k-space traversed in the oscillating
gradient direction.
Since the excitation k-space is discretely sampled in the blipped direction it results in
aliasing along that dimension and the spacing of the replicas (Δ𝑦 𝑟𝑒𝑝𝑙𝑖𝑐𝑎𝑠 ) is given by:
Δ𝑦 𝑟𝑒𝑝𝑙𝑖𝑐𝑎𝑠 =
𝑁 𝑏𝑙𝑖𝑝 𝐾 𝑏𝑙𝑖𝑝 =
𝑁 − 1
𝐾 𝑏𝑙𝑖𝑝 , (39)

Background RF Pulse Design
20
where N is the total number of lobes.
Another important design parameter in the 2DRF is the shift in the spatial profile with off-
resonance (𝛿 𝑝𝑟𝑜𝑓𝑖𝑙𝑒 ) and it can be derived from equations 37 and 39 as:
𝛿 𝑝𝑟𝑜𝑓𝑖𝑙𝑒 =
𝑁 𝑏𝑙𝑖𝑝 Δ𝑓 Δ𝑓 𝑠𝑙𝑜𝑤 𝛿 𝑠𝑙𝑜𝑤 , (40)
where Δ𝑓 is the off resonant frequency . This shift given in equation 40 and can be used to shift
the fat profile (440 Hz off resonant at 3T) outside the excited slice by carefully choosing pulse
parameters to give 𝛿 𝑝𝑟𝑜𝑓𝑖𝑙𝑒 > slice thickness. These RF pulses are used in reduced FOV imaging
and have recently been demonstrated for several application related to fast spine and cardiac
Figure 2.7: Two-dimensional RF pulse design. A) RF pulse with phase encoding and slice selective gradients. Red
arrow shows increasing time. B) Excitation k-space trajectory for the 2DRF pulse. Red arrows indicate the traversal
of excitation k-space and corresponds to increasing time shown in A). C) (left) 2D Pulse profile for the 2DRF pulse,
(center) 1D profiles in the slice encoding direction shows fat and water profiles at 3T, and (right) 1D profile in the
phase encoding direction. The simulated 2DRF pulse was designed with 23 sub-pulses, TBW of slow pulse 2.2, TBW
of fast pulse 10, slice thickness 10mm, phase FOV 80mm, fat shift 33mm, and max b1 of 12uT.
Background RF Pulse Design
21
imaging (31,32). Interested reader can also consult work by Alley et al. (30) for details on the
derivation of equations 36-40.
Spectral Spatial Pulses
Spectral spatial pulses or spatial spectral pulses (SPSP) are a class of 2DRF pulses that
excite magnetization in one spatial dimension at a specified location with specified spectral
content. They are often used in single shot echo planar imaging (EPI) to excite water only signal
without exciting the fat signal. SPSP pulses offer better tolerance to B1 in homogeneity compared
to conventional fat saturation pulses because the fat signal is never excited in SPSP pulses.
Interested readers can find detailed descriptions about the different types of SPSP pulses and how
to optimize their design in the works by Meyer et al., Bernstein et al., Zur et al. (33–35). Here we
will only provide a quick overview of SPSP pulses.
There are several different types of SPSP pulses but almost all of them use an oscillating
gradient much like the fast gradient in the 2DRF pulses discussed in section 2.2.3, a fast RF pulse
and a slow RF pulse (that serves as the envelope). An example of a SPSP pulse is shown in Figure
2.8A with the corresponding spatial and spectral slice profile. Excitation k-space can be used for
analyzing SPSP pulses and the two axes in k-space are kz and kf, where the former is the spatial
slice selective dimension (without loss of generality) and later is the spectral dimension. Phase in
the spatial dimension is accumulated by traversal through excitation k-space as a function of z 𝜙 =
2𝜋 𝑘 𝑧 𝑧 , from equation 33. Similarly, phase in the spectral dimension is essentially accumulated by
just waiting and is given by 𝜙 = 2𝜋𝑓𝑡 , where f is the frequency offset. By convention, we set
𝑘 𝑓 (𝑡 = 𝑇 𝑒𝑛𝑑 ) = 0 so 𝑘 𝑓 = 𝑇 𝑒𝑛𝑑 − 𝑡 , which implies that separating the RF sub-pulses in time
ensures that they are distributed along 𝑘 𝑓 .
Like the replicas in 2DRF pulses, SPSP also have replicas in the spectral dimension
because the pulses are discretely sampled along 𝑘 𝑓 . The replicas for the main peak appear at:
𝑓 = 0,±
2
𝑇 ,±
4
𝑇 ,±
6
𝑇 ,…
whereas secondary peaks appear halfway between the main replicas at:
𝑓 = ±
1
𝑇 ,±
3
𝑇 ,±
5
𝑇 ,…
where T is the period of the oscillating gradient and is an important parameter that has
important ramifications on parameters such as minimum slice thickness, and spectral selectivity
Background RF Pulse Design
22
of the pulse. The secondary peaks have smaller amplitude than the larger peaks and have odd
symmetry about 𝑧 = 0. Between the main peak and the secondary peak are the valleys with true
nulls, where no signal is excited. Depending on the desired profile and hardware constraints, the
lipid Larmor frequency can be centered on the true null to design a pulse known as true-null SPSP
pulse. Lipid Larmor frequency can also be centered on the secondary peaks to create a opposed-
null SPSP pulse. For more details on the tradeoff between the two designs and possible use cases
refer to work by Zur et al. (35).
There are two additional constraints to consider while designing SPSP pulses. Firstly, the
playing of RF during both the positive and negative lobes of the oscillating gradients or just playing
the RF during the positive lobe (fly-back design). The former requires high gradient and RF
fidelity. Presence of eddy currents or RF imperfections can result in severe artifacts due to Nyquist
ghosting in the spectral dimension. Whereas the fly-back design reduces the gradient and RF
fidelity constraints but increases the period of oscillation there by the designer may need to
minimum slice thickness for achieving true null design and so on. Secondly, some of the increases
in period of oscillations can be remedied by playing RF during gradient ramps which may allow
minimum possible slice thickness to be decreased. Like the 2DEP pulse, we can use equation 35
to determine the 𝐵 1
(𝑡 ) for the SPSP pulse:
𝐵 1
(𝑡 ) = 𝐶 (𝜃 )𝐴 𝑠𝑝𝑎𝑡 (𝑘 𝑧 (𝑡 ))𝐴 𝑠𝑝𝑒𝑐 (𝑡 )|𝑮 (𝑡 )| (41)
Figure 2.8: Spectral spatial RF pulse. A) SPSP RF pulse and slice select gradients. The pulse was designed with 13
sub-pulses for a slice thickness of 10mm with linear phase envelope. B) 2D frequency and spatial profile for the SPSP
pulse shows replicas in the frequency dimension at ±900Hz and has a flat profile over ±0.5cm and ±150 Hz. The pulse
also has -35dB of attenuation at -440 Hz which is the off-resonant frequency of fat, therefore it doesn’t excite any fat.
Background Velocity Based Dephasing
23
2.3 Velocity Based Dephasing
Generally, magnetic field gradients are used for encoding spatial position but they can also
be used to encode macroscopic flow into the phase of the MR signal. The class of gradients used
for this purpose are referred to as flow-sensitizing (FS) gradients. The most common type of FS
gradients are bipolar gradients, i.e. they consist of a pair of gradient lobes with equal area and
opposite polarity. For static spins bipolar gradients cause no net phase accumulation but for
moving spins they result in net phase accumulation along the direction of the gradients. The
accumulated phase is directly proportional to the velocity of the moving spins.
To illustrate this, consider a spin moving towards the isocenter, spins closer to isocenter
gather phase more slowly than spins further away from the isocenter. If a spin is moving towards
the isocenter during the first lobe of the bipolar gradient, it will accumulate a phase which would
be partially reversed by the second lobe of the bipolar gradient. All the phase from the first lobe
will not be reversed because the spin is closer to isocenter during the second lobe of the bipolar
and gathers phase more slowly. Hence, a moving spin will accumulate net phase in the presence
of bipolar gradients. Whereas if there were two spins that were static at the isocenter and at a
location away from the isocenter they will have no net phase accumulation from the bipolar
gradients. This is shown in Figure 2.9.

Figure 2.9: Phase evolution for static and moving spins in presence of bipolar gradients. a) Bipolar gradients of
duration 𝛥𝑡 , strength G0, Area A, and edge to edge distance T. b) phase accumulation for two spins not at the isocenter
as they experience the bipolar gradients. There is zero phase accumulated if the spins remain static but if the spin
from x1 moves to x2 then there is some net phase accumulation. This figure was adapted from Bernstein et al. (36).
Background Velocity Based Dephasing
24
Quantitatively, we can express the location of a spin at a time as a Taylor’s series expansion
if we know the velocity 𝑣 0
, acceleration 𝑎 0
, and location 𝑧 0
at time t=0.
𝑧 (𝑡 ) = 𝑧 0
+ 𝑣 0
𝑡 +
1
2
𝑎 0
𝑡 2
(42)
If the bipolar gradient is applied along z at time t>0 then the accumulated phase is:
𝜙 (𝑡 ) = 𝛾 ∫ 𝐺 (𝑢 )𝑧 (𝑢 )𝑑𝑢
𝑡 0
(43)
where u is the variable of integration denoting time. Inserting z(t) into 𝜙 (𝑡 ) we can write the phase
as:
𝜙 (𝑡 ) = 𝛾 ∫ 𝐺 (𝑢 )
𝑡 0
(𝑧 0
+ 𝑣 0
𝑢 +
1
2
𝑎 0
𝑢 2
)𝑑𝑢 (44)

𝜙 (𝑡 ) = 𝛾 𝑚 0
(𝑡 )𝑧 0
+ 𝛾 𝑚 1
(𝑡 )𝑣 0
+ 𝛾 𝑚 2
(𝑡 )𝑎 0
+ ⋯ (45)
where,
𝑚 𝑛 (𝑡 ) = ∫ 𝐺 (𝑢 )𝑢 𝑛 𝑑𝑢
𝑡 0
(46)
then for a spin moving a constant velocity, 𝑣 0
:
𝜙 (𝑡 ) = 𝛾 𝑚 1
(𝑡 )𝑣 0
(47)
Lets assume bipolar gradients of G0 amplitude with rectangular lobes of width Δ𝑡 starting at t=0,
and t=T respectively. The first moment can be calculated using:
𝑚 1
(𝑡 ) = ∫ −𝐺 0
𝑢𝑑𝑢 +
Δ𝑡 0
∫ 𝐺 0
𝑢𝑑𝑢 T+Δ𝑡 Δ𝑡 (48)
𝑚 1
(𝑡 ) = 𝐺 0
Δ𝑡𝑇 = 𝐴𝑇 (49)
where A is the area of each gradient lobe. This result can be generalized to any gradient waveform
by decomposing the waveform into the sum of infinitely small rectangles.
These flow sensitive gradients can be used to suppress flow from moving spins. In a basic
implementation following a 90° excitation the bipolar gradients can be used to induce a velocity
dependent phase. When the intravoxel velocity is heterogeneous this produces cancellation of
signal within the voxel and can be referred to as a saturation. This concept is often used in velocity-
selective ASL (37), discussed in the next section, and motion-sensitized driven equilibrium
(MSDE) flow suppression pulses (38). Using this technique moving spins above a certain cut off
velocity, 𝑉 𝑐 can be suppressed.
Background Velocity Based Dephasing
25
In the transverse plane, the signal equation has a Fourier relationship with velocity
distribution of spins within a voxel. In this case, instead of sampling in spatial frequency, we
sample in the velocity frequency domain 𝑘 𝑣 . Using the Fourier velocity encoding the relationship
between signal and velocity distribution, m(v) can be written as :
𝑀 (𝑘 𝑣 ) = ∫𝑚 (𝑣 )𝑒 −𝑗 2𝜋 𝑘 𝑣 (𝑡 )𝑣 𝑣 𝑑𝑣 (50)

𝑘 𝒗 (𝑡 ) =
𝛾 2𝜋 ∫ 𝜏𝐺 (𝜏 )𝑑𝜏
𝑡 0
=
𝛾𝐴𝑇 2𝜋 (51)
where assuming laminar flow with mean velocity v0, m(v) is a rect function with uniformly
distributed velocities within a voxel from 0 to 2𝑣 0
.
𝑚 (𝑣 ) =
𝑀 0
2𝑣 0
⊓ (
𝑣 − 𝑣 0
2𝑣 0
) (52)

Figure 2.10: Distribution of spins within a voxel under assumptions of laminar flow. Spins are uniformly distributed
from 0 to 2v0 where v0 is the mean velocity.
using the scaling and shift property of the Fourier transform.

𝑀 (𝑘 𝑣 ) =
sin(2𝜋 𝑣 0
𝑘 𝑣 )
2𝜋 𝑣 0
𝑘 𝑣 𝑒 −𝑗 2𝜋 𝑣 0
= 𝑠𝑖𝑛𝑐 (2𝑣 0
𝑘 𝑣 )𝑒 −𝑗 2𝜋 𝑣 0
𝑘 𝑣 (53)
The lowest 𝑘 𝑣 that gives zero signal is the first zero crossing of the sinc shown in Figure
2.11, 𝑘 𝑣 =
1
2𝑣 0
. Velocity cutoff is defined as:
𝑉 𝑐 ≜
1
2𝑘 𝑣 (54)
Background Rapid Imaging Sequences
26

Figure 2.11: The velocity profile plotted on a log scale is sinc shaped and the first zero-crossing is the cutoff velocity.
Using the definition of 𝑉 𝑐 and 𝑘 𝑣 we can derive an expression for 𝑉 𝑐 for any set of bipolar gradients:

2.4 Rapid Imaging Sequences
Balanced Steady State Free Precession
Steady state sequences form a class of MR sequences that are based on fast gradient
recalled echo (GRE) sequence. In steady state sequences the transverse and longitudinal
magnetization is kept in a steady state. These sequences have revolutionized cardiac imaging and
have become the standard for cardiac imaging both functional and anatomical. Primarily, due to
their good signal to noise ratio and increased speed of imaging.
Balanced steady state free precession (bSSFP) is a special case of steady state sequences
in which gradients on all three axes are fully refocused such that the gradient induced dephasing
on each axis is zero. The sequence is flow compensated and intrinsically insensitive to flow making
it ideal for cardiac imaging. The steady state signal for bSSFP is dependent on T1,T2, FA, TR, and
proton density as shown in equation below. Interested readers can learn more about bSSFP signal
evolution and derivation of steady state signal in (39,40). The steady state is generally reached
after 5*T1/TR and signal exhibits exponential decay to steady state during the transient phase after
the first excitation. This approach to steady state is dependent on T1 and T2 of the tissue. Imaging
during this transient state can exhibit various contrasts dependent of FA and number of imaging
𝑉 𝑐 =
𝜋 𝛾 ∫ 𝜏𝐺 (𝜏 )𝑑𝜏
𝑡 0
=
𝜋 𝛾 𝐺 0
𝛥𝑡𝑇 =
𝜋 𝛾𝐴𝑇 (55)

Background Rapid Imaging Sequences
27
TRs. Interested reader can read more about signal evolution of bSSFP during its transient state in
works by Scheffler et al. and Worters et al. (41,42).
𝑀 𝑥𝑦
+
= 𝑀 0
𝑠𝑖𝑛𝛼 (1 − 𝐸 1
)
1 − (𝐸 1
− 𝐸 2
)𝑐𝑜𝑠𝛼 − 𝐸 1
𝐸 2
, (56)
where 𝐸 1,2
= 𝑒 −
𝑇𝑅
𝑇 1,2
and 𝑀 𝑥𝑦
= 𝑀 𝑥𝑦
+
√𝐸 2
.
Transient state imaging is especially important for quantitative cardiac imaging which uses
snapshot bSSFP to acquire a complete image within the quiescent period of the cardiac cycle
(<300ms). Imaging during the transient phase is sensitive to signal fluctuations which can be
reduced using catalyzation techniques (42). Catalyzation techniques use either linear or Kaiser
Bessel windowed ramps (42) to slowly increase imaging flip angle. Data is acquired once the
desired/prescribed imaging FA is reached. The data acquired during catalyzation is discarded.
Background Rapid Imaging Sequences
28
Sensitivity of bSSFP imaging to prepared longitudinal magnetization is a function of
imaging FA and number of TRs required to reach the center of k-space (43). This is important in
quantitative MRI techniques, e.g. ASL and diffusion tensor imaging (DTI). Steady state SSFP has
almost zero sensitivity to prepared longitudinal magnetization. Due to the number of imaging TRs
needed to reach steady state before data acquisition. With fewer TRs needed to reach the center of
k-space, transient bSSFP has become the standard imaging method used for these techniques.
Figure 2.12: Balanced SSFP sequence diagram and signal profile. A) bSSFP pulse sequence diagram with Z,X,Y
representing slice select, phase encoding, and readout gradient axis. RF pulses are shown for 𝜋 phase cycling. B) The
steady-state signal in bSSFP as a function of off-resonant frequency for FAs of 20°,40°, and 60°. The profile has a
flat plateau over 0 Hz and sharp drops at off-resonant frequencies of integer multiples ±1/2TR. These drops results in
banding artifacts shown in C).C) A 4-chamber view of the heart with banding artifact shown with white arrows.
(Figure in A) was adapted from Worters et al. (42).

Background Rapid Imaging Sequences
29
Smith et al. (43) presented an optimization of the FA to maximize the sensitivity of bSSFP to
prepared longitudinal magnetization.
Another key characteristic of bSSFP sequences is the dependence of the signal on off-
resonance, as shown in Figure 2.12B. Phase cycling is used to center the off-resonance profiles of
SSFP on 0 Hz , in most common implementations. The phase of excitation pulses is incremented
by 𝜋 or alternatively altered between 0 and 𝜋 . With 0,𝜋 phase cycling the nulls of the SSFP off
resonance profile occur at 1/2TR and the bandwidth of the off-resonance profile is 1/TR. The
shorter the TR, the wider the off-resonance profile. If the nulls of the profile fall in an image they
appear as banding artifacts. Robust shimming is needed to minimize these off-resonance related
banding artifacts in bSSFP imaging sequences. In cardiac imaging, at high field banding artifacts
are often present and care is taken to shift the off-resonance profile (using phase cycling) such that
the bands fall outside the region of interest. An example of banding artifacts is shown in Figure
2.12.
Single Shot Echo Planar Imaging
Echo planar imaging (EPI) is a fast imaging technique that can acquire a complete MR
image in 20-100 ms. It was amongst the first MRI techniques developed by Sir Peter Mansfield in
the early 1980s (44). EPI is capable of essentially freezing motion due to its speed and is ideal for
imaging fast physiological processes and moving organs such as the heart. It is extensively used
for brain imaging and has already been demonstrated to be useful in diagnosis of stroke and for
studying brain function with fMRI, amongst other applications. Advances in gradient performance
and artifact correction techniques have allowed for robust EPI in double oblique slice orientations
enabling cardiac EPI. This has recently been used for cardiac DTI (45,46) and ASL (9).
In EPI, multiple lines of k-space are generally acquired after a single RF excitation. Data
is usually acquired with a fast oscillating readout gradient that oscillates between positive and
negative amplitude forming gradient echoes. Blips on the phase encoding axis encode each echo
as a different phase encode, which corresponds to a different line in k-space. Data can be acquired
immediately after the RF excitation to minimize echo time (TE) in gradient echo EPI. GRE-EPI is
sensitive to phase shifts from magnetic field in-homogeneity, chemical shifts and/or static tissue
susceptibility gradients which are not cancelled. Additionally, GRE-EPI image contrast is dictated
by T2* weighting where T2* < T2. To reduce sensitivity to phase shifts and to get T2 contrast a
180° pulse can be used to refocus magnetization in spin echo (SE) EPI, at the expense of TE. In
Background Rapid Imaging Sequences
30
cardiac imaging SE-EPI generally gives much better performance and is frequently used. During
data acquisition with GRE or SE EPI, all k-space lines may be acquired after a single excitation,
in single shot EPI or with multiple excitations in multi-shot EPI, as shown in Figure 2.13.

Figure 2.13: K-space coverage in EPI. (left) Entire k-space is acquired in single shot EPI.(right) Portions of k-space
are acquired in multi-shot EPI. An example of sequential three shot EPI is shown with k-space coverage of each shot
represented by a different color.
Single shot EPI is ideal for quantitative MRI due to its sensitivity to prepared longitudinal
magnetization. It is also more susceptible to imaging artifacts compared to conventional imaging,
primarily, due to reversal of every other echo with fast gradient switching and long readout
durations.
In EPI, every other k-space line is acquired under a negative gradient and must be reflected
with respect to time. Misalignment in acquired k-space lines can lead to Nyquist ghosting artifacts
at half of the field of view in the phase encoding direction. Generally, the misalignment occurs
due to imperfections in the RF receive chain, gradient timing errors and/or eddy currents that can
affect gradient fidelity. This misalignment can be corrected by using linear phase correction
(47,48), which is implemented on most scanners. For oblique imaging 2D correction methods can
further reduce ghosting artifacts (49–51), recently significant progress has been made in this area
and interested reader can consult works by Lee et al. and Xie et al. (52–54). An example of
ghosting artifact and both linear and 2D phase correction methods is shown in Figure 2.14.
The bandwidth per pixel in the phase encoding direction for an EPI readout is smaller than
the bandwidth per pixel in the readout direction by a factor equivalent to the number of k-space
lines per shot. Lower bandwidth per pixel and the long acquisition times can lead to larger-fat
water shifts, geometric distortion due to B0 inhomogeneity (from the magnet or the patient), signal
loss due to dephasing, and resolution loss due to T2 filter effects. The interested reader can learn
more about each of these effects in chapter 6 of the book by Schmitt et al. (55). Fat suppression,
Background Rapid Imaging Sequences
31
SPSP, or carefully designed 2DRF pulses can be used to limit artifacts from fat shifts. Geometric
distortion, signal loss due to dephasing and T2 filter effects can be reduced by increasing per pixel
bandwidth with parallel imaging or using reduced FOV imaging. Software methods also exist to
correct geometric distortion by using a B0 map but their use to moving organs is limited (56–61).
An example of geometric distortion with one possible correction algorithm is presented in Figure
2.15.

Figure 2.14: Example of phase correction methods for Nyquist ghost correction. (left) Axial image of the brain shown
with normal scaling, (right) the same image is shown with harsher thresholding to highlight lower intensity ghosting
artifacts. a) Image without any correction. b) Image with linear phase correction, we can observe a reduction in
artifacts but substantial ghosting is still visible in the right column image. c) Image corrected with 2D phase correction
as presented in (56), ghosting artifacts are almost completely removed. This figure was adapted from Chen et al. (56).
Reduced FOV imaging is an effect method to reduce the long readouts and subsequently
minimize artifacts, if the region of interest is smaller than the sensitive region of the transmit and
receiver coils. Reduced FOV imaging reduces the phase encodes required to achieve a desired
resolution and can significantly reduce readout times. Several methods exist for reduced FOV
imaging and interested readers can find a good survey of the methods and their performance in
work by Wargo et al. (62).

Figure 2.15: Example of geometric distortion correction in a
phantom with seven test tubes. a) Magnitude phantom image with
geometric distortion in the up down direction (phase encoding).
b) Geometric distortion was corrected using the algorithm
presented by Hu et al. (48).

Background Arterial Spin Labeling
32

2.5 Arterial Spin Labeling
Perfusion refers to the delivery of blood to supply oxygen and nutrients and to remove
waste from tissues. It is an important indicator of disease and can be used for diagnosis of stroke,
coronary artery disease, and renal insufficiency. It is generally measured with proton emission
tomography (PET), SPECT, and MRI using externally administered tracers that can exchange
between the vascular compartment and tissues. Arterial spin labeling (ASL) is a unique MRI
technique that uses RF pulses to generate an endogenous tracer. Therefore, it does not require
injection of a contrast agent, involves no ionizing radiation, and poses no incremental risk to the
patient which makes it infinitely repeatable.
Perfusion in ASL in measured by acquiring two images known as the ‘label’ and ‘control’.
In both images the static tissue signal is the same but inflowing blood signal is modulated to
generate the perfusion signal. Specifically, in the label images the blood magnetization is labeled
using either an inversion or saturation RF pulse. A fixed amount of time known as post labeling
delay, is allowed to pass between labeling and imaging to allow labeled blood to perfuse the tissue.
Control and label images are subtracted to calculate the perfusion signal which is proportional to
inflowing blood.
Arterial Spin Labeling in the Brain
ASL methods for measuring cerebral blood flow have found widespread acceptance in the
research community over the past 20 years. Over the years brain ASL has been used for several
applications including assessment of stroke (63), and functional magnetic resonance imaging
(fMRI) (64). More recently with improvements in sequence robustness, image resolution,
decreased acquisition times, and advances in automated post-processing capabilities ASL is ready
for clinical use. Currently, there are two main types of labeling methods used for brain ASL; 1)
Continuous ASL (CASL) and 2) Pulsed ASL (PASL). CASL was the first implementation of ASL
and all other methods have been developed to address limitations and technical challenges
associated with CASL (65).
CASL uses long continuous RF pulses (2-4s) to induce a flow-driven adiabatic inversion
to label spins close to the imaging plane (66). PASL uses short RF pulses (5-20ms) to saturate or
invert blood in a labeling slab that is proximal to the imaging plane. While CASL techniques have
Background Arterial Spin Labeling
33
been shown to have theoretically higher SNR than PASL (65), they do suffer from increase SAR
deposition, Magnetization transfer (MT) effects, and lower labeling efficiency (~80%) (65). PASL
techniques are more sensitive to transit delay effects which generally wipeout gains from improved
labeling efficiency (>95%) compared to CASL.
Pseudo-continuous ASL (PCASL) was developed to make use of the high SNR in CASL
and high labeling efficiency in PASL while lowering the MT effects and SAR compared to CASL.
PCASL uses discrete RF pulses for labeling with gradients in between to mimic the flow driven
adiabatic inversion used in CASL (67). PCASL has found widespread support from the brain ASL
community and is the recommended labeling methods for clinical use (68). All these methods use
spatial labeling and are sensitive to transit delay effects which are one of the largest sources of
errors in perfusion quantification. Velocity selective ASL (VSASL), which is a PASL technique,
labels spins everywhere including in the imaging plane based on the velocity of the spins thus is
theoretically insensitive to transit delay effects. Velocity selective tag pulses can be used for
saturation (37) or inversion (69,70) and label spins moving above a cutoff velocity 𝑉 𝑐 .
Due to the insensitivity to transit delays, VSASL pulses may be ideal for use in pathologies
where collateral or slower flow may increase transit delay effects e.g. stroke.
Arterial Spin Labeling in the Heart
Success of ASL methods in the brain provides motivation for their use in the heart but due
to significant physiological differences several modifications are needed to adapt ASL methods to
the heart. Briefly, the pathway that blood travels to perfuse the myocardium is more complicated
Figure 2.16: Schematic representation of labeling and imaging planes and pulse sequences. a) Labeling planes for
CASL/PCASL (red), PASL (green), and VSASL (blue) is shown with the imaging plane (white). The VSASL label is
shown for demonstration but VSASL label is spatially non-selective. b) Timing diagrams for the all four ASL labeling
methods in the brain. Inversion time (TI) is the same as PLD but is shown to clarify terminology used in literature.
It’s the duration from center of labeling pulse to center of imaging. This picture was adapted from Alsop et al. (68).
Background Arterial Spin Labeling
34
compared to the brain which impacts choice of labeling strategies. The flow of blood is pulsatile
and happens only during part of the cardiac cycle which impacts timing of labeling. The heart is
moving rapidly during the cardiac cycle and has short quiescent periods which impacts choice of
imaging methods and timing of imaging. Additionally, the intrinsic SNR in the heart is lower
which impacts the sensitivity of ASL in the heart. For more details on the physiological and
practical considerations when moving from brain to the heart for ASL please refer to the review
paper by Kober et al. (71).
ASL methods in the heart have been limited to PASL due to the complex geometry of the
heart which makes it difficult to find a suitable labeling plane of CASL and PCASL techniques.
Possible labeling candidates for CASL in the heart include the pulmonary vein or the ascending
aorta. Pulmonary veins are unsuitable because they are difficult to locate, have long transit times
to the myocardium, and labeling them may spuriously label the myocardium. The pulsatile flow
in the aorta and motion during the cardiac cycle also make it unsuitable for CASL.
Amongst the PASL techniques flow alternating inversion recovery (FAIR) is the most
common labeling strategy used in the heart (9,10,12,13). Figure 2.17 shows the common
implementation of FAIR in the heart. In FAIR, during the label acquisition a non-selective
adiabatic inversion is used to label the myocardium and all in-flowing blood. In the control
acquisition, a slice-selective slab is used to invert magnetization just around the imaging slice. In
both acquisitions, images are acquired after PLD of 1-2 cardiac cycle (RR intervals). The inflow
is positive during the control acquisition and negative during the label acquisitions. Subtracting
the label and control images yields the perfusion signal which is twice the in-flow, after correcting
for T1 recovery of blood. FAIR has been the mainstay for ASL in the heart because of its ease of
implementation and simple quantification. However, FAIR is sensitive to transit delay effects
especially for multi-slice imaging and has low sensitivity.
To resolve the transit delay sensitivity of FAIR Jao et al. recently demonstrated VSASL
for the heart (72). VSASL in the heart was combined with background suppression to lower
physiological noise and improve sensitivity. Background suppression uses a non-selective
inversion pulse after labeling to suppress the myocardial signal. The timing of inversion is chosen
to center imaging around the zero-crossing of the myocardial T1. With background suppression
VS-ASL works similar to FAIR but labeling is dependent on velocity rather than spatial position.
In the label acquisition, VSSAT labels blood moving above Vc whereas during the control
Background Arterial Spin Labeling
35
acquisition VSSAT is played with a infinite Vc so no blood is labeled, but T2 effects are balanced.
In both acquisitions, following the VSSAT a non-selective labeling pulse is used for background
suppression. Hence in the VSASL control acquisition inflow is negative blood and VSASL label
acquisition inflow is saturated blood (instead of blood at equilibrium in FAIR control). The
difference in control and label gives us the ASL signal which is half the signal in FAIR. Therefore,
VSASL with VSSAT is expected to have half the sensitivity of FAIR ASL but is theoretically
insensitive to transit delays.
Recently, Capron et al. (15) demonstrated steady pulsed labeling (SPASL) to improve the
sensitivity of ASL in the heart. In SPASL, labeling and imaging is performed every cardiac cycle
to achieve a flow driven steady state. Blood is labeled in the aortic root during the label acquisition
and no blood is labeling during the control acquisition. To balance MT effects SPASL is
compatible with both EPISTAR (73) and PICORE labeling (74). In EPISTAR, label acquisition is
performed by placing the labeling slab proximal to the imaging slice whereas the control
acquisition is performed by placing the labeling slab distally. To balance the MT effects the
labeling slabs are placed symmetrically on either side of the imaging slice. In PICORE, the labeling
is performed similar to EPISTAR but the control acquisition is performed using an off-resonant
pulse. To balance MT effects the off-resonant control pulse is played at the same frequency as the
Figure 2.17: FAIR Labeling scheme in the heart uses a label and control image to measure the myocardial perfusion.
FAIR’s label acquisition uses an RF pulse to non-selectively invert all tissue magnetization. After waiting for delivery
of labeled blood to the myocardium, a short axis image is obtained with inverted myocardium and blood flow. In the
control acquisition, a slice-selective inversion is used to invert signal only around the desired short axis slice. After
waiting for delivery of non-labeled blood to the myocardium, an image is obtained with inverted myocardium and
non-inverted fresh blood flow. The difference between control and labeled acquisitions is taken to yield ASL signal
that is proportional to 2 times myocardial blood flow.
Background Arterial Spin Labeling
36
labeling pulse, but without any gradients. PICORE unlike EPISTAR is compatible with multi-slice
imaging and has half the residual MT effects compared to EPISTAR (74,75).
Challenges and Limitations of ASL in the heart
ASL is a low signal technique with ASL signal ranging from 0.3%-3% of the M0 under
physiological blood flow conditions of 0.5 ml/g/min - 4 ml/g/min. This makes ASL very sensitive
to small fluctuations in signal which can cause errors in perfusion measurements. ASL in the heart
is then extra challenging due to physiological noise that arises from cardiac and respiratory motion,
heart rate variations, imperfect subtraction between control and label images and physiological
fluctuations in MP. PN is the dominant source of noise and have been found to be 2-3x higher
than TN for cardiac imaging. High PN and low ASL signal limit the sensitivity of existing ASL
techniques in the heart and significant improvements are needed in sensitivity to make ASL in the
heart clinically viable.
Figure 2.18: MT balanced strategies for PASL. In EPISTAR and PICORE labeling is performed the same way on the
proximal end of blood flow. Control images in EPISTAR are acquired by playing a labeling pulse symmetrically on
the other side of the imaging slice (ideally distal end of blood flow). Where as in PICORE the control acquisition is
performed by playing the labeling pulse without slice selective gradients but at the same frequency as the labeling
pulse which implies that labeling pulse is played off-resonant.
Background Arterial Spin Labeling
37
Current ASL techniques in the heart are also limited to a single slice. To make ASL in the heart
clinically viable, coverage needs to be extended to at least 3 myocardial short axis slices (Basal,
Mid, and Apical), as shown in figure below. This is the minimum requirement for covering all
coronary artery territories needed to guide treatment decision in perfusion CMR (76). Increasing
coverage is challenging because changes in perfusion can only be detected during stress and the
duration of pharmacological stress is limited to 3 min, the time currently needed to acquire one
slice.

Figure 2.19: Spatial coverage requirements for ASL-CMR. (left) 5-chamber view of the heart is shown with desired
prescription of a basal, mid, and apical short-axis slice. (right) Each slice is shown with corresponding segment based
on the AHA 16 segment model. This figure was adapted from Cerqueria et al. (75).
Saturation Steady Pulsed Labeling Introduction
38
3. Saturation Steady Pulsed Labeling
3.1 Introduction
The original myocardial SPASL technique (15) had three notable limitations: 1) Inversion
labeling pulses were used to label blood in the aortic root. However, due to the complex blood
flow patterns in the heart, inversion pulses label blood multiple times before entering the
myocardium, and this leads to complex dependence on spin history. This makes quantification
challenging and can affect labeling efficiency in the aortic root. 2) echo-planar MR imaging and
signal targeting with alternating frequency (EPISTAR) labeling (73) was used to balance
magnetization transfer (MT) effects. EPISTAR balances MT for single-slice studies, but does not
generalize to multi-slice which is ultimately needed for ischemia assessment. 3) High flip angle
fully sampled imaging was used. This can reduce the sensitivity to prepared longitudinal
magnetization due to the partial saturation effects of the imaging pulses.
In this work, we present an implementation of SPASL based on Saturation labeling, to
address key limitations of the previous approach. We demonstrate that saturation labeling pulses,
surprisingly, do not compromise labeling efficiency and provide insensitivity to spin history. We
employ an improved approach to balance magnetization transfer effects (74) that will enable future
multi-slice acquisition. We also experimentally validate the proposed approach in healthy swine,
and experimentally optimize imaging parameters to maximize sensitivity.
3.2 Methods
Experimental Methods
Experiments were performed in adult Yorkshire swine (N=8, 18-22 kg). Prior to CMR
imaging, swine were intubated and sedated. Respiration was controlled using a mechanical
ventilator. Isoflurane (1-5%) was administered to maintain the anesthetic plane throughout the
experiment. The protocol was approved by the Animal Care Committee of Sunny Brook Research
Institute.
Image Acquisition Methods
Experiments were performed on a 3 T scanner (MR750, General Electric Healthcare,
Waukesha, Wisconsin, USA) with an 8-channel cardiac receiver coil. Plethysmograph (PG) gating
was used in all experiments. The precise timing of stable diastole and systole were identified using
Saturation Steady Pulsed Labeling Methods
39
PG-gated CINE scans. End-systole was identified based on closure of the aortic valve; stable
diastole was identified based on the longest stationary period of the LV myocardium, identified on
a SAX cine. Trigger delays were set as ratios of the RR duration to place imaging at the center of
stable diastole and labeling for SPASL was centered at end-systole.

Figure 3.1: Location and timing of ECG-gated myocardial SPASL. Graphic prescription is performed using a three-
chamber cine; (top right) imaging is performed in a mid short-axis slice during mid-diastole (blue block); (top left)
labeling is performed in a slab parallel to the imaging slice at end-systole such that the leading edge covers the entire
proximal aorta and bulb of the aortic valve (green block). Timing diagrams are shown for imaging intervals of 1, 2,
and 3 RR. Labeling is always performed every RR.
ASL-CMR data was acquired using single-gated FAIR (10) and SPASL (15,77). Figure
3.1 shows the sequence and timing diagrams of SPASL methods along with locations of the
imaging slice and labeling slab. FAIR ASL was performed in seven 15 sec breath-holds (BH) in a
total of 4 min (10,12). A fully sampled baseline image, under-sampled baseline image, and a noise
image were acquired in the first BH. Six pairs of control and label images were acquired in
subsequent BHs. SPASL was performed in seven BH and in 4 minutes. Baseline and noise images
were acquired in the first BH. Three control and three label series were acquired in the subsequent
six BH. A 60 mm labeling slab was used for all SPASL experiments unless stated otherwise. MT
effects in SPASL were balanced using PICORE (74). We also acquired proton density weighted
Saturation Steady Pulsed Labeling Methods
40
baseline images using a fast GRE sequence before ASL-CMR exams. These images were used as
estimates of M0.
All ASL acquisitions utilized snapshot balanced steady state free precession imaging
(39,78) with the following parameters: TE/TR: 1.5ms/3.2ms, matrix size: 96 x 96 matrix, 1.88mm-
2.08mm spatial resolution, 18cm-20cm FOV, 1cm slice thickness, 651Hz bandwidth per pixel,
1.6x GRAPPA factor with 24 autocalibration signal lines, 192 ms image acquisition window, and
50° prescribed flip angle unless specified otherwise. M0 images were acquired using a spoiled fast
GRE sequence with TE/TR: 1.2ms/4ms and same matrix size and resolution as bSSFP but without
acceleration.
Reconstruction and Data Analysis
Reconstruction and data analysis were performed in MATLAB (MathWorks Inc., South
Natick, MA). Images were reconstructed using a custom implementation of GRAPPA (79).
Channel images were combined using optimal B1 combination (80). Similar to previous ASL-
CMR studies all images were sinc interpolated by a factor of 4 to facilitate segmentation. Sinc
interpolation was performed by zero-padding k-space before taking the inverse Fourier transform.
Semi-automated segmentation of the left ventricular myocardium was performed using offline
motion correction with advanced normalization tools (81) with previously published settings (82).
This generated masks for global and per-segment (six segment) analysis. The entire left ventricle
myocardium from the slice was used as the region of interest (ROI) for global analysis, and the
AHA six-segment model (76) of a mid-short-axis slice was used for per-segment analysis (83).
In SPASL, the first 5, 4, or 3 images were rejected for each breath-hold as they occur during
transient approach to steady state for imaging intervals of 1RR, 2RR, or 3RR, respectively. ASL
signal (S) was calculated as the difference between control and label image signal normalized by
M0. PN for FAIR was calculated as described by Zun et al. (10), whereas for SPASL PN was
calculated using the equation below:
𝜎 𝑠𝑝𝑎𝑠𝑙 =
√
𝜎 𝑐 2
+ 𝜎 𝑙 2

𝑁 ,
where 𝜎 𝑐 and 𝜎 𝑙 is the standard deviation of control and label signal, respectively and N is the
number of control and label images after rejecting images during transient approach to pseudo-
steady state.
Saturation Steady Pulsed Labeling Methods
41
Statistical equivalence of measurements between different ASL-CMR techniques was
established using a two one-sided test (TOST) (84). Statistical difference was established using a
paired T-test. Segments were visually rejected on a per-animal basis during analysis due to
banding and thinning of the myocardium. Banding can result in over or underestimation of ASL
signal and can increase PN. Segments with thin myocardium were rejected to avoid corruption of
ASL signal due to partial voluming from the LV blood pool, which may artificially increase ASL
signal and PN.
Inversion versus Saturation SPASL
Labeling efficiency for inversion SPASL was simulated as a function of heart rate, ejection
fraction, and incidental labeling of left-atrial blood pool. Labeling efficiency of saturation and
inversion SPASL were compared in simulation. ASL signal and PN in both SPASL techniques
were compared with FAIR ASL in 8 healthy swine. Two pigs were scanned twice resulting in 10
datasets.

Figure 3.2: Labeling pulse performance. (left) simulated labeling efficiency. The red box indicates b1 (0.6-1.2) and
∆f (±150Hz) values observed in swine and human CMR at 3 Tesla (85). The inversion pulse provides inversion
efficiency >75% and saturation pulse provides saturation efficiency >75% in this region. (center) simulated slab
profile with 60mm thickness. The saturation pulse is designed to have less than 0.5% tipping of the imaging slice i.e.
25mm from the edge of the labeling slab (red dotted lines, red arrows). (right) representative maps of in-vivo labeling
efficiency. We measured >75% inversion efficiency and >85% saturation efficiency in the aortic root in all swine in
this study.
Inversion labeling was performed with adiabatic (hyperbolic secant) pulses because of their
insensitivity to B0 and B1variation (86). Saturation labeling was achieved using a saturation pulse
Saturation Steady Pulsed Labeling Results
42
train of three pulses. The pulse train was optimized using Bloch simulations to maximize labeling
efficiency with less than 0.5% saturation 2.5cm from the edge of the labeling slab for B1 and B0
variation of 0.5-1.2, and ±150 Hz, respectively. FAs of 80° was used for all three pulses in the
saturation pulse train. B1 and B0 variations were measured from B0 and B1 maps. Both pulses
and their profiles are shown in Figure 3.2.
Optimization of Imaging Interval and Flip Angle
Five healthy swine were scanned with saturation SPASL sequence with imaging FAs of
10° to 70°, 20° to 70°, and 20° to 70° for imaging intervals of 1RR, 2RR, and 3RR. The sequence
timing diagram for imaging intervals of 1-3RR is shown in Figure 1. Only two series each of
control and label images were acquired, instead of three, to accommodate this experiment within
our protocol. This resulted in four BH per setting instead of six BH in the previous experiment.
Data acquisition of different imaging intervals was interleaved to compensate for drift in perfusion
and subsequently ASL signal due to hemodynamic changes during the long experiment. We
acquired one pair of control and label image series at all FAs for each imaging interval followed
by the second pair. The two pairs of data for each setting were then combined and average ASL
signal and PN was used for comparison. Bloch simulations with flow were also performed to
estimate the maximum ASL signal for FAs of 0° to 100° and imaging intervals of 1RR to 10 RR.
Verification of Inflow Signal
Saturation SPASL experiments were also performed with a labeling slab of 5 mm in five
of the eight animals. The proximal edge of the labeling slab was matched with previous
experiments. ASL signal and PN obtained from this experiment were compared to the saturation
SPASL experiment with a 60 mm slab.
3.3 Results
Figure 3.3 shows representative control and label images. Qualitatively image quality is
comparable to previous ASL-CMR studies (12,15). CNR for both control and label images was
highest in FAIR. As expected, CNR was comparable for control images in SPASL and was slightly
higher for label images in inversion SPASL. bSSFP related off resonance artifacts were observed
in the inferior and/or lateral wall of some animals, resulting in rejection of sixteen out of sixty
segments. Fourteen segments were rejected due to banding, all inferior or lateral wall segments,
which are known to be the most susceptible to banding due to close proximity to draining veins
Saturation Steady Pulsed Labeling Results
43
(87). Two septal segments were rejected due to spurious appearance of thinning that could have
result in corruption of ASL signal due to partial volume effects.

Figure 3.3: Representative image quality in healthy swine at rest. Image quality for all settings was comparable to
previous published studies. Compared to FAIR, SPASL has reduced poor blood-myocardium contrast in label images
(bottom row), which makes myocardial segmentation and registration challenging. FAIR parameters: PLD 1RR, flip
angle 50°; SPASL parameters: 1RR, flip angle of 50°.
Figure 3.4 compares numerical simulations of the labeling efficiency of inversion-SPASL
and saturation-SPASL as a function of HR, ejection-fraction (EF) and incidental labeled of left-
atrial (ILLA) blood. In healthy pigs, HR was approximately 90 bpm and EF was assumed to be
60% based on previous literature (88). Inversion-SPASL and saturation-SPASL should provide
comparable labeling efficiency in healthy pigs under these physiological conditions based on our
simulations. The reduction in labeling-efficiency for inversion-SPASL is due to incidental labeling
of left-atrial blood. Unlike inversion SPASL, labeling efficiency of saturation SPASL is not a
function of incidental labeling of left-atrial blood, HR, or EF.
Figure 3.5 compares performance of saturation SPASL, inversion SPASL, and FAIR in
terms of ASL signal and PN. Red bars in the box and whisker plots represent median ASL signal
and PN, box edges represent 25th and 75th percentile, and whiskers represent 2.7 times the standard
deviation of data for ASL signal and PN. Bar plots show average ASL signal. Error bars represent
PN, averaged over all animals. ASL signal as percentage of M0 was 0.30%±0.32%, 0.28%±0.44%,
and 0.43%±0.80% for saturation SPASL, inversion SPASL, and FAIR, respectively. ASL signal
in inversion SPASL was statistically equivalent to saturation SPASL using TOST at a signal
difference of 0.10% with p<0.0001. PN as percentage ASL signal was 0.09%±0.13%,
Saturation Steady Pulsed Labeling Results
44
0.10%±0.17%, and 0.27%±0.52% for saturation SPASL, inversion SPASL, and FAIR,
respectively. PN was significantly higher in FAIR compared to both SPASL techniques with
p<0.05.

Figure 3.4: Simulation of SPASL labeling efficiency. Labeling efficiency of (A) inversion SPASL depends on heart
rate (HR), ejection fraction (EF), and incidental labeling of left-atrial (ILLA) blood; and is plotted: (left) as a function
of HR and ILLA, for an ejection fraction of 60%, and (right) as a function of EF and ILLA, for a HR of 90 bpm.
Labeling efficiency for (B) saturation SPASL does not depend on HR, EF, or ILLA. Blue contours in the inversion
SPASL plots represent where labeling efficiency is equivalent to saturation SPASL. The red boxes on the plots
represent the operating point (OP) observed in our animal models (EF 50-65%, and ILLA 80-95%).
Figure 3.6 demonstrates that ASL signal in saturation SPASL is due to labeled inflowing
blood. Median ASL signal is represented by the red bars in the box and whisker plot. Edges of the
box represent 25th and 75th percentile signal, whereas whiskers cover 93.5% of all data. Mean ASL
signal as percentage of M0 averaged over all animals was 0.32%±0.32% and 0.11%±0.28% for
saturation SPASL with labeling slab of 60mm and 5mm, respectively. The difference in ASL
signal was statistically significant with p<0.0001.
Figure 3.7 shows the simulated and in-vivo results for optimization of ASL signal in
saturation SPASL. Simulated ASL signal increases with imaging intervals. It is maximum for FAs
Saturation Steady Pulsed Labeling Discussion
45
of 35°, 45°, and 50° for intervals of 1,2, and 3 RR, respectively. Increasing imaging interval beyond
3RR gives <0.05% increase in ASL signal due to T1 recovery of the labeled blood. Simulated
signal efficiency decreases with increasing imaging intervals and is highest for imaging interval
of 1RR.

Figure 3.5: Comparison of myocardial perfusion measurements in swine, using saturation SPASL, inversion SPASL,
and FAIR. A) Box whisker plot of normalized ASL signal (S/M0). B) Box whisker plot of physiological noise (PN).
Red lines represent the median values; box edges represent 25th and 75th percentile; whiskers represent 2.7 times the
standard deviation. C) Bar plot of (blue) myocardial, and (orange) chest-wall ASL signal (S/M0) averaged across all
animals. Error bars represent ± PN averaged over all animals.*Represents statistical equivalence with
p<0.05.†Represents statistical difference with p<0.05. Statistical comparisons were made between all techniques, but
only showing p-values where statistically significant difference or equivalence was found.
In-vivo ASL signal increases with imaging interval and FA, as expected. ASL signal, as a
percentage of M0 , is maximum for imaging FA of 60° and interval of 3RR at 0.67%±0.64%. The
decrease in signal with increasing FA was not observed in vivo because higher FAs were not
acquired. The highest signal for interval of 1RR was 0.35%±0.54% with FA of 50°. Signal
efficiency for imaging interval of 1RR was maximum at 0.35%±0.54% where as it was higher for
interval of 2RR at 0.42%±0.42% and 3RR at 0.39%±0.37%, with FA of 50°.
3.4 Discussion
We developed saturation SPASL and carefully optimized it for ASL-CMR. We
demonstrate that the ASL signal with saturation SPASL is equivalent to inversion SPASL and
saturation SPASL has 1.75x higher signal efficiency compared to FAIR. We also present an
optimization of imaging flip angles and intervals to maximize signal strength in saturation SPASL.
We demonstrate the trends in ASL signal are comparable between simulations and in-vivo
Saturation Steady Pulsed Labeling Discussion
46
experiments. Our results show that staying within SAR limits we should try to use the maximum
possible prescribed FAs for imaging interval of 1RR or 2RR to maximize signal efficiency.
Figure 3.6: Verification of inflow based SPASL signal.
SPASL measurements were obtained using 60mm and
5mm labeling slabs, with the same leading edge. Inflow
signal is demonstrated by reduction in ASL signal when
using a thinner labeling slab. A) Box whisker plot of
normalized myocardial ASL signal (100% x S/Mo). Red
lines represent median ASL signal; box edges represent
25th and 75th percentile; whiskers cover 93.5% of data
points. B) Bar plot of (blue) myocardial and (orange)
chest-wall ASL signal averaged over all animals. Error
bars represent ± PN averaged over all animals. The
signal is chest wall was small and comparable for both
slab thicknesses, ruling out contribution of MT and
labeling slab profile to the ASL signal. Represents a
statistically significant difference with p<0.0001.
ASL signal in our studies was surprisingly equivalent for both saturation and inversion
SPASL. Inversion pulses are typically used in ASL because they provide twice the labeling
efficiency of saturation labeling. In ASL-CMR, labeling pulses label blood multiple times before
it reaches myocardium due to the complicated movement of blood through the heart. This leads to
a complicated spin history which makes labeling efficiency in inversion SPASL dependent on HR,
EF, and amount of labeled blood in the left atrium. In this work, we showed that for physiological
conditions in healthy pigs it is beneficial to use saturation labeling because it provides equivalent
labeling efficiency to inversion while being insensitive to spin history. This is an important design
consideration for ASL-CMR and distinction from brain ASL. A 2D adiabatic inversion labeling
pulse can potentially be used to improve the labeling efficiency of ASL-CMR (89), and its use will
be explored in future work.
We expected FAIR to have twice the signal compared to saturation labeling because it
benefits from the full dynamic range of an inversion pulse. However, ASL signal in saturation
SPASL was approximately 70% of the signal obtained with FAIR. This 20% increase in signal
was likely due to an accumulation of ASL signal from labeling and imaging every heartbeat in
SPASL. This can be verified by performing an experiment that compares saturation SPASL, FAIR,
Flow sensitive saturation recovery (FASR).
We verified inflow as the source of the ASL signal in saturation SPASL. We found that
the ASL signal was significantly lower when a labeling slab of 5 mm was used compared to a
labeling slab of 60 mm. The small signal observed with the labeling slab of 5mm was likely from
Saturation Steady Pulsed Labeling Discussion
47
the small amount of blood that is labeled in the aortic root, which can contribute to the ASL signal.
The absence of signal in the chest wall for all saturation SPASL experiments also confirms 1)
imaging slices were not labeled with our labeling pulses and 2) inflow was the source of the ASL
signal. Otherwise, we would’ve observed signal in the chest wall.

Figure 3.7: Optimization of saturation SPASL imaging. (top left) Simulated ASL signal and (top right) simulated ASL
signal efficiency for imaging intervals of 1-10 RR. (bottom left) Bar plot shows average ASL signal and (bottom right)
average ASL signal efficiency measured in-vivo for imaging intervals of 1-3 RR and FA of 10-70° prescribed and 20-
50° actual FAs. The in-vivo FA for each segment was rounded up for averaging and visualization. Error bars represent
physiological noise. Legends for each row are shown to the right. The red box in the top row highlights the FA’s
shown in the bottom row.
Another key component in our work was the use of the PICORE labeling scheme instead
of EPISTAR. We found that using EPISTAR labeling scheme in the heart can also limit the
freedom in placement of the labeling slab with respect to the imaging slice. To maintain symmetry
either the labeling slab for control will label in the LV blood pool or the labeling slab for label
images may not be placed on the aortic root. This can cause a loss in labeling efficiency. Although
not shown here, reduction in ASL signal was observed for inversion SPASL with EPISTAR
Saturation Steady Pulsed Labeling Discussion
48
compared to PICORE in 3 healthy swine. Alternatively, PICORE labeling is compatible with
multi-slice ASL and can allow for precise control of labeling slab placement.
Trends in ASL signal for the optimization of imaging interval and FA were consistent
within each imaging interval between simulations and in-vivo experiments. ASL signal increased
with increasing FA and was maximized at a FA for both simulations and in-vivo experiments. We
did not observe the decrease in ASL signal with FA in the experimental data because we were
unable to achieve the actual FAs >60° due to SAR constraints. We expected the signal efficiency
to be maximum for imaging interval of 1RR at FA of 35°. From our experiments we observed
highest signal efficiency with imaging interval of 1RR at around FA of 50°. Imaging intervals of
2RR and 3RR also had ASL signal and signal efficiency maximized at FAs of 50°. ASL signal and
efficiency were maximized at 10° higher FAs for in-vivo experiments compared to simulations
likely due to the rounding up of FAs that was done to group data for visualization and analysis.
Imaging interval of 2RR and 3RR had higher signal efficiencies than imaging interval of
1RR which was not consistent with simulations. But the difference between maximum ASL signal
efficiency for intervals of 1RR and 2RR was only 0.07%, which was not significant. Since at higher
imaging intervals PN is also higher, lower imaging intervals should be used to improve TSNR.
Simply put using actual FA of 50° (prescribed FA of 60°) with imaging interval of 1RR maximizes
signal efficiency in-vivo, where as if SAR is an issue then imaging interval of 2RR can be used
with an actual FA of 60° (prescribed FA of 70°).
Our study had several limitations. We did not acquire higher FAs for intervals of 2RR and
3RR and were not able to completely validate the signal behavior observed in our simulations. The
labeling slab was fixed to be parallel to the imaging slice for ease of implementation. Future work
will allow for this flexibility, so the imaging slab is parallel to the aortic valve. A relatively large
number of segments had to be rejected due to banding. This is a limitation of cardiac bSSFP
approaches at 3T and is not unique to ASL-CMR. Our saturation labeling used a saturation pulse
train and we had to trade off saturation efficiency for minimizing the effect of labeling pulses on
the imaging slice. This worked for mid short-axis slices that were >3 cm away from the edge of
the labeling slab but even this optimized labeling pulse will not be compatible with multi-slice
imaging. Future work will likely explore saturation labeling pulses with sharper pulse profiles. We
used PG for triggering and while it worked in our set of animals its prone to drift in the gating
signal that can cause timing errors for both labeling and imaging. We had difficulty in getting a
Saturation Steady Pulsed Labeling Conclusion
49
good ECG signal in our animals so had to rely on PG but human studies will use ECG gating which
can reduce timing errors and potentially lower PN. Lastly, this was a long study where animals
were anaesthetized for a long duration. This may have caused changes in hemodynamics whose
effect on perfusion are not completely known and could’ve cause errors in our measurements.
3.5 Conclusion
We have demonstrated the feasibility of using saturation SPASL for ASL-CMR with
PICORE labeling scheme. We compared saturation SPASL to existing ASL-CMR techniques in
healthy swine. Saturation SPASL improves the signal efficiency and sensitivity of ASL-CMR by
1.75x and 2x, respectively compared to FAIR. It gives comparable performance to inversion
SPASL in terms of signal efficiency and sensitivity but is not affected by spin history which can
cause errors in quantification. We have also demonstrated that using an actual FA of 50° and
imaging interval of 1RR gives the highest signal efficiency which is comparable to imaging
interval of 2RR at FA of 50° and imaging interval of 3RR at FA of 50°.

Single Shot Echo Planar Imaging Introduction
50
4. Single Shot Echo Planar Imaging
4.1 Introduction
Current ASL-CMR stress test protocols provide only single-slice coverage which is
insufficient for clinical imaging (90). The American Heart Association guidelines recommend at
least three short axis slices to assess all coronary artery territories (91). Spatial coverage can be
increased using either sequential or simultaneous multi-slice imaging. Currently, both of these
techniques are not feasible with snapshot balanced steady state free precession (bSSFP) imaging,
the most common ASL-CMR acquisition. Sequential multi-slice imaging with bSSFP would
require an imaging window of ~600 ms, which is much longer than stable diastole. Simultaneous
multi-slice (SMS) imaging with bSSFP imaging has many unresolved artifacts due to off-
resonance and imprecise excitation. These artifacts can corrupt the ASL signal (92,93).
In this work, we revisit single-shot echo-planar imaging (SS-EPI) for ASL-CMR. SS-EPI
was the earliest imaging scheme proposed for ASL-CMR at 1.5T by Poncelet et al. (9) and remains
a promising alternative to snapshot bSSFP. With modern high-performance gradients, it has the
potential to reduce the imaging window (from ~200ms) to ~55ms without compromising spatial
resolution. The short imaging window can potentially enable sequential multi-slice imaging to
improve spatial coverage. In recent years, SS-EPI has been successfully applied to cardiac
diffusion tensor imaging, which, like ASL, is a low SNR and highly motion-sensitive technique
(46,94). In cardiac diffusion tensor imaging SS-EPI achieved in-plane resolution of 2.7 mm x 2.7
mm for both single slice and SMS systolic imaging with inner-volume excitation (95), readout
durations of 11-14ms and using parallel imaging with partial k-space. In this study, we make three
modifications to these published SS-EPI implementations that are relevant to ASL. First, ASL
benefits from sequential multi-slice imaging which requires a change in the implementation of
inner-volume imaging. Second, ASL is sensitive to changes in myocardial signal intensity,
requiring systematically study of the velocity-selective effects of crusher gradients on both
myocardial and blood pool signal. Third, ASL requires higher in-plane spatial resolution to
minimize partial voluming effects from the blood pool that can bias MP measurements.
In this work, we present a practical implementation SS-EPI tailored to the needs of ASL.
We achieve a reduced field of view (FOV) using with 2DRF excitation, in a way that is compatible
with sequential multi-slice imaging. We analyze the velocity-selective effect specific to cardiac
Single Shot Echo Planar Imaging Methods
51
imaging and determine the SS-EPI settings that maximizes temporal SNR (TSNR) for both
systolic and diastolic imaging. Optimized SS-EPI is then compared with bSSFP for single-slice
ASL-CMR. We also demonstrate sequential multi-slice ASL-CMR with no increase in scan time.
4.2 Methods
Pulse Sequence
A product EPI pulse sequence (GE Healthcare, Waukesha, WI, USA) was modified to
include a novel two-dimensional radiofrequency (2DRF) excitation that achieves reduced field of
view (rFOV) SS-EPI (30,31), as shown in Figure 4.1. The 2DRF was designed with a fly-back
echo planar trajectory in excitation k-space to avoid Nyquist ghosts along the slice-select axis due
to gradient and timing imperfections. Oscillating and blipped gradients were played on the phase-
encoding axis and slice-select axis, respectively. Selection along the phase-encode direction is
controlled by the sub-pulses. Each sub-pulse was designed using Shinnar Le-Roux (SLR)
algorithm (25) with a time bandwidth product (TBW) of 10, 1% passband ripple, 1% stopband
ripple, 8 cm passband for a 14 cm phase-encode FOV, and 35 dB attenuation outside the phase-
encode FOV. Selection along the slice-select direction is controlled by the envelope by which the
23 sub pulses are scaled. This envelope was also designed using SLR with TBW of 2.2. The sub
pulse spacing was chosen to shift the fat excitation by at least 30 mm. This fat signal was
Figure 4.1: SS-EPI Pulse Sequence (A) is comprised of a 90° spectral-spatial 2DRF excitation pulse (green), 1D
selective 180° spin echo pulse (pink) with crusher gradients that provide blood suppression, and a single-shot partial
Fourier echo-planar readout (gray). The 2DRF pulse (B) contains 23 subpulses. Selection along the phase-encode
direction is controlled by the subpulse shape and Gy gradient (oscillating). A fly-back trajectory was used to avoid
ghosting due to gradient and timing imperfections. Selection along slice direction is controlled by the RF envelope
and Gz gradient (blipped). The individual subpulses and the envelope were both designed using Shinnar Le-Roux (19).
Single Shot Echo Planar Imaging Methods
52
subsequently suppressed by only refocusing a 20 mm slice. The overall pulse duration was
minimized by 1) using maximum gradient slew rate for fly-back gradients, and 2) using the
variable rate selective excitation (VERSE) algorithm (28,96) and playing RF during gradient
ramps. These specifications were heuristically optimized, and in our experience, provide consistent
image quality.

Figure 4.2: Simulated water and lipid signal profiles overlaid on a diastolic 4-chamber image to scale. (Left) Water
signal for 2DRF pulse. (Center) Lipid signal for 2DRF pulse. (Right) Water and lipid combined signal after the slab-
selective 180° refocusing pulse. Red lines represent the FWHM of the 180° refocusing pulse. Simulated signal
amplitude (S/Mo) is depicted in color. Lipid excitation is shifted by ~3.5cm relative to water, moving it outside the
passband of the refocusing pulse. Replicas for both water and lipid signal are 10cm apart in the slice encoding
direction. Only the water signal at the desired slice location is refocused by the 180° pulse.
Figure 4.2 illustrates the simulated RF pulse profiles for excitation and refocusing. To
enable sequential multi-slice imaging slice replicas were kept 10 cm away which ensures that only
one slice within the LV will be excited with each excitation. Based on simulation, the pulse profile
in the phase-encode direction has an 8 cm passband and <1% signal in the stopband. Fat excitation
is shifted by 35 mm and not included in the refocusing slab.
Optimization of Velocity Suppression
The proposed SS-EPI sequence provides inherent flow suppression. Specifically, the
crusher gradients around the 180-SE pulse, shown in Figure 4.1, create velocity-dependent phase.
When intravoxel velocity is heterogenous, this produces saturation. This mechanism has been
exploited in velocity-selective ASL (37) and motion-sensitized driven equilibrium (MSDE) flow
suppression pulses (38). A common approximation is that uniform velocity distributions [0, Vmax]
with peak velocities Vmax above 𝑉 𝑐 = 𝜋 /2𝛾 𝑀 1
are suppressed, where gamma is the gyromagnetic
ratio and 𝑀 1
is the 1st moment of crusher gradients. A low Vc is expected to suppress the LV blood
pool signal, which reduces flow artifacts, and partial voluming effects for myocardial imaging.
Single Shot Echo Planar Imaging Methods
53
However, this can potentially reduce the myocardial signal. The tradeoff between myocardial
signal loss and blood suppression was experimentally characterized in 4 volunteers for both
systolic and diastolic cardiac phases, leading to a Vc of 25cm/s in systole and 12cm/s in Diastole.
Details can be found in the Supporting Information.
Evaluation of Geometric Distortion
SS-EPI is susceptible to geometric distortion artifacts along the phase-encode direction due
to off-resonance (97). This presents a well-known challenge in myocardial imaging at 3T (98).
These artifacts can degrade image quality and cause signal loss especially in the lateral and inferior
LV walls due to proximity to the lungs and draining coronary veins (87). The artifact can persist
elsewhere if there are implanted metal such as valve clips (99). Rotating the scan plane will change
the direction of geometric distortion and improve image quality. Four to six scan plane rotations
(30° increment) were acquired for each subject. The final rotation angle was selected by the scan
operator by visual inspection, based on uniformity of LV myocardium thickness, and on least
amount of signal loss. This process was repeated for both systole and diastole to account for
potential differences in off-resonance with cardiac phase.
ASL Data Collection
Double gated flow alternating inversion recovery (DG-FAIR) ASL (9,11) was performed
in short-axis slices, shown in Figure 4.3. Adiabatic hyperbolic-secant pulses were used for
labeling. SS-EPI and bSSFP were used for imaging, in separate scans. Labeling and imaging were
performed during end-systole for systolic scans and mid-diastole for diastolic scans. A
plethysmograph gated cine bSSFP scan was used to visually determine stable systolic and diastolic
phases.
DG-FAIR experiments were performed with post labeling delay (PLD) of 1-RR and 2-RR
(RR is the duration of one cardiac cycle). PLD was defined as the duration from center of labeling
to center of the imaging window, such that measured MP reflects average perfusion during PLD
period, as shown in Figure 4.1. This enables pairwise comparisons with one independent variable,
for example 1RR bSSFP can be directly compared with 2RR bSSFP, and 2RR bSSFP can be
directly compared with 2RR SS-EPI. SS-EPI benefits from post-labeling delays >1RR because at
1RR there is inadequate contrast in control and label images to perform LV segmentation. Direct
subtraction of control and label images is not feasible in human ASL-CMR due to variations in
breath hold position. Therefore some image contrast is required to facilitate image-to-image
Single Shot Echo Planar Imaging Methods
54
registration. In SS-EPI with PLD of 1RR, both blood pool (due to velocity selective suppression)
and myocardium are near zero and difficult to distinguish. SS-EPI requires a PLD > 1RR to
provide blood-myocardium contrast sufficient for reliable LV segmentation and/or registration of
control and label images for ASL analysis.
SS-EPI acquisition parameters: FOV 28x14 cm, TE/TR 31.3 ms /55ms, flip angle 90º,
matrix size 128x64, partial Fourier factor 5/8th, and readout time= 25 ms. bSSFP acquisition
parameters: FOV 18-24 cm, slice thickness 1cm, TE/TR 1.3ms/3.1ms, flip angle 50º, matrix size
96x96, GRAPPA with 24 auto-calibration lines, and effective acceleration R=1.6.

Figure 4.3: Graphic prescription and sequence timing. (Bottom) timing of diastolic DG-FAIR ASL with post labeling
delay of 1RR, and 2RR. (Top) graphic prescription performed using a four-chamber cine. The imaging slices (purple)
are short axis slices at mid-diastole. Images were acquired at mid-short axis for single slice experiments and at basal,
mid, and apical short axis for diastolic multi-slice experiments. The labeling slab (green) for control labeling covers
the imaging slices for both single and multi-slice experiments. Labeling slab of 30 mm was used for single slice and
70 mm was used for diastolic multi-slice experiments. Prescription for systolic experiments was similar, but was done
on an end-systole 4-chamber view. For systolic multi-slice experiments, only mid, and apical short axis slices were
acquired, using a 50 mm labeling slab.
Six-pairs of control and labelled images were acquired for each DG-FAIR scan using slice
selective and non-selective labeling, respectively. Each pair of images was acquired during a 10-
12s BH to avoid mis-registration. A delay of 6 seconds between each image acquisition allowed
complete T1 recovery. A separate baseline image (i.e. an image without any preparation pulse),
Single Shot Echo Planar Imaging Methods
55
noise image, and an image with minimum possible PLD was also acquired. These images were
used to estimate coil sensitivity, noise covariance, and for quantification, respectively. Sensitivity
maps for SS-EPI were acquired using the vendor’s calibration scan.
All experiments were performed on a clinical 3T scanner (GE Signa Excite HDxT,
Waukesha, WI, USA) with 8-channel cardiac receiver coil. Four healthy volunteers (1M/3F Age:
27-36) were recruited for this study. The imaging protocol was approved by our institutions review
board and informed consent was obtained from all the volunteers. In all subjects, we performed
single-slice DG-FAIR SS-EPI (PLD 2RR) and single-slice DG-FAIR bSSFP (PLD: 1RR and
2RR), during both stable systole and diastole. We also performed sequential multi-slice DG-FAIR
SS-EPI. Basal, mid, and apical short axis (SAX) slices were acquired for diastolic imaging. Mid
and apical SAX slices were acquired for systolic imaging. The order of slice acquisition was base
to apex. The labeling slab thickness was 30mm, 50mm and 70mm for the single, multi-slice
systolic and multi-slice diastolic experiments, respectively.
ASL Data Analysis
Standard SS-EPI distortion correction and reconstruction were performed using the
Orchestra reconstruction toolbox (GE Healthcare, Waukesha, WI, USA). This included linear
phase correction and gridding of ramp-sampled data. bSSFP images were reconstructed using a
custom implementation of GRAPPA(79). In both settings, channel images were combined using
optimal B1 combination (80). Per pixel SNR maps were computed using the pseudo-replica method
(100). One pair of control and label images were manually segmented and the masks were
propagated to all other images in the series using image registration (82) implemented using
Advanced Normalization Tools (ANTs) (101). Spatial filtering was performed to calculate
regional signal intensities which were then used for MP quantification (83). MP for both single
slice and multi slice data was calculated in the same way.
MP was calculated using the interpolated signal difference (ΔM) between control and label
T1 curves as described by Poncelet et al. (9) and Do et al. (11). T1 curves for control and label
images were fitted using established three parameter model for T1 mapping (102). ΔM between
control and label T1 curves was estimated at the average inversion time (TIavg) of control and label
images. MP was then calculated using the following equation:
f =
ΔM
αM
0
TI
avg
e
−TI
avg
/T1
blood

Single Shot Echo Planar Imaging Results
56
derived from Buxton’s general kinetic model (103), where α is inversion efficiency, M0 is
equilibrium magnetization, f is myocardial perfusion, TIavg is mean inversion time of control and
label images, and T1blood is the T1 of arterial blood (1950 ms at 3T based on Weingartner et al.
(104)). α was estimated with an image acquired immediately after a non-selective inversion (TI of
~160ms) and a baseline image (representing complete recovery, TI of 9500ms). In multi-slice
studies, the slice specific TIavg was used during MP quantification. Mis-triggered images were
visually identified and rejected. Physiological noise was calculated as the standard error in flow
measurements as described by Poncelet et al (9). MP measurement quality was evaluated using
temporal SNR (TSNR, MP divided PN).
Segmental analysis was performed in the mid-short axis slices for single slice experiments,
using the standard AHA model (76). Global analysis was performed for multi-slice experiments.
Statistical equivalence of MP measurements between different ASL-CMR techniques was
established using a two one-sided test (TOST) (84) at a difference of 0.4 ml/g/min. Statistical
differences in TSNR between techniques was established using a paired T-test.
4.3 Results
Data Quality

Figure 4.4: Illustration of SS-EPI geometric distortion from one healthy volunteer, and its dependence on graphic
prescription. Baseline images are shown with different in-plane rotations (RO: readout and PE: phase-encode
directions are labeled). As expected, the geometric distortion stemming from off-resonance is always along the phase-
encode direction. Arrows (red) show most severe geometric distortion in the lateral wall, which is adjacent to the
lungs and draining veins (87). Rotation angle that produced LV myocardium with least amount of signal loss, and
most uniform thickness was selected. Based on these criteria, in the above example rotation angle of 0°was selected.
Figure 4.4 illustrates the impact of prescription angle on geometric distortion in SS-EPI
images. As expected, the geometric distortion stemming from off-resonance was always along the
phase-encode direction. Geometric distortion was most severe in the lateral wall, which is adjacent
to the lungs and draining veins (87). Changing the rotation angle affected the amount of signal
loss, and uniformity of the myocardium. In this example, rotation angle of 0° was selected because
Single Shot Echo Planar Imaging Results
57
it provided images with most uniform LV thickness and least amount of signal loss upon visual
inspection. The rotation angle scout scan was completed in under three minutes for all subjects.

Figure 4.5: Representative SS-EPI and bSSFP images from one healthy volunteer: (top) baseline, (center) control,
and (bottom) labelled images, during (left) diastole and (right) systole. The images are cropped to 12cm x 12cm.
Colormap represents pixel values in SNR units. Baseline myocardial SNR is presented as mean ± std across four
subjects. bSSFP images were acquired with 1RR post-labeling delay, as this is the reference method. SS-EPI images
were acquired with 2RR post-labeling delay to provide adequate contrast for LV segmentation. Systolic bSSFP images
show motion artifacts (yellow arrow), likely due to the long acquisition window. SS-EPI images show -20dB diastolic
and -7dB systolic blood suppression, approximately two-fold higher baseline SNR, and stronger blood-myocardium
contrast, compared to bSSFP. SS-EPI images show the typical geometric distortions in a mid-SAX slice due to off-
resonance.
Figure 4.5 contains representative control and label images for SS-EPI and bSSFP. These
illustrate typical image quality, geometric distortion artifacts, and blood suppression. Average
Myocardial SNR for SS-EPI (diastole: 61.1 ± 9.3, and systole: 67 ± 7.9) was higher than bSSFP
(diastole: 34.8 ± 6.8, and systole: 36.1 ± 6.6). bSSFP images acquired during systole showed
motion artifacts (see yellow arrows), likely due to long acquisition window (192ms) which was
greater than stable end-systole. These artifacts were not observed in SS-EPI images. SS-EPI
Single Shot Echo Planar Imaging Results
58
images showed -20dB diastolic and -7dB systolic blood suppression which resulted in higher blood
myocardium contrast than bSSFP images.
Validation

Figure 4.6: Comparison of SS-EPI and bSSFP ASL in mid-short axis slices of four healthy volunteers. A) Illustrates
left ventricular regions of interest. MP and TSNR measurements are summarized for B) global C) septal, and D)
lateral myocadium. Globally, diastolic and systolic SS-EPI were statistically equivalent to diastolic bSSFP 1RR with
P-values of 0.022 and 0.031 using the TOST procedure at a MP difference of 0.4 ml/g/min. In septal regions, MP
measured with diastolic SS-EPI, systolic SS-EPI, and diastolic bSSFP 2RR were equivalent to diastolic bSSFP 1RR
with P-values 0.015,0.012, and 0.019, respectively. In lateral regions, MP in systolic SS-EPI and diastolic bSSFP
1RR were equivalent to diastolic bSSFP 2RR with P-value of 0.004, and 0.043. TSNR in diastolic SS-EPI, systolic
bSSFP 1RR, and bSSFP 2RR was significantly lower compared to diastolic bSSFP 1RR with P-values of 0.017,0.039,
and 0.022. Error bars for MP were calculated by averaging PN across subjects and demonstrate measurement
variability (average intra-subject variability). Error bars for TSNR were calculated as standard deviation in TSNR
across subjects (inter-subject variability).
Figure 4.6 summarizes validation experiments comparing DG-FAIR SS-EPI against
bSSFP. Six out of twenty-four and two out of twenty-four control and label pairs were rejected
Single Shot Echo Planar Imaging Results
59
for diastolic and systolic SS-EPI, respectively, due to mis-triggering. Error bars for MP and TSNR
represent the standard deviation across subjects (e.g. inter-subject variability). These values are
also presented in Table 4.1. Global MP for diastolic SS-EPI (1.67±0.72 ml/g/min) and systolic SS-
EPI (1.50±0.36 ml/g/min) were found to be statistically equivalent to diastolic bSSFP (1RR:
1.59±0.80 ml/g/min) based on TOST with P-values of 0.022 and 0.031, respectively. MP values
for diastolic SS-EPI, systolic SS-EPI and diastolic bSSFP were comparable to previous ASL-CMR
studies (11) with diastolic 1RR bSSFP, but were slightly higher than previous PET (105) and first
pass CMR (106) studies. CMR first pass studies (107) have shown that resting MP in systole and
diastole should be equivalent. In systole, global MP of bSSFP (1RR: 0.83±0.51 ml/g/min, and
2RR: 0.41±0.52 ml/g/min) was lower than SS-EPI (1.50±0.36 ml/g/min).
Global TSNR was highest for diastolic bSSFP (1RR: 13.46±7.98). It was similar in
diastolic bSSFP (2RR: 8.59±3.69), diastolic SS-EPI (7.97±3.62), and systolic SS-EPI (9.59±9.22).
In systole, global TSNR of bSSFP (1RR: 5.54±3.05, and 2RR: 2.45±2.93) was lower than SS-EPI.
Our sample size was too small to establish statistical significance of any differences in global
TSNR.

Global Septal Segments Lateral Segments Segmental
MP TSNR MP TSNR MP TSNR MP TSNR
Diastole
bSSFP 1RR 1.60±0.80 13.46±8.0 1.40±0.94 7.75±5.6 1.61±0.92 11.45±7.4 1.64±0.81 10.0±5.1
bSSFP 2RR 1.21±0.87 8.59±3.69 1.26±0.86 6.79±2.6 1.30±0.80 6.36±1.35 1.13±0.86 5.55±1.65
SS-EPI 2RR 1.66±0.73 7.97±3.62 1.25±0.24 6.24±0.7 0.84±1.33 1.89±1.79 1.53±0.65 4.32±1.38
Systole
bSSFP 1RR 0.82±0.51 5.54±3.05 0.93±0.50 4.81±1.9 0.89±0.64 3.78±2.74 0.71±0.13 3.37±0.97
bSSFP 2RR 0.41±0.53 2.45±2.93 0.74±0.73 5.09±2.5 0.36±0.45 1.51±1.96 0.67±0.46 3.64±1.64
SS-EPI 2RR 1.50±0.36 9.59±9.22 1.31±0.55 7.09±5.9 1.24±0.53 4.63±2.88 1.53±0.41 5.51±3.43
Table 4.1: Myocardial perfusion (MP) and temporal signal-to-noise ratio (TSNR) in four healthy volunteers. Values
for global, septal and lateral segments are presented as mean ± standard deviation across all subjects (inter-subject
variability). Mean segmental values are presented as average over all subjects and segments, whereas segmental
standard deviation is the standard deviation over six-segments within each subject averaged over all subjects (intra-
subject variability).
Figure 4.6B and Figure 4.6C show MP and TSNR measured in the septal and lateral
myocardium. MP measured in the septal myocardium for SS-EPI systole (1.31±0.55 ml/g/min)
and diastole (1.25±0.24 ml/g/min) was statistically equivalent to 1RR diastolic bSSFP (1.40±0.94)
based on TOST with P-values of 0.0012 and 0.0015, respectively. Septal TSNR was similar for
Single Shot Echo Planar Imaging Results
60
systolic SS-EPI (7.09±5.93), diastolic SS-EPI (6.24±0.74) and diastolic bSSFP 1RR (7.75±5.68),
and 2RR (6.79±2.6). MP in the lateral myocardium for SS-EPI systole (1.24±0.53 ml/g/min) was
statistically equivalent to diastolic bSSFP (2RR: 1.30±0.80 ml/g/min) based on TOST with a P-
value of 0.004. Surprisingly, TSNR of diastolic 1RR bSSFP (11.45±7.41) was significantly higher
than diastolic SS-EPI (1.89±1.79) based on t-test with a P-value of 0.017. Diastolic 1RR bSSFP
had higher TSNR compared to systolic bSSFP 1RR (3.78±2.74), and systolic bSSFP 2RR
(1.51±1.96), based on t-tests with P-values of 0.039, and 0.022, respectively.
Multi-Slice Demonstration
Figure 4.7 contains sequential multi-slice SS-EPI images for three subjects. These were all
acquired in the same scan duration as a single-slice acquisition. Basal, mid, and apical mid-SAX
slices are shown for diastole whereas mid and apical slices are shown for systole. Image quality,
blood suppression, and contrast between blood pool and myocardium was visually comparable to
single slice SS-EPI images. Geometric distortion in the apical slice during diastole was severe and
resulted in lower image quality. This was not an issue for apical slices acquired during systole.

Figure 4.7: : Baseline multi-slice SS-EPI images from three healthy volunteers. Basal, Mid, and Apical slices are
shown for diastole. Mid and apical slices are shown for systole. The images are cropped to 12cm x 12cm. The SS-EPI
velocity cutoff was 25cm/s for systolic and 12cm/s for diastolic images. Similar to single-slice SS-EPI, multi-slice SS-
EPI images show -20dB diastolic and -7dB systolic blood suppression, higher baseline SNR, and high blood-
myocardium contrast. Geometric distortion in apical slice during diastole was severe and resulted in poor image
quality. Pixel intensities are shown in SNR units and were calculated using the pseudo-replica method.
Single Shot Echo Planar Imaging Discussion
61
Figure 4.8 shows global MP for sequential multi-slice SS-EPI. Global MP for sequential
multi-slice experiments was 1.64±0.47, 1.34±0.29, and 1.88±0.58 for basal, mid, and apical slices,
respectively, during diastole. Global MP was 1.61±0.35, and 1.66±0.49 for mid, and apical slices,
respectively, during systole. MP measured in all slices during both diastole and systole was
comparable to previous ASL-CMR (108) and PET studies (106). MP measured during systole,
however, was less variable across slices compared to MP measured during diastole. The variability
across slices in estimated MP during diastole was within the ASL measurement variability.

Figure 4.8: Demonstration of sequential multi-slice ASL-CMR using SS-EPI. Bar plots of global MP in basal (blue),
mid (orange), and apical (yellow) slices for three volunteers are shown (left) for each subject, and (right) averaged.
(top) Systolic global MP is shown for mid, and apical slices. (bottom) Diastolic global MP is shown for basal, mid,
and apical slices. Error bars represent global PN per slice (left) global PN in each subject and (right) global PN
averaged over all subjects.
4.4 Discussion
We present a careful analysis of SS-EPI for ASL-CMR and optimize its settings for 3T
human myocardial perfusion imaging. We demonstrate that it provides equivalent measurements
of myocardial perfusion compared to bSSFP, which is the current reference standard. We also
demonstrate that its speed (~55ms per image) enables sequential multi-slice imaging, which is a
Single Shot Echo Planar Imaging Discussion
62
major step towards achieving whole-heart coverage with the duration of standard adenosine
vasodilation (approximately 3 minutes).
Magnetization transfer (MT) effects of the proposed 2DRF SS EPI sequence were
comparable to a standard SS-EPI spin echo sequence. The simulated MTR was 0.1%, 2.76% and
5.46% for slices 1, 2 and 3, respectively (109). The simulated MTR for slice 1 would result in a
negligible loss in sensitivity and can be ignored. The effects for slices 2 and 3 are quite small
compared to other sources of variability in ASL-CMR. Because the signal reduction will be
identical for both control and label images, we expect this to slightly lower sensitivity (<6%)
without introducing bias in MP measurements.
One limitation of the proposed 2DRF pulse is its sensitivity to off-resonance. The
bandwidth (BW) of the pulse in the slice-select direction was only 120Hz. Shifts in the excitation
profile could result in signal loss in the LV myocardium because only the signal in pass-band of
the 180-refocusing (BW ~1KHz) pulse is refocused. The lateral wall of the myocardium typically
experiences the highest off-resonance due to proximity to lungs and draining veins (87). It is
therefore, most susceptible to signal loss from shifts in excitation profile. In this study, we
mitigated this issue by setting the thickness of the 180°-refocusing pulse to 2x the imaging slice
thickness. This provided tolerance to ± 62.5Hz off resonance before experiencing any signal loss.
The slice thickness of the refocusing pulse was not increased further to prevent refocusing fat
signal that could also degrade image quality. This tradeoff in slice thickness of 180° pulse, off
resonance sensitivity, and fat suppression warrants further study.
Another important feature of this study was the careful assessment of velocity selective
saturation effects of the crusher gradients, specifically on myocardium and blood pool signal. SS-
EPI has previously been demonstrated for cardiac DTI (94) where the large motion sensitizing
gradients provided blood pool suppression. Previous works have anecdotally demonstrated blood
pool suppression using crusher gradients (9), but did not systematically study the tradeoff between
undesirable myocardial signal loss and desirable blood pool signal suppression based on the choice
of Vc. We experimentally characterized myocardial signal loss in systolic and diastolic cardiac
phases, and systematically arrived at the velocity cutoffs of 12.5 cm/s for diastole and 25 cm/s for
systole. This study was performed in healthy volunteers and baseline images. Such a study may
need to be carried on a larger cohort, including patients, for generality.
Single Shot Echo Planar Imaging Discussion
63
We demonstrate sequential multi-slice SS-EPI. A natural extension would be SMS SS-
EPI, because of its higher SNR efficiency and even shorter acquisition duration. Previous works
have demonstrated cardiac DTI imaging with SMS SS-EPI (94). Lau et al. used an acceleration
factor of 3 (R=3) to acquire slices simultaneously. We chose sequential multi-slice for an initial
demonstration because it is compatible with all available cardiac coil geometries and does not
require parallel imaging.
Systolic SS-EPI provided statistically equivalent MP measurements to reference diastolic
1RR bSSFP method whereas MP was underestimated for systolic bSSFP. This underestimation in
MP during systolic bSSFP may be due to motion artifacts, as shown in Figure 4.4. The imaging
window in systolic bSSFP (~192ms) is longer than stable end-systole (80-100ms @ HR 60bpm),
which leads to us to expect substantial cardiac motion during the imaging window. SS-EPI has a
much shorter imaging window (55ms) with a readout duration of <25 ms and is therefore, less
susceptible to motion artifacts during systole or in subjects with higher HR.
We did not find MP in diastolic 2RR bSSFP to be equivalent to 1RR bSSFP at the
difference level of 0.4 ml/g/min. Diastolic 1RR bSSFP was used as the reference method and 2RR
bSSFP was acquired to perform a comparison with 2RR SS-EPI with one degree of variation. MP
measured using diastolic and systolic SS-EPI was statistically equivalent to the reference method.
MP measured using diastolic 2RR bSSFP was underestimated by approximately 20%. Bloch
simulations with flow effects (not shown) suggest that an increase in PLD from 1RR to 2RR leads
to roughly 5% underestimation of MP, due to partial labeling of the LV blood pool during control
acquisitions that eventually reaches the target tissue. This may partially explain the observed
underestimation.
We observed a difference in estimated MP in lateral and septal segments using diastolic
SS-EPI. Diastolic SS-EPI yielded MP values equivalent to both diastolic 1RR and 2RR bSSFP for
the septal myocardium. In the lateral myocardium, however, MP was underestimated for diastolic
SS-EPI compared to diastolic bSSFP. This could be due to signal loss in the lateral myocardium.
Lateral myocardium is more susceptible to signal loss due to 1) saturation effects of flow-
suppressing crusher gradients because it moves at higher velocity than septal myocardium
(110,111), 2) off-resonance sensitivity of the 2DRF excitation profile because it has higher off-
resonance due to interface with lungs and draining veins (87), and 3) lower B1 which causes lower
signal due to both imperfect excitation and refocusing. Higher off resonance in the lateral wall also
Single Shot Echo Planar Imaging Discussion
64
causes geometric distortion which can cause signal loss. MP in diastolic bSSFP and systolic SS-
EPI was equivalent using TOST even in the lateral wall, likely due to increased number of
myocardial pixels which allow us to gain back some SNR by averaging.
SS-EPI images showed 2-3 fold variation in SNR across the myocardium where as bSSFP
images showed only 1.5-2 fold variation. SNR variation in SS-EPI was higher likely due to 1) pile-
up artifact from off-resonance which can cause local signal enhancement, 2) B1+ sensitivity of
both the excitation and refocusing RF pulses that can amplify the signal variation due to B1+, and
3) Off-resonance sensitivity of the 2DRF pulse which can cause signal loss. Pile up artifact from
off-resonance can be reduced using parallel imaging. Composite pulses can be used to reduce the
sensitivity of refocusing pulses to B1+ variations. These will be explored in future work to improve
the image quality of SS-EPI at 3T.
MP measured for sequential multi-slice experiments showed very high variability for the
apical slice acquired during diastole. This was likely due to poor image quality observed in
diastolic apical slice due to poor blood pool suppression, and higher off-resonance (87). We
hypothesize that less blood is suppressed in the apical slice blood pool due to lower blood flow
velocities (112), and this blood pool signal can cause flow artifacts in SS-EPI images. This was
not an issue for systolic imaging because as shown in Figure 4.4, during end-systole blood pool is
very small and myocardium is very thick for the apical slice. We did not experimentally evaluate
the effect of slice acquisition order on image quality and MP quantification, and the apical slice
was always acquired last. Therefore, slice acquisition order cannot be ruled out as a potential
reason for high variability in the apical slice.
The MP quantification model used for myocardial ASL assumes constant flow during the
cardiac cycle which is an acceptable assumption when PLD is multiples of the RR interval. It
yields average MP during the cardiac cycle. In sequential multi-slice ASL, PLD is a multiple of
RR ±55ms. The assumption of constant flow can cause a slight bias in MP quantification. The
effect of this change is minimized by using a slice specific PLD. The residual error of the constant
flow assumption is <5% which is small compared to other sources of variability in ASL-CMR
(~10-30%).
In multi-slice studies, we did not observe lower MP in slices that were further from the
leading edge of the labeling slab in sequential multi-slice imaging. It has previously been
hypothesized that transit delay sensitivity of FAIR ASL-CMR makes it incompatible with multi-
Single Shot Echo Planar Imaging Conclusions
65
slice imaging (78). For both imaging during systole and diastole this reduction in MP due to transit
delay was not observed. This is likely due to the longer inversion time (2RR) used for SS-EPI
ASL-CMR experiments, and the fact that all subject were healthy.
Additionally, in multi-slice studies wide slice selective inversion slabs (70 mm and 50 mm)
were used in the control acquisition. These will label large amounts of blood in the LV blood pool.
This blood then perfuses the myocardium for the next few cardiac cycles (because ejection fraction
is < 100%), and will result in underestimation of MP. We performed Bloch simulations with flow
effects (not shown for brevity), including the effect of partial labeling of LV blood pool.
Underestimation is roughly 20% and 10% for diastolic and systolic imaging (PLD: 2RR),
respectively. This underestimation is a significant limitation of the FAIR technique. This effect
was not detected in the data, likely because variation in MP observed between subjects is >30% ,
which is large but consistent with previous ASL-CMR studies (11,113). Larger sample sizes will
be required to detect differences on the order of <30 % with existing ASL-CMR techniques.
Our study had several limitations: 1 – A small number of healthy volunteers were studied,
and further validation in a larger population including patients will be valuable. 2– Segmental
analysis to understand regional variability of multi-slice MP measurements was not performed in
this study. Such analysis is very important for clinical use, and an in depth study of regional
measurements including their sensitivity to motion and partial voluming will be valuable. 3–
Reference acquisitions (DG-FAIR bSSFP) for multi-slice experiments were not acquired.
Acquisition of such reference data would have made this experiment prohibitively long (several
hours) for human volunteers. This would benefit from validation in large animals using reference
CMR methods and/or gold standard measurements such as microsphere. 4 – Sensitivity of SS-EPI
ASL-CMR measurements to Vc was not explored. We only explored the effect of Vc on SNR of
LV myocardium and blood pool in baseline images. ASL-CMR is sensitive to <1% changes in
image signal and future would should explore the effect of Vc on MP and TSNR. 5 – Roughly 18%
of data was rejected due to mis-triggering, which we partially attribute to the use of
plethysmograph. Electrocardiogram (ECG) triggering could be used in future works to overcome
this limitation.
4.5 Conclusions
SS-EPI is an efficient acquisition scheme for ASL-CMR, with many benefits compared to
bSSFP, including a shorter acquisition window and a simple multi-slice implementation. We have
Single Shot Echo Planar Imaging Supporting Information
66
demonstrated SS-EPI with carefully optimized settings for human myocardial ASL-CMR at 3T
during both systole and diastole. Single slice MP measured using SS-EPI was statistically
equivalent to diastolic bSSFP, which is the current reference method. We have also demonstrated
sequential multi-slice imaging using SS-EPI for ASL-CMR without increasing in scan time.
4.6 Supporting Information
Evaluation of velocity selective saturation effects
Three healthy subjects (2M/1F,Age: 23-26) were scanned using the rFOV SS-EPI
sequence, one subject was scanned twice on separate days, resulting in four datasets. TE was fixed
at 51.3ms based on the largest crusher gradients. Eleven images were acquired at each of 14
kv values of 0.0125–0.5s/cm during systole and diastole, for a total of 308 images.
Data was converted to SNR units for analysis (114). Images for all kv were aligned using
advanced normalization tools (ANTs) (101). One image for each subject was manually segmented,
with the ROI propagated to all aligned images. The myocardium was divided into 6 segments (76)
and the data for each segment was fitted to the model of velocity-selective modulation of image
signal (37).
Results are shown in Supporting Information Figures 4.9, 4.10 and 4.11. The estimated
average vz of the myocardium was 0.3-1.6cm/s during stable-diastole and 0.9–4.3cm/s during
stable-systole. The residue for the fitting (not shown here) was uniform and low for both systole
and diastole. Subject 1 had a higher HR of 70-85 (vs subject 2 and 3 with HR of 50-55 and 60-
65) and had higher estimated diastolic and systolic velocities.
Supporting Information Figure 4.10 contains representative images from Subject 2 and
shows that the model fits the data well for both diastolic and systolic images. Figure 4.11 shows a
scatter plot of normalized signal from all segments and all subjects plotted. This can be used to
infer the choice of Vc that minimizes the myocardial velocity-saturation.
Single Shot Echo Planar Imaging Supporting Information
67

Figure 4.9: Bullseye plots showing estimated velocities for 3 subjects. The average velocity during stable diastole is
estimated to be 0.3-1.6 cm/s across subjects. The average velocity during stable systole is estimated to be 0.9 – 4.3
cm/s. In general, the residue was low and about the same throughout the myocardium, indicating goodness of fit for
this simple model. Additionally, myocardial velocity during systole was slightly higher in the lateral wall compared
to the septal wall, consistent with the literature (4).

Figure 4.10: Six segment measured data (dots) and fitted data (solid line) for a representative subject during diastole
and systole. Increasing velocity cutoff increased the signal in diastole and systole. Also shown are representative
images demonstrating the change in SNR with Vc. Qualitatively, this also helps verify the difference in cardiac wall
motion as predicted by the velocity maps shown in the Fig. S1( i.e. the cardiac velocity is slightly higher in the lateral
wall than the septal wall).
Single Shot Echo Planar Imaging Supporting Information
68

Figure 4.11: Scatter plot of normalized signal from all segments and all subjects plotted. There was no detectable
saturation of myocardium using the SS-EPI sequence, using Vc ≥12cm/s for diastolic and ≥25 cm/s for systolic.

Coronary Endothelial Function Assessment Introduction
69
5. Coronary Endothelial Function Assessment
5.1 Introduction
The coronary endothelium plays a crucial role in vasomotor and blood flow regulation. In
coronary endothelial dysfunction (CED), vasodilation is impaired and paradoxical
vasoconstriction may occur with endothelial dependent stressors(116). This may limit the increase
in coronary blood flow (CBF) or cause a reduction in CBF with endothelial dependent stress. CED
occurs in the initial stages of coronary atherosclerosis presumably in the presence of traditional
cardiovascular risk factors(117) and is an independent predictor of cardiovascular disease
(CVD)(118–122), the leading cause of mortality and morbidity in the developed world. These
findings suggest the potential of CED detection to improve cardiovascular risk stratification or as
a therapeutic target itself.
Despite the potential of CED assessment to improve CVD management, clinical
application of coronary endothelial function testing has been limited due to a continued
dependence of invasive techniques in the absence of proven non-invasive techniques. The gold
standard for assessing CED remains coronary angiography(121,123), which involves radiation,
uses contrast agents, and has procedure-related complications. Peripheral endothelial function can
be non-invasively studied with ultrasound, however, these vessels rarely develop plaque rupture
and their function does not strongly correlate to that of the coronary arteries(124–126). Current
cardiac magnetic resonance (CMR) imaging methods to assess coronary endothelial function are
effective but rely on high spatial resolution for visualization of the coronary arteries in cross-
section(127–129). Hence there remains a continued need to develop novel non-invasive techniques
to directly study coronary endothelial function in at-risk asymptomatic populations(128).
Arterial spin labeling (ASL)-CMR is a non-contrast technique for measuring myocardial
perfusion (MP) and may be a viable alternative to assess CED that overcomes many of the
limitations of current CMR techniques(11,12,14,71,130). ASL-CMR requires lower spatial
resolution than coronary MRI because signals are measured in the myocardium and not in the
coronary tree. Additionally, MPI captures changes in the entire coronary arterial tree and, in
essence, measures CBF to large regions of myocardium. Therefore, ASL-CMR may provide a
more sensitive, non-invasive assessment of coronary endothelial function. We hypothesized that
Coronary Endothelial Function Assessment Methods
70
change in MP in response to endothelial dependent stress would be impaired in patients with
coronary artery disease (CAD) and those at high-risk for CAD compared to healthy controls.
5.2 Methods
Subject Population
Thirty-four subjects with no contraindications to MRI underwent ASL-CMR imaging and
were eligible for this study. The study subjects were divided among three groups: CAD (n=11),
high-risk for CAD (n=13), and healthy (n=10). CAD was defined as a (1) history consistent with
angina or myocardial infarction based on clinical symptoms and non-invasive imaging and (2) a
>50% stenosis seen on prior coronary angiography in at least one major epicardial coronary artery.
We defined high-risk patients as those with at-least 2 traditional CVD risk factors (hypertension,
diabetes mellitus, active tobacco use, and hyperlipidemia) but (1) neither evidence of angina or
myocardial infarction by clinical history and non-invasive imaging (2) nor angiographically-
proven CAD (as defined above). Healthy participants were <35 years of age and free of any
traditional CVD risk factors or known CAD. This was established using a brief questionnaire
documenting the absence of known heart disease, hypertension, smoking, diabetes and high
cholesterol. The University of Southern California institutional review board approved the study
protocol and informed consent was obtained from all subjects. Patients were also requested to
refrain from caffeine for at-least 12 hours before the exam.
Study Protocol
Maximum voluntary contraction (MVC) was determined for each participant prior to
beginning the imaging protocol and defined as the average of 3 maximum strength contractions of
the isometric handgrip (IHG) device (Stoelting Co., Wood Dale, IL). All participants were scanned
in the supine position. Images were acquired in a mid-short axis slice during stable diastasis
throughout this study. ASL-CMR was used to measure MP continuously for 5-8 min of rest, 5 min
of stress, and 10-12 min of recovery as shown in Figure 5.1. Stress was induced by sustained IHG
exercise at 30% MVC. A research team member was present in the scan room with the subject to
ensure compliance with IHG at the appropriate exertion level. This was done by directly reading
force measurement on the IHG display or via verbal confirmation from the subjects who could
read the IHG display. IHG device has previously been shown to be a repeatable endothelial
dependent stressor(127,129,131,132). Subjects were provided instructions to perform frequent 3-
Coronary Endothelial Function Assessment Methods
71
4 second breath-holds using a pre-recorded voice prompt from the scanner. Compliance with
breathing instructions was confirmed by the scan operator monitoring the respiratory bellows
signal in real-time. This reduced signal variation compared to free breathing and allowed for
shorter breath-holds compared to 12-16 seconds for previous ASL-CMR protocols(10–12,14).
Variation in the duration of rest and recovery period was primarily due to scan efficiency.
Scan efficiency is defined as the number of control and label image pairs acquired per-unit of time
(e.g. minute). A total for 54 MP measurements were collected for each subject over a duration of
20-25 min. For each subject; 18 MP measurements were collected during rest, 8-13 MP
measurements were collected during 5 min of stress, and the remaining perfusion measurements
were collected during recovery.

Figure 5.1: Scan protocol for coronary endothelial function testing using ASL-CMR. ASL-CMR was performed
continuously for approximately 20-25 minutes: 5-8 min rest, 5 min handgrip stress, and 8-12 min rest. Data acquired
during the periods marked rest, stress and recovery were each averaged to calculate MP for rest, stress, and recovery
respectively. The initial 1.5 min of data during the stress and recovery phases reflect a transient a state and were
therefore discarded.
MRI Protocol
All experiments were performed on a 3 Tesla system (Signa Excite HDxt, GE Healthcare)
using an eight-channel cardiac array and plethysmograph (PG) gating. ASL-CMR was performed
using a double gated flow alternating inversion recovery (FAIR) with balanced steady state free
precession (bSSFP) imaging(11,71). Double gated FAIR sequence was used because it is less
sensitive to expected heart rate variations during IHG stress(11). Labeling was performed using a
hyperbolic secant adiabatic inversion pulse with an inversion efficiency > 85% in all subjects.
Inversion efficiency was measured in-vivo by applying the inversion label immediately prior to
image acquisition, as described by Do et al.(11). MR imaging parameters were : TE/TR =
1.5ms/3.2ms, matrix size : 96 x 96, spatial resolution: 1.88mm-2.50mm ,FOV: 18cm-24cm, slice
thickness: 1cm, bandwidth per pixel: 651Hz, GRAPPA R = 1.6 with 24 autocalibration signal
(ACS) lines, image acquisition window : 192 ms, and flip angle = 50. The total duration for each
experiment was ~45min.
Coronary Endothelial Function Assessment Methods
72
Data Analysis
All data analysis was performed in a blinded fashion. Data was processed using MATLAB
(MathWorks Inc., South Natick, MA) as shown in Figure 5.2. Similar to previous ASL-CMR
studies all images were sinc interpolated by a factor of two to facilitate segmentation. Sinc
interpolation was performed by zero-padding k-space before inverse Fourier transform. Semi-
automated segmentation of the left ventricular myocardium was performed using offline motion
correction with advanced normalization tools with previously published settings(82).This
generated masks for global and per-segment (six segment) analysis. The entire left ventricle
myocardium from the slice was used as the region of interest (ROI) for global analysis, and the
AHA six-segment model(76) of a mid-short-axis slice was used for per-segment analysis(132).
MP was calculated using the interpolated signal difference (ΔM) from control and label T1
curves as previously described by Poncelet et al.(9) and Do et al.(11). T1 curves for control and
label images were fitted using established three parameter model for T1 mapping(102).
ΔM between control and label T1 curves was estimated at the average inversion time (TIavg) of
control and label images. MP was then calculated using the following equation:
ΔM = αM
0
fTI
avg
e
TI
avg
/T1
blood
,
derived from Buxton’s general kinetic model(103), where α is inversion efficiency, M0 is
equilibrium magnetization, f is myocardial perfusion, TIavg is mean inversion time of control and
label images, and T1blood is T1 is the T1 of arterial blood (assumed to be 1950ms at 3T based on
Weingartner et al. (103). Physiological noise (PN) ( in ml/g/min) was calculated as the standard
deviation of MP as described by Zun et al.(10) and is a measure of measurement variability.
Temporal SNR (tSNR) was calculated as MP divided by PN and was used as a measure of data
quality. MP values for rest, peak stress, and recovery period were calculated from the initial 5-8
min of rest data (18 pairs), 3.5 min of stress data (~6-8 pairs), and 10-12 min of recovery data (~20
pairs), respectively. The first 1.5 min of data during stress and recovery was not used in the
calculation of MP because it was collected during a transient state. MP response was calculated
as the difference in MP between stress and rest (i.e. MP stress – MP rest). Subjects were excluded
from the study if: 1) the subject was unable to complete the study and verified this verbally after
the scan, 2) there were technical difficulties with the imaging sequence, or if 3) the global tSNR
was < 1 during rest or stress, which indicates very low confidence (measurement variability is
Coronary Endothelial Function Assessment Methods
73
greater than the measured value). Individual pairs of MP measurements within a scan were rejected
in case of a mis-trigger.

Figure 5.2: Flowchart for semi-automatic calculation of global and segmental MP for rest, stress and recovery data.
Reference images are automatically selected from the control and label ASL-CMR images. Reference images are
manually segmented to generate masks. Control and label images are then registered using motion correction
software to their respective reference images and a reverse transformation is applied to the manually segmented
masks to generate masks for each control and label image. These masks along with the control and label images are
used for spatial averaging. These segmental and global values are used in double gated ASL-CMR analysis to generate
MP measurements for each pair of control and label images. The data is then sorted into rest, stress, and recovery as
shown in Figure 5.1

.
Coronary Endothelial Function Assessment Methods
74
Healthy High-risk CAD
Subjects

n 6 10 7
Age, yrs 24±1.7 68.5±7.3 61.4±11.8
Sex

Male 5 6 5
Female 1 4 2
Hypertension - 10 7
Hyperlipidemia - 10 7
Diabetes Mellitus - 3 1
CAD

Single-vessel - - 5
Multi-vessel - - 2
ASL

# of segments 36 60 42
MP Rest (ml/g/min) 1.4 ± 0.97 1.61 ± 1.12 1.31 ± 1.23
MP Stress (ml/g/min) 2.84 ± 1.77 2.31 ± 1.61 1.64 ± 1.49
MP Recovery (ml/g/min) 1.53 ± 1.14 1.34 ± 1.29 1.13 ± 0.96
MP Response (ml/g/min) 1.44 ± 1.46 0.69 ± 1.34 0.32 ± 1.93
HR Rest 60.4 ± 13.4 65.3 ± 14.1 65.1 ± 13.6
HR Stress 72.3 ± 14.2 75 ± 14.7 71.6 ± 14.1
Table 5.1: Demographic and measured MP, MP response and HR. Values are reported as mean ± standard deviation
and are only for subjects whose data were included in the final analysis.
Statistical Analysis
Analysis was performed on the segmental MP measurements based on the AHA six-
segment model for a mid-short axis slice. Data normality was examined using histograms. Mixed
effect model was used to test the global intra- and inter-group interaction. Contrast tests were used
for post hoc comparisons, which included the tests for intra-group changes from rest to stress,
inter-group difference for each measurement phase, and inter-group difference in changes from
rest to stress. Bonferroni Step-down (Holm) correction was used to control the false discovery
from multiple comparisons. Mixed model validity was examined using residual plots. Cook's D
and restricted likelihood distance (RLD) were used to detect the potential outliers. If outliers
Coronary Endothelial Function Assessment Results
75
detected, the sensitivity analysis was conducted by removing outliers. All data analyses were
conducted using SAS 9.4.
5.3 Results
Thirty-four subjects were enrolled in the study. One healthy subject was unable to complete
the MRI scan. Three CAD patients and one high-risk patient were excluded due to technical
difficulties with the MRI pulse sequence, which resulted in poor image quality. Additionally, 6
subjects (3 healthy, 2 high-risk, and 1 CAD) were excluded due to inadequate tSNR. The remaining
data from 23 subjects (6 healthy, 10 high-risk, and 7 CAD) were analyzed. This resulted in 42
CAD, 60 high-risk and 36 healthy myocardial segments available for analysis. Segmental tSNR
was comparable for the CAD, high-risk, and healthy groups at 8.39±8.1, 6.94±5.1, and 8.57±5.35,
respectively. There was no substantial difference between the groups. These values were also
comparable to previous studies(10,12). Patient demographics and results are summarized in Table
5.1. Figure 5.5 shows representative image quality, MP, and tSNR in ASL-CMR.
The HR during rest, stress, and recovery is shown in Figure 5.3A with a statistically
significant interaction (p<0.05). There was a statistically significant increase in heart rate (HR) in
all three patient groups following IHG stress. MP during rest, stress, and recovery is shown with
group analysis in Figure 5.3B and for individual data in Figure 5.4 with a borderline statistically
significant interaction (p=0.055). The resting MP values reported in the study were comparable to
previous ASL-CMR (10,12,134) and PET studies (135). After correcting for multiple comparisons,
the inter-group comparison showed no differences in MP between rest and recovery. Stress MP
was significantly lower in both CAD (p=0.0005) and high-risk groups (p=0.05) compared to
healthy volunteers. MP was significantly higher during stress compared to rest (p=0.0002) for
healthy controls but not for CAD or high-risk groups. Trends showed stronger MP response
(stress-rest) in healthy (1.44 ± 1.46) and high-risk group (0.69 ± 1.34) compared to CAD group
(0.32 ± 1.93). The statistical significance diminished after multiple comparison correction (p=0.12
and 0.22 respectively).
Histograms showed that data from each measurement phase followed a bell shape
distribution, but with slightly heavy tails, which may indicate the presence of outliers. After
removing outliers the results remained the same.
Coronary Endothelial Function Assessment Discussion
76

Figure 5.3: A) HR measurements during rest (blue), stress (orange) and recovery (yellow) during the ASL-CMR scans
for CAD, high-risk and healthy groups. The error bars depict standard error. The HR difference between rest and
stress was statistically significant (**p< 0.05) for all groups. B) MP measurements for the CAD, high-risk, and healthy
groups. The difference between rest and stress MP was statistically significant for the healthy group (**p<0.05).
Stress MP was lower for the CAD and high-risk groups compared to the healthy group (*p<0.05). Rest and recovery
MP values were not statistically different in any of the three groups. C) MP response in the CAD, high-risk, and
healthy groups. There was a statistically significant difference in MP response for healthy vs. CAD and healthy vs.
high-risk with ***p<0.05 before correction for multiple comparisons and not significant after correction. * denotes
significance in comparison to healthy, ** denotes significance in relation to rest, and *** denotes significance only
before correction when compared to healthy.
5.4 Discussion
Coronary endothelial function, as determined by ASL-CMR, was significantly impaired in
individuals with established CAD and those at increased risk for CAD compared to healthy
participants. This study demonstrates the feasibility of using ASL-CMR determined MP to non-
invasively measure of coronary endothelial function. MP during IHG exercise, an established
endothelial-dependent stressor, was significantly lower for both the CAD and high-risk groups
compared to the healthy volunteers. Additionally, stress MP was significantly higher than the rest
MP only for the healthy volunteers.

Coronary Endothelial Function Assessment Discussion
77

Figure 5.4: Individual subject MP data during rest, stress, and recovery for all three subject groups. (A,B) MP
response was positive in all healthy subjects. In the high-risk and CAD groups, MP response was positive in some
subjects and negative in some subjects, but on average the MP response was small. (C,D) We detected no significant
difference between MP measured during rest and recovery, for all groups. This can serve as a measure of repeatability
for the resting MP measurement.
The use of ASL-CMR to evaluate changes in MP to endothelial-independent vasodilation
has previously been established and results presented here extend use for endothelial-dependent
vasodilation(14). Prior study has demonstrated the ability of CMR to assess coronary endothelial
function using other methods. Phase contrast flow velocity imaging of the right coronary artery
detected a significantly lower rise in coronary flow velocity in response to cold pressor test (CPT)
for asymptomatic diabetic females compared to healthy females (136). Hays and colleagues have
extensively reported on the use of coronary MRA imaging to formally assess endothelial
function(127,128,137,138). In these studies, average CBF, based off of diameter and flow velocity
Coronary Endothelial Function Assessment Discussion
78
changes in proximal coronary artery segments, was determined at rest, in response to IHG stress
and in one study during recovery. They found that average CBF on cardiac MRI increased by
>30%, decreased by 4%, and decreased by 13% in patients with no CAD, mild CAD, and
significant CAD respectively. In patients where 2 separate coronary arteries were imaged, a greater
degree of endothelial dysfunction was observed in the coronary artery with more severe stenosis
compared to the contralateral, minimally diseased artery. Additionally, no difference in CBF was
reported during rest and recovery periods(127). Phase contrast imaging can also measure global
coronary flow reserve in the coronary sinus.(139,140), which in combination with IHG stress, may
be viable for the assessment of CED.
Figure 5.5: Representative images, MP maps, and tSNR maps for (left) CAD, (middle) high-risk and (right)
healthy subject groups. (Top) Control and label images acquired during rest, stress, and recovery. Image quality
and SNR was high with good contrast between blood and myocardium. (Bottom) Resting MP maps and
corresponding resting tSNR maps overlaid on control images. In these representative cases, resting MP values
were comparable to previous ASL studies and segmental tSNR was high. Note that global tSNR was approximately
2x higher than segmental tSNR. Global tSNR <1 was used as a rejection criterion in this study.
Coronary Endothelial Function Assessment Discussion
79
While coronary endothelial function assessment with CMR can be determined with these
other methods, some significant limitations exist. Imaging of the coronary arteries in CMR is
limited by suboptimal spatial resolution, sensitivity to cardio-respiratory motion, and the ability to
only image one or a handful of proximal coronary segments. CED may not affect all coronary
arteries or microvasculature of a major coronary artery uniformly. ASL-CMR could potentially
improve CED detection because it is not limited by spatial resolution requirements, is less sensitive
to cardio-respiratory motion, and can potentially capture CED in the entire coronary tree by
measuring changes in MP.
Positron emission tomography (PET) imaging has already validated the use of absolute
MP to non-invasively determine coronary endothelial function and our findings are consistent with
these prior studies(141–143). PET requires injection of radiopharmaceuticals and involves
radiation exposure, making it a poor choice for screening asymptomatic patients. Therefore,
application of the MRI technique described here would be preferred in an asymptomatic population
who requires further long-term cardiovascular risk stratification.
Despite the potential advantages of ASL-CMR compared to other currently available non-
invasive techniques, adoption and widespread clinical use of this method will depend on not only
its ability to reliably detect CED in asymptomatic individuals without evidence of CAD, but also
whether such early identification results in therapies to improve coronary endothelial function and,
ultimately, long-term cardiovascular outcomes. Prior studies involving PET-measured MP do
suggest, however, that improvements in endothelium-related coronary artery function may have
direct preventive effects on coronary atherosclerosis progression. Initiation of glucose-lowering
treatment in diabetic patients significantly improved MP response during CPT testing, and the
magnitude of MP improvement inversely correlated with progression of coronary artery calcium,
an important subclinical CAD measure that strongly predicts development of future CVD
events(144,145). Similarly, decreases in body-mass index following gastric bypass surgery in
morbidly obese individuals were associated with an improvement in MP during CPT. Finally,
coronary endothelial function also improves following smoking cessation(146). In a population of
healthy, young individuals, the MP response to CPT significantly improved after just 1 month of
smoking cessation such that no difference was seen between ex-smokers and non-smokers(147).
Our study has limitations. ASL-CMR signal is primarily from MP, but also includes a
contribution from blood volume which may be of independent value in the context of CED.
Coronary Endothelial Function Assessment Discussion
80
Separation of the blood volume signal in ASL-CMR remains future work. ASL-CMR derived
measures of coronary endothelial function were not directly compared to those obtained from
coronary angiography. Since all study participants either had no known CHD or stable CHD,
invasive testing would not have been appropriate. MP reserve assessment with pharmacologic
vasodilation was not performed in this population and we cannot comment on whether co-existing
microvascular disease may have also been present. Only a small number of subjects were recruited
and findings presented need further validation in larger, well-characterized populations. The lack
of a detected difference in rest vs. stress MP in CAD and high-risk patients could be due to the
small sample size. Systematic evaluations of ASL-CMR on a larger cohort including patients with
CAD are warranted for future study. We did not capture blood pressure at rest or during stress and
MP-based assessment of coronary endothelial function may have been limited in the setting of
severe hypertension. The healthy volunteers were not age-matched to the patient groups and, as a
result, they were substantially younger. Considering the known age-related decline in coronary
flow reserve that occurs, this age difference may have confounded the differences in coronary
endothelial function observed here(148–150). We wanted to ensure the healthy volunteers were
free of cardiac risk factors and recruitment of such individuals with a similar average age to the
patient groups was difficult. Difficulties with the IHG device itself may have interfered with the
ability to maintain 30% MVC as the device was bulky and participants needed to maintain an
uncomfortable position in order to maintain a safe distance from the chest coil. Rate-pressure
product could not be calculated because blood pressure data was not recorded. Therefore, only HR
information was used to assess the hemodynamic response under stress. Finally, late gadolinium
enhanced imaging was not performed, therefore we were unable to rule out the possible existence
of infarct within the imaging slice for CAD patients.
In this study, we used GRAPPA acceleration (R=1.6) with 24 auto-calibration lines (ACS),
which we have found maximizes temporal SNR (tSNR) as shown in previous studies (134).
Acceleration beyond R=1.6 leads to reduction in tSNR with our particular scanner setup and 8-
channel cardiac coil. This is likely due to significant increase in thermal noise from g-factor losses.
Further acceleration may be feasible with higher channel counts (16- or 32-channel coils) or other
fast imaging sequences such as echo planar imaging (EPI).
In this study, the high failure rate was due to three reasons. 1) The first four subjects (3
CAD, 1 high-risk) were rejected due to poor image quality and un-resolved image artifacts. This
Coronary Endothelial Function Assessment Conclusion
81
was due to a sequence programming error. This implementation error was resolved completely
and will not cause failures in future studies. 2) Six subjects (1 CAD, 2 high-risk, and 3 healthy)
were rejected due to low tSNR. Previous ASL-CMR studies have reported global resting tSNR of
16±8. In this study, the six subjects had resting tSNR < 1. 3) One subject was rejected due to
inability to complete the IHG stress.
5.5 Conclusion
This study demonstrates the potential utility of ASL-CMR for non-invasive assessment of
CED. The change in MP in response to IHG stress was progressively lower when going from
healthy individuals to those at high-risk for CAD and, finally, to those with established CHD.
Additional study with larger populations to further validate this technique for assessment of
coronary endothelial function, including direct comparison with other available non-invasive
CMR and PET methods, and to better establish reference range values with this method will be
important.
Concluding Remarks
82
6. Concluding Remarks
Myocardial perfusion imaging with MRI is arguably the best technique for assessment and
screening of CAD. CMR first pass perfusion was recently used to guide treatment in CAD patients
and was shown to be non-inferior to fractional flow reserve imaging, the leading invasive
assessment technique. However, CMR first pass perfusion cannot be used to assess CAD in
patients with kidney disease. ASL-CMR can be the method of choice in kidney disease patients. It
can also provide benefit as a screening tool in patients without kidney disease. If it can be made as
robust as CMR first pass perfusion, it could be a lower cost and safe alternative to all existing
techniques.
It is worth noting that technology for CMR first pass perfusion is already mature and
technical developments to improve accuracy will only be incremental. ASL-CMR provides an
exciting area for research and development which is ripe for technical advances. In this thesis, we
addressed two major limitations of ASL-CMR 1) low sensitivity and 2) limited spatial coverage,
which restrict its clinical adoption. We demonstrated saturation SPASL to improve sensitivity of
ASL-CMR. We showed that single shot EPI can be used to improve spatial coverage of ASL-CMR
to three-slices in diastole and two-slices in systole. This was the first demonstration of multi-slice
ASL-CMR with MP measurements and PN that was comparable to single slice ASL-CMR studies.
Finally, we showed that ASL-CMR can be used for assessment of coronary endothelial function
which can likely improve risk stratification for CAD patients.
The work presented in this thesis lays the groundwork for future work in ASL-CMR but
had many limitations. Our SPASL technique used saturation pulses which were optimized for
single slice imaging. Saturation efficiency was sacrificed to reduce labeling of the mid-SAX
imaging slice to by -50dB. These pulses are not compatible with multi-slice imaging because they
will label the basal imaging slice in a multi-slice experiment which can significantly bias the ASL
measurements. We believe the best approach is to use VS labeling for SPASL because it eliminates
sensitivity to transit delay and will not introduce a spatial bias in MP due to labeling of slices closer
to the labeling slab.
We demonstrated the feasibility of SS-EPI based ASL-CMR but the sensitivity of this
technique was significantly lower in the lateral wall for diastolic single-slice and multi-slice
imaging. Systolic imaging had higher sensitivity than diastolic imaging especially in the lateral
Concluding Remarks
83
wall but there was a loss in sensitivity compared to single slice bSSFP with PLD of 1RR. We
recommend use of systolic EPI to achieve 3-4 slice coverage during systole and further
optimization of imaging parameters to improve sensitivity.
FAIR ASL-CMR was used in our multi-slice studies because it is the most stable and
common labeling scheme for ASL-CMR. However, we recognize that multi-slice FAIR ASL-
CMR underestimates MP and acknowledge this limitation. Our goal was to demonstrate the
feasibility of SS-EPI which was demonstrated with multi-slice FAIR ASL. However, we don’t
recommend multi-slice FAIR ASL-CMR for clinical imaging. Future work should explore EPI
with other labeling schemes that are compatible with multi-slice.
The inter-segment variation in MP should be low for MPI techniques, especially in healthy
volunteers at rest. In our work, we observed that there was significant inter-segment variation in
MP measured with existing ASL-CMR techniques. Carefully designed studies are needed to study
causes of this variation and to address them.
We also showed that velocity selective crushers can suppress the blood pool signal in SS-
EPI. This technique, however, will be challenging to implement in a clinical setting. Careful
selection of the imaging window is needed with empirical knowledge of velocity cutoff to suppress
blood pool and to minimize risk of labeling the myocardium. Such care may not be feasible for
widespread clinical adoption and in patients with high heart rate variation. Other techniques for
blood pool suppression should be explored e.g. velocity selective pulses designed with Fourier
velocity encoding to achieve robust blood suppression while minimizing the risk of labeling the
myocardium (70).
Gibbs ringing related artifacts in SPASL and FAIR experiments with bSSFP are also a
significant challenge. These artifacts make the analysis very difficult and required rejection of
segments boundary pixels cannot be excluded from the segmentation masks. It is likely that
suppressing the blood pool or matching contrast between control and label images can be used to
minimize effects of these artifacts on the ASL signal. Novel post processing techniques can also
be explored to correct these artifacts.
ASL data is limited by low sensitivity and can only be analyzed with segmental analysis
which requires segmentation. MP measurements with existing techniques are extremely sensitive
to segmentation, which is a major challenge for widespread clinical adoption. Recently, work by
Do et al. demonstrated machine learning based methods to automate segmentation, but these are
Concluding Remarks
84
not generalizable (151). Do et al.’s work can be expanded to automate segmentation of ASL-CMR,
to enable stable and consistent ASL measurements with minimal user intervention.
Much technical development and validation of ASL-CMR imaging and labeling schemes
is needed before ASL-CMR can be made clinically viable. Our collaborators at Sunny Brook
Research Institute are developing an ischemia model which will allow us to test the diagnostic
accuracy of ASL-CMR techniques for detection of varying levels of ischemia. This will offer a
unique opportunity to perform more advanced validation studies for ASL-CMR techniques.
Optimization of SS-EPI methods for cardiac imaging at 3T is needed and can have a wider impact
than just ASL-CMR. Improvements can include 1) development of a SMS 2DRF for 3-4x slice
coverage in systolic imaging using a combination of sequential and SMS imaging, 2) optimization
of imaging parameters to include parallel imaging to reduce geometric distortion, 3) use of
saturation FVE velocity selective pulses for robust blood pool suppression, and 4) development of
machine learning based methods to correct geometric distortion which may reduce PN. Robust
SS-EPI can be used for improving quantitative CMR it can be easily applied to T1 mapping, T2
mapping, and DTI, among other applications. Opportunity exists to combine developments in
SPASL and EPI to enable ASL-CMR with improved sensitivity and better spatial coverage.
SPASL should also be explored with VS saturation labeling, using VS pulses designed using FVE
methods and optimized to minimize the labeling of the myocardium, as done by Landes et al.
Methods to automate the processing pipeline in ASL should also be explored similar to BOLD
CMR to improve repeatability and reproducibility of ASL-CMR (152). Eventually, repeatability
and reproducibility of ASL-CMR should also be assessed in both humans and swine. We believe
that, ultimately, clinically viable ASL-CMR technique needs to be repeatable, reproducible, and
easy to use with a completely automated processing pipeline.
Work presented in this thesis has contributed to the body of knowledge related to ASL-
CMR. We have contributed to moving ASL-CMR a few steps closer towards clinical adoption by
developing saturation SPASL and demonstrating sequential multi-slice SS-EPI at 3T. It is my
belief that despite its challenges, ASL-CMR is an exciting area of research with very promising
prospects. I hope future work will focus of making ASL-CMR reproducible and clinically viable.

Bibliography
85
7. Bibliography
1. Benjamin EJ, Blaha MJ, Chiuve SE, Cushman M, Das SR, et al. Heart Disease and Stroke
Statistics—2017 Update. Circulation 2017;135:e146–e603.
2. Rosamond W, Flegal K, Furie K, Go A, Greenlund A, et al. Heart Disease and Stroke
Statistics—2008 Update. Circulation 2008;117.
3. Nichols M, Townsend N, Scarborough P, Rayner M. Cardiovascular disease in Europe 2014:
epidemiological update. European Heart Journal 2014;35:2950–2959.
4. Garber AM, Hlatky MA, Chareonthaitawee P, and Askew JW. Stress testing for the diagnosis
of obstructive coronary heart disease.; 2017.
5. Hussain ST, Paul M, Plein S, et al. Design and rationale of the MR-INFORM study: stress
perfusion cardiovascular magnetic resonance imaging to guide the management of patients with
stable coronary artery disease.; 2012.
6. Miskulin D RM. Nephrogenic systemic fibrosis/nephrogenic fibrosing dermopathy in advanced
renal failure.; 2017.
7. System USRD. USRDS 2013 Annual Data Report: Atlas of Chronic Kidney Disease and End-
Stage Renal Disease in the United States.; 2013.
8. Afzali B, Haydar AA, Vinen K, and Goldsmith DJ. From Finland to Fatland: Beneficial
Effects of Statins for Patients with Chronic Kidney Disease. Journal of the American Society of
Nephrology 2004;15:2161–2168.
9. Poncelet BP, Koelling TM, Schmidt CJ, et al. Measurement of human myocardial perfusion by
double-gated flow alternating inversion recovery EPI. Magnetic Resonance in Medicine
1999;41:510–519.
10. Zun Z, Wong EC, Nayak KS. Assessment of myocardial blood flow (MBF) in humans using
arterial spin labeling (ASL): Feasibility and noise analysis. Magnetic Resonance in Medicine
2009;62:975–983.
11. Do HP, Yoon AJ, Fong MW, Saremi F, Barr ML, Nayak KS. Double-gated myocardial ASL
perfusion imaging is robust to heart rate variation: Double-Gated Myocardial ASL is Robust to
HRV. Magnetic Resonance in Medicine 2017;77:1975–1980.
12. Do H, Jao TR, Nayak KS. Myocardial arterial spin labeling perfusion imaging with improved
sensitivity. Journal of Cardiovascular Magnetic Resonance 2014;16:15.
13. Wacker CM, Fidler F, Dueren C, et al. Quantitative assessment of myocardial perfusion with
a spin-labeling technique: Preliminary results in patients with coronary artery disease. J. Magn.
Reson. Imaging 2003;18:555–560.
Bibliography
86
14. Zun Z, Varadarajan P, Pai RG, Wong EC, Nayak KS. Arterial Spin Labeled CMR Detects
Clinically Relevant Increase in Myocardial Blood Flow With Vasodilation. JACC: Cardiovascular
Imaging 2011;4:1253–1261.
15. Capron T, Troalen T, Robert B, Jacquier A, Bernard M, Kober F. Myocardial perfusion
assessment in humans using steady-pulsed arterial spin labeling: Human MBF Using Steady-
Pulsed Arterial Spin Labeling. Magnetic Resonance in Medicine 2015;74:990–998.
16. Nishimura D. Principles of magnetic resonance imaging. 2010.
17. Signal, Contrast, and Noise. In: Magnetic Resonance Imaging. John Wiley & Sons, Ltd; 2014.
pp. 325–373.
18. Macovski A. Noise in MRI. Magnetic Resonance in Medicine 1996;36:494–497.
19. Krüger G, Glover GH. Physiological noise in oxygenation‐sensitive magnetic resonance
imaging. Magnetic Resonance in Medicine 2001;46:631–637.
20. Triantafyllou C, Polimeni JR, Wald LL. Physiological noise and signal-to-noise ratio in fMRI
with multi-channel array coils. NeuroImage 2011;55:597–606.
21. Triantafyllou C, Hoge RD, Krueger G, et al. Comparison of physiological noise at 1.5 T, 3 T
and 7 T and optimization of fMRI acquisition parameters. NeuroImage 2005;26:243–250.
22. Terrence Jao KN. Analysis of physiological noise in quantitative cardiac magnetic resonance.
PLOS ONE 2019;14:e0214566.
23. Pauly J, Nishimura D, Macovski A. A k-space analysis of small-tip-angle excitation. Journal
of Magnetic Resonance (1969) 1989;81:43–56.
24. Bernstein MA, King KF, Zhou XJ. Chapter 3 - basic radiofrequency pulse functions. Handbook
of MRI Pulse Sequences. Burlington: Academic Press; 2004. pp. 67–95.
25. Roux PL, Nishimura D, Macovski A. Parameter Relations for the Shinnar-Le Roux Selective
Excitation Pulse Design Algorithm. IEEE Transactions on Medical Imaging 1991;10:53–65.
26. Hardy CJ, Cline HE. Spatial localization in two dimensions using NMR designer pulses.
Journal of Magnetic Resonance (1969) 1989;82:647–654.
27. Stenger VA, Boada FE, Noll DC. Multishot 3D slice-select tailored RF pulses for MRI. Magn.
Reson. Med. 2002;48:157–165.
28. Conolly S, Nishimura D, Macovski A, Glover G. Variable-rate selective excitation. Journal of
Magnetic Resonance (1969) 1988;78:440–458.
29. Brian A. Hargreaves SMC Charles H Cunningham, Dwight G Nishimura. Variable‐rate
selective excitation for rapid MRI sequences. Magnetic Resonance in Medicine 2004;52:590–597.
Bibliography
87
30. Alley MT, Pauly JM, Sommer FG, Pelc NJ. Angiographic Imaging with 2D RF Pulses.
Magnetic Resonance in Medicine 1997;37:260–267.
31. Banerjee S, Nishimura DG, Shankaranarayanan A, Saritas EU. Reduced field-of-view DWI
with robust fat suppression and unrestricted slice coverage using tilted 2DRF excitation: rFOV
DWI with Tilted 2D RF Excitation. Magnetic Resonance in Medicine 2016;76:1668–1676.
32. Sui Y, Arunachalam SP, Arani A, et al. Cardiac MR elastography using reduced-FOV, single-
shot, spin-echo EPI. Magnetic Resonance in Medicine 2018;80:231–238.
33. Meyer CH, Pauly JM, Macovskiand A, Nishimura DG. Simultaneous spatial and spectral
selective excitation. Magnetic Resonance in Medicine 1990;15:287–304.
34. Bernstein MA, King KF, Zhou XJ. CHAPTER 4 - SPECTRAL RADIOFREQUENCY
PULSES. Handbook of MRI Pulse Sequences. Burlington: Academic Press; 2004. pp. 96–124.
35. Zur Y. Design of improved spectral-spatial pulses for routine clinical use. Magnetic Resonance
in Medicine 2000;43:410–420.
36. Bernstein MA, King KF, Zhou XJ. CHAPTER 10 - CORRECTION GRADIENTS. Handbook
of MRI Pulse Sequences. Burlington: Academic Press; 2004. pp. 292–362.
37. Wong EC, Cronin M, Wu W-C, Inglis B, Frank LR, Liu TT. Velocity-selective arterial spin
labeling. Magnetic Resonance in Medicine 2006;55:1334–1341.
38. Wang J, Yarnykh VL, Hatsukami T, Chu B, Balu N, Yuan C. Improved suppression of plaque-
mimicking artifacts in black-blood carotid atherosclerosis imaging using a multislice motion-
sensitized driven-equilibrium (MSDE) turbo spin-echo (TSE) sequence. Magnetic Resonance in
Medicine 2007;58:973–981.
39. Bieri O, Scheffler K. Fundamentals of balanced steady state free precession MRI:
Fundamentals of Balanced SSFP MRI. Journal of Magnetic Resonance Imaging 2013;38:2–11.
40. Bieri O, Scheffler K. Optimized balanced steady‐state free precession magnetization transfer
imaging. Magnetic Resonance in Medicine 2007;58:511–518.
41. Scheffler K. On the transient phase of balanced SSFP sequences. Magnetic Resonance in
Medicine 2003;49:781–783.
42. Worters PW, Hargreaves BA. Balanced SSFP transient imaging using variable flip angles for
a predefined signal profile. Magnetic Resonance in Medicine 2010;64:1404–1412.
43. Smith T, Zun Z, Wong EC, Nayak KS. Design and use of variable flip angle schedules in
transient balanced SSFP subtractive imaging. Magn. Reson. Med. 2010;63:537–542.
44. Mansfield P. Multi-planar image formation using NMR spin echoes. J. Phys. C: Solid State
Phys. 1977;10:L55–L58.
Bibliography
88
45. Lau AZ, Tunnicliffe EM, Frost R, Koopmans PJ, Tyler DJ, Robson MD. Accelerated human
cardiac diffusion tensor imaging using simultaneous multislice imaging: Multiband Human
Cardiac Diffusion Tensor Imaging. Magnetic Resonance in Medicine 2015;73:995–1004.
46. Ferreira PF, Kilner PJ, McGill L-A, et al. In vivo cardiovascular magnetic resonance diffusion
tensor imaging shows evidence of abnormal myocardial laminar orientations and mobility in
hypertrophic cardiomyopathy. Journal of Cardiovascular Magnetic Resonance 2014;16:87.
47. Buonocore MH, Gao L. Ghost artifact reduction for echo planar imaging using image phase
correction. Magn. Reson. Med. 1997;38:89–100.
48. Hu X, Le TH. Artifact reduction in EPI with phase-encoded reference scan. Magn. Reson.
Med. 1996;36:166–171.
49. Hoge WS, Polimeni JR. Dual-polarity GRAPPA for simultaneous reconstruction and ghost
correction of echo planar imaging data: Dual-Polarity GRAPPA for EPI. Magnetic Resonance in
Medicine 2016;76:32–44.
50. Xiang Q-S, Ye FQ. Correction for geometric distortion and N/2 ghosting in EPI by phase
labeling for additional coordinate encoding (PLACE). Magnetic Resonance in Medicine
2007;57:731–741.
51. Kim Y-C, Nielsen J-F, Nayak KS. Automatic correction of echo-planar imaging (EPI) ghosting
artifacts in real-time interactive cardiac MRI using sensitivity encoding. Journal of Magnetic
Resonance Imaging 2008;27:239–245.
52. Lee J, Han Y, Ye JC. k-Space Deep Learning for Reference-free EPI Ghost Correction.
arXiv:1806.00153 [cs, stat] 2018.
53. Lee J, Jin KH, Ye JC. Reference-free single-pass EPI Nyquist ghost correction using
annihilating filter-based low rank Hankel matrix (ALOHA): Reference-Free Single-Pass EPI
Nyquist Ghost Correction using ALOHA. Magn. Reson. Med. 2016;76:1775–1789.
54. Xie VB, Lyu M, Liu Y, Feng Y, Wu EX. Robust EPI Nyquist ghost removal by incorporating
phase error correction with sensitivity encoding (PEC-SENSE): Nyquist Ghost Removal by PEC-
SENSE. Magn. Reson. Med. 2018;79:943–951.
55. Franz Schmitt RL Michael K Stehling, Robert Turner, H Fischer. Echo-Planar Imaging,
Theory, Technique and Application. 1998:179–200.
56. Chen N, Wyrwicz AM. Removal of EPI Nyquist ghost artifacts with two-dimensional phase
correction. Magnetic Resonance in Medicine 2004;51:1247–1253.
57. Wan X, Gullberg GT, Parker DL, Zeng GL. Reduction of geometric and intensity distortions
in echo-planar imaging using a multireference scan. Magn. Reson. Med. 1997;37:932–942.
58. David L. Foxall JH Paul R Harvey. Rapid iterative reconstruction for echo planar imaging.
Magnetic Resonance in Medicine 1999;42:541–547.
Bibliography
89
59. Michael H. Buonocore DCZ. Image‐based ghost correction for interleaved EPI. Magnetic
Resonance in Medicine 2001;45:96–108.
60. Lee KJ, Barber DC, Paley MN, Wilkinson ID, Papadakis NG, Griffiths PD. Image-based EPI
ghost correction using an algorithm based on projection onto convex sets (POCS). Magn. Reson.
Med. 2002;47:812–817.
61. Scott B. Reeder ERM Anthony Z Faranesh, Ergin Atalar. A novel object‐independent
“balanced” reference scan for echo‐planar imaging. Journal of Magnetic Resonance Imaging
1999;9:847–852.
62. Wargo CJ, Moore J, Gore JC. A comparison and evaluation of reduced-FOV methods for
multi-slice 7T human imaging. Magnetic Resonance Imaging 2013;31:1349–1359.
63. Chen J, Licht DJ, Smith SE, et al. Arterial spin labeling perfusion MRI in pediatric arterial
ischemic stroke: Initial experiences. J. Magn. Reson. Imaging 2009;29:282–290.
64. Wolf RL, Detre JA. Clinical neuroimaging using arterial spin-labeled perfusion magnetic
resonance imaging. Neurotherapeutics 2007;4:346–359.
65. Sasitorn Petcharunpaisan MC Joana Ramalho. Arterial spin labeling in neuroimaging. World
Journal of Radiology 2010;2:384–398.
66. Williams DS, DETREtt JA, LEIGHt JS, Koretsky AP. Magnetic resonance imaging of
perfusion using spin inversion of arterial water. Proc. Natl. Acad. Sci. USA 1992:5.
67. Dai W, Garcia D, de Bazelaire C, Alsop DC. Continuous flow-driven inversion for arterial spin
labeling using pulsed radio frequency and gradient fields. Magn. Reson. Med. 2008;60:1488–
1497.
68. Alsop DC, Detre JA, Golay X, et al. Recommended implementation of arterial spin‐labeled
perfusion MRI for clinical applications: A consensus of the ISMRM perfusion study group and the
European consortium for ASL in dementia. Magnetic Resonance in Medicine 2015;73:102–116.
69. Shin T, Hu BS, Nishimura DG. Off-resonance-robust velocity-selective magnetization
preparation for non-contrast-enhanced peripheral MR angiography: Off-Resonance-Robust VS
Preparation for NCE MRA. Magnetic Resonance in Medicine 2013;70:1229–1240.
70. Qin Q, Shin T, Schär M, Guo H, Chen H, Qiao Y. Velocity-selective magnetization-prepared
non-contrast-enhanced cerebral MR angiography at 3 Tesla: Improved immunity to B0/B1
inhomogeneity: VS Prepared Cerebral MRA with Improved B0/B1 Immunity. Magnetic
Resonance in Medicine 2016;75:1232–1241.
71. Kober F, Jao T, Troalen T, Nayak KS. Myocardial arterial spin labeling. Journal of
Cardiovascular Magnetic Resonance 2016;18.
72. Jao TR, Nayak KS. Demonstration of velocity selective myocardial arterial spin labeling
perfusion imaging in humans. Magnetic Resonance in Medicine 2018;80:272–278.
Bibliography
90
73. Edelman RR, Chen Q. EPISTAR MRI: Multislice mapping of cerebral blood flow. Magnetic
Resonance in Medicine 1998;40:800–805.
74. Wong EC, Buxton RB, Frank LR. Implementation of quantitative perfusion imaging
techniques for functional brain mapping using pulsed arterial spin labeling. NMR in Biomedicine
1997;10:237–249.
75. Jahng G-H, Zhu X-P, Matson GB, Weiner MW, Schuff N. Improved perfusion-weighted MRI
by a novel double inversion with proximal labeling of both tagged and control acquisitions.
Magnetic Resonance in Medicine 2003;49:307–314.
76. Cerqueira MD. Standardized Myocardial Segmentation and Nomenclature for Tomographic
Imaging of the Heart. Circulation 2002;105:539–542.
77. Javed A, Jao T, Ghugre N, Wright G, Wong EC, Nayak KS. “Saturation”-based Steady Pulsed
Myocardial ASL Perfusion Imaging. In: Journal of Cardiovascular Magnetic Resonance. Seattle;
2019.
78. Jao TR, Nayak KS. Demonstration of velocity selective myocardial arterial spin labeling
perfusion imaging in humans: Demonstration of Velocity Selective Myocardial Arterial Spin
Labeling Perfusion Imaging in Humans. Magnetic Resonance in Medicine 2018;80:272–278.
79. Uecker M, Lai P, Murphy MJ, et al. ESPIRiT-an eigenvalue approach to autocalibrating
parallel MRI: Where SENSE meets GRAPPA. Magnetic Resonance in Medicine 2014;71:990–
1001.
80. Roemer PB, Edelstein WA, Hayes CE, Souza SP, Mueller OM. The NMR phased array.
Magnetic Resonance in Medicine 1990;16:192–225.
81. Avants BB, Tustison N, Song G. Advanced Normalization Tools (ANTS). Insight j 2009;2:1–
35.
82. Javed A, Jao TR, Nayak KS. Motion correction facilitates the automation of cardiac ASL
perfusion imaging. Journal of Cardiovascular Magnetic Resonance 2015;17:1–2.
83. Jao T, Zun Z, Varadarajan, Padmi, Pai, Ramdas, Nayak K. Myocardial ASL Data Filtering for
Improved Detection of CAD. In: ISMRM 20th Scientific Sessions. Melbourne; 2012. p. 3892.
84. Walker E, Nowacki AS. Understanding Equivalence and Noninferiority Testing. Journal of
General Internal Medicine 2011;26:192–196.
85. Sung K, Nayak KS. Design and use of tailored hard-pulse trains for uniformed saturation of
myocardium at 3 Tesla. Magn. Reson. Med. 2008;60:997–1002.
86. Garwood M, DelaBarre L. The Return of the Frequency Sweep: Designing Adiabatic Pulses
for Contemporary NMR. Journal of Magnetic Resonance 2001;153:155–177.
Bibliography
91
87. Reeder SB, Faranesh AZ, Boxerman JL, McVeigh ER. In vivo measurement of T*2 and field
inhomogeneity maps in the human heart at 1.5 T. Magnetic Resonance in Medicine 1998;39:988–
998.
88. van Essen GJ, te Lintel Hekkert M, Sorop O, et al. Cardiovascular Function of Modern Pigs
Does not Comply with Allometric Scaling Laws. Sci Rep 2018;8:792.
89. Zun Z, Wong EC, Nayak KS. Myocardial ASL Perfusion Imaging Using Pulsed 2D Tagging
of the Proximal Aorta. In: International Society of Magnetic Resonance in Medicine. Vol. 17. ;
2009. p. 1737.
90. Epstein FH, Meyer CH. Myocardial Perfusion Using Arterial Spin Labeling CMR. JACC:
Cardiovascular Imaging 2011;4:1262–1264.
91. Cerqueira MD, Weissman NJ, Dilsizian V, et al. Standardized Myocardial Segmentation and
Nomenclature for Tomographic Imaging of the Heart. Journal of Cardiovascular Magnetic
Resonance 2002;4:203–210.
92. Jao TR, Nayak K. Simultaneous Multi-slice Cardiac ASL. In: Proc. ISMRM 25th Scientific
Sessions. Honolulu; 2017. p. 1888.
93. Landes V, Jao T, Nayak K. Simultaneous Multi-slice bSSFP CMR: Is it feasible? In: SCMR
21st Scientific Session. Barcelona; 2018. p. 333.
94. Lau AZ, Tunnicliffe EM, Frost R, Koopmans PJ, Tyler DJ, Robson MD. Accelerated human
cardiac diffusion tensor imaging using simultaneous multislice imaging. Magnetic Resonance in
Medicine 2015;73:995–1004.
95. Feinberg DA, Hoenninger JC, Crooks LE, Kaufman L, Watts JC, Arakawa M. Inner volume
MR imaging: technical concepts and their application. Radiology 1985;156:743–747.
96. Hargreaves BA, Cunningham CH, Nishimura DG, Conolly SM. Variable-rate selective
excitation for rapid MRI sequences. Magnetic Resonance in Medicine 2004;52:590–597.
97. Jezzard P, Balaban RS. Correction for geometric distortion in echo planar images from B0
field variations. Magnetic Resonance in Medicine 1995;34:65–73.
98. Schär M, Kozerke S, Fischer SE, Boesiger P. Cardiac SSFP imaging at 3 Tesla. Magnetic
Resonance in Medicine 2004;51:799–806.
99. Schenck JF. The role of magnetic susceptibility in magnetic resonance imaging: MRI magnetic
compatibility of the first and second kinds. Medical Physics 1996;23:815–850.
100. Robson PM, Grant AK, Madhuranthakam AJ, Lattanzi R, Sodickson DK, McKenzie CA.
Comprehensive quantification of signal‐to‐noise ratio and g‐factor for image‐based and k‐space‐
based parallel imaging reconstructions. Magnetic Resonance in Medicine 2008;60:895–907.
101. Avants B, Tustison N, Song G. Advanced normalization tools (ANTS). Insight J 2009:1–35.
Bibliography
92
102. Kellman P, Hansen MS. T1-mapping in the heart: accuracy and precision. J Cardiovasc Magn
Reson 2014;16:2.
103. Buxton RB, Frank LR, Wong EC, Siewert B, Warach S, Edelman RR. A general kinetic
model for quantitative perfusion imaging with arterial spin labeling. Magnetic Resonance in
Medicine 1998;40:383–396.
104. Weingärtner S, Meßner NM, Budjan J, et al. Myocardial T1-mapping at 3T using saturation-
recovery: reference values, precision and comparison with MOLLI. J Cardiovasc Magn Reson
2017;18:84.
105. Zaglavara T, Haaverstad R, Cumberledge B, et al. Dobutamine stress echocardiography for
the detection of myocardial viability in patients with left ventricular dysfunction taking β blockers:
accuracy and optimal dose. Heart 87:329–335.
106. Chareonthaitawee P. Heterogeneity of resting and hyperemic myocardial blood flow in
healthy humans. Cardiovascular Research 2001;50:151–161.
107. Motwani M, Kidambi A, Uddin A, Sourbron S, Greenwood JP, Plein S. Quantification of
myocardial blood flow with cardiovascular magnetic resonance throughout the cardiac cycle.
Journal of Cardiovascular Magnetic Resonance 2015;17:4.
108. Zun Z, Wong EC, Nayak KS. Assessment of myocardial blood flow (MBF) in humans using
arterial spin labeling (ASL): feasibility and noise analysis. Magnetic Resonance in Medicine
2009;62:975–983.
109. Gloor M, Scheffler K, Bieri O. Quantitative magnetization transfer imaging using balanced
SSFP. Magnetic Resonance in Medicine 2008;60:691–700.
110. Codreanu I, Robson MD, Golding SJ, Jung B a, Clarke K, Holloway CJ. Longitudinally and
circumferentially directed movements of the left ventricle studied by cardiovascular magnetic
resonance phase contrast velocity mapping. Journal of cardiovascular magnetic resonance : official
journal of the Society for Cardiovascular Magnetic Resonance 2010;12:48.
111. Greaves K, Puranik R, O’Leary JJ, Celermajer DS. Myocardial Tissue Velocities in the
Normal Left and Right Ventricle: Relationships and Predictors. Heart, Lung and Circulation
2004;13:367–373.
112. Buchenberg WB, Markl M, Bauer S, Bock J, Lorenz R, Jung BA. Dual VENC Phase Contrast
MRI for Simultaneous Assessment of Blood Flow and Cardiac Motion. In: Proc. ISMRM 19th
Scientific Sessions. Montréal; 2011. p. 3303.
113. Zun Z, Varadarajan P, Pai RG, Wong EC, Nayak KS. Arterial Spin Labeled CMR Detects
Clinically Relevant Increase in Myocardial Blood Flow With Vasodilation. JACC: Cardiovascular
Imaging 2011;4:1253–1261.
114. Kellman P, McVeigh ER. Image reconstruction in SNR units: A general method for SNR
measurement. Magnetic Resonance in Medicine 2005;54:1439–1447.
Bibliography
93
115. Greaves K, Puranik R, O’Leary JJ, Celermajer DS. Myocardial tissue velocities in the normal
left and right ventricle: relationships and predictors. Heart, lung & circulation 2004;13:367–373.
116. L. Ludmer P, Selwyn A, L. Shook T, et al. Paradoxical Vasoconstriction Induced by
Acetylcholine in Atherosclerotic Coronary Arteries.; 1986.
117. Gutierrez E, Flammer AJ, Lerman LO, Elizaga J, Lerman A, Fernandez-Aviles F. Endothelial
dysfunction over the course of coronary artery disease. European Heart Journal 2013;34:3175–
3181.
118. Halcox JPJ, Schenke WH, Zalos G, et al. Prognostic Value of Coronary Vascular Endothelial
Dysfunction. Circulation 2002;106:653–658.
119. Nitenberg A, Chemla D, Antony I. Epicardial coronary artery constriction to cold pressor test
is predictive of cardiovascular events in hypertensive patients with angiographically normal
coronary arteries and without other major coronary risk factor. Atherosclerosis 2004;173:115–123.
120. Schächinger V, Britten MB, Zeiher a M. Prognostic impact of coronary vasodilator
dysfunction on adverse long-term outcome of coronary heart disease. Circulation 2000;101:1899–
1906.
121. Suwaidi J a, Hamasaki S, Higano ST, Nishimura R a, Holmes DR, Lerman a. Long-term
follow-up of patients with mild coronary artery disease and endothelial dysfunction. Circulation
2000;101:948–954.
122. Schindler TH, Nitzsche EU, Munzel T, et al. Coronary vasoregulation in patients with various
risk factors in response to cold pressor testing. Journal of the American College of Cardiology
2003;42:814–822.
123. Hasdai D, Gibbons RJ, Holmes DR, Higano ST, Lerman A. Coronary Endothelial
Dysfunction in Humans Is Associated With Myocardial Perfusion Defects. Circulation
1997;96:3390 LP – 3395.
124. Hirooka Y, Egashira K, Imaizumi T, et al. Effect of l-arginine on acetylcholine-induced
endothelium-dependent vasodilation differs between the coronary and forearm vasculatures in
humans. Journal of the American College of Cardiology 1994;24:948–955.
125. Anderson TJ, Uehata A, Gerhard MD, et al. Close relation of endothelial function in the
human coronary and peripheral circulations. Journal of the American College of Cardiology
1995;26:1235–1241.
126. Takase B, Uehata A, Akima T, et al. Endothelium-dependent flow-mediated vasodilation in
coronary and brachial arteries in suspected coronary artery disease. The American Journal of
Cardiology 1998;82:1535–1539.
127. Hays AG, Stuber M, Hirsch G a., et al. Non-Invasive Detection of Coronary Endothelial
Response to Sequential Handgrip Exercise in Coronary Artery Disease Patients and Healthy
Adults. PLoS ONE 2013;8:1–8.
Bibliography
94
128. Hays AG, Hirsch G a., Kelle S, Gerstenblith G, Weiss RG, Stuber M. Noninvasive
visualization of coronary artery endothelial function in healthy subjects and in patients with
coronary artery disease. Journal of the American College of Cardiology 2010;56:1657–1665.
129. Hays AG, Iantorno M, Soleimanifard S, et al. Coronary vasomotor responses to isometric
handgrip exercise are primarily mediated by nitric oxide: a noninvasive MRI test of coronary
endothelial function. American Journal of Physiology - Heart and Circulatory Physiology
2015;308:H1343–H1350.
130. Do HP, Javed A, Jao TR, Kim H, Yoon AJ, Nayak KS. Arterial spin labeling CMR perfusion
imaging is capable of continuously monitoring myocardial blood flow during stress. Journal of
Cardiovascular Magnetic Resonance 2015.
131. Klocke FJ. Epicardial Coronary Artery Vasomotion. Journal of the American College of
Cardiology 2010;56:1666–1667.
132. Brown BG, Lee AB, Bolson EL, Dodge HT. Reflex constriction of significant coronary
stenosis as a mechanism contributing to ischemic left ventricular dysfunction during isometric
exercise. Circulation 1984;70:18–24.
133. Jao T, Zun Z, Varadarajan P, Pai RG, Nayak KS. Spatiotemporal Filtering of Myocardial
ASL Data: Implications in Detection and Diagnosis of Coronary Artery Disease. In: ISMRM 20th
Scientific Sessions. Vol. 20. Melbourne; 2012. p. 3892.
134. Jao TR, Do HP, Nayak KS. Myocardial ASL-CMR perfusion imaging with improved
sensitivity using GRAPPA. Journal of Cardiovascular Magnetic Resonance 2016;18:P100.
135. Chareonthaitawee P, Kaufmann PA, Rimoldi O, Camici PG. Heterogeneity of resting and
hyperemic myocardial blood flow in healthy humans. Cardiovascular Research 2001;50:151–161.
136. Maroules CD, Chang AY, Kotys-Traughber MS, Peshock RM. Coronary Endothelial
Dysfunction in Women With Type 2 Diabetes Measured by Coronary Phase Contrast Flow
Velocity Magnetic Resonance Imaging: Journal of Investigative Medicine 2015;63:856–861.
137. Hays AG, Iantorno M, Schär M, et al. Local coronary wall eccentricity and endothelial
function are closely related in patients with atherosclerotic coronary artery disease. Journal of
Cardiovascular Magnetic Resonance 2017:19–51.
138. Hays AG, Kelle S, Hirsch GA, et al. Regional Coronary Endothelial Function Is Closely
Related to Local Early Coronary Atherosclerosis in Patients With Mild Coronary Artery Disease.
Circulation: Cardiovascular Imaging 2012;5:341 LP – 348.
139. Kato S, Saito N, Nakachi T, et al. Stress Perfusion Coronary Flow Reserve Versus Cardiac
Magnetic Resonance for Known or Suspected CAD. Journal of the American College of
Cardiology 2017;70:869–879.
Bibliography
95
140. Kato S, Saito N, Kirigaya H, et al. Impairment of Coronary Flow Reserve Evaluated by Phase
Contrast Cine‐Magnetic Resonance Imaging in Patients With Heart Failure With Preserved
Ejection Fraction. J Am Heart Assoc 2016;5.
141. Schindler TH, Cardenas J, Prior JO, et al. Relationship Between Increasing Body Weight,
Insulin Resistance, Inflammation, Adipocytokine Leptin, and Coronary Circulatory Function.
Journal of the American College of Cardiology 2006;47:1188–1195.
142. Prior JO. Coronary Circulatory Dysfunction in Insulin Resistance, Impaired Glucose
Tolerance, and Type 2 Diabetes Mellitus. Circulation 2005;111:2291–2298.
143. Alexanderson E, Ochoa JM, Calleja R, et al. Endothelial Dysfunction in Systemic Lupus
Erythematosus: Evaluation with 13N-Ammonia PET. Journal of Nuclear Medicine 2010;51:1927–
1931.
144. Detrano R, Guerci AD, Carr JJ, et al. Coronary Calcium as a Predictor of Coronary Events in
Four Racial or Ethnic Groups. New England Journal of Medicine 2008;358:1336–1345.
145. Schindler TH, Cadenas J, Facta AD, et al. Improvement in coronary endothelial function is
independently associated with a slowed progression of coronary artery calcification in type 2
diabetes mellitus. European Heart Journal 2009;30:3064–3073.
146. Morita K, Tsukamoto T, Naya M, et al. Smoking Cessation Normalizes Coronary Endothelial
Vasomotor Response Assessed with 15O-Water and PET in Healthy Young Smokers. Journal of
Nuclear Medicine 2006;47:1914–1920.
147. Quercioli A, Montecucco F, Pataky Z, et al. Improvement in coronary circulatory function in
morbidly obese individuals after gastric bypass-induced weight loss: relation to alterations in
endocannabinoids and adipocytokines. European Heart Journal 2013;34:2063–2073.
148. Uren NG, Camici PG, Melin J, Bol A, Bruyne B, Radvan J. Effect of Aging on Myocardial
Perfusion Reserve. Journal of Nuclear Medicine 1995;36:2032–2036.
149. Czernin J, Müller P, Chan S, et al. Influence of age and hemodynamics on myocardial blood
flow and flow reserve. Circulation 1993;88:62 LP – 69.
150. Galderisi M, Rigo F, Gherardi S, et al. The impact of aging and atherosclerotic risk factors
on transthoracic coronary flow reserve in subjects with normal coronary angiography.
Cardiovascular Ultrasound 2012;10.
151. Hung P Do KSN Yi Guo, Andrew J Yoon. Accuracy, Uncertainty, and Adaptability of
Automatic Myocardial ASL Segmentation using Deep CNN. 2018.
152. Oksuz I, Mukhopadhyay A, Dharmakumar R, Tsaftaris SA. Unsupervised Myocardial
Segmentation for Cardiac BOLD. IEEE Transactions on Medical Imaging 2017;36:2228–2238.

Abstract (if available)

Abstract Coronary artery disease (CAD) is one of the leading causes of death in the United States. It is responsible for approximately one third of all deaths in individuals over the age of 35. Clinical manifestations of CAD including chest pain (angina) and myocardial infarction are a result of narrowing of coronary arteries, which reduces blood flow to the heart. The gold standard for diagnosis of CAD is via direct visualization of the epicardial vessels using coronary angiography. However, coronary angiography is unsuitable for screening and routine monitoring because it is invasive, involves ionizing radiation, and is expensive. In general, it is reserved for the most high-risk patients. An alternative is to use myocardial perfusion imaging (MPI) which can assess CAD by comparing myocardial perfusion (MP) under rest and stress. All existing MPI techniques to evaluate CAD are limited by either low sensitivity, use of ionizing radiation and/or the need for contrast injection which are not well tolerated in patients with kidney disease. Nevertheless, patients with kidney disease would benefit immensely from routine monitoring of CAD because they have a 10x higher likelihood of developing CAD and 3x higher likely hood of dying from CAD. ❧ Arterial spin labeling (ASL) is an alternative contrast free MR technique which uses blood as an endogenous contrast agent. However, ASL cardiac magnetic resonance imaging (ASL-CMR) techniques have several limitations. Firstly, perfusion measurements made with ASL-CMR are noisy because the observed signal is only 1-4% of the total myocardial signal. Secondly, to make ASL clinically viable at least 3-slice coverage in needed based on AHA’s recommendations. However, in humans, pharmacologic stress is limited to 3-4 min due to patient comfort and safety concerns which is the time currently required for single slice ASL exam. In this thesis, we try to address the above two limitations by developing ASL techniques that are compatible with multi-slice imaging and improve sensitivity of ASL-CMR. We also explore new applications of ASL-CMR for early diagnosis of CAD. ❧ We improved the steady pulsed labeling (SPASL) technique developed by Capron et al to improve the sensitivity of ASL in the heart. The original technique was not compatible with multi-slice imaging and was sensitive to spin history. In our work, we eliminated the sensitivity of the technique to spin history without sacrificing improvements in sensitivity and we demonstrated SPASL that would be compatible with multi-slice imaging. We further optimized the labeling scheme to maximize signal strength and signal efficiency. ❧ Improving spatial coverage in ASL is very challenging because the duration of pharmacologic stress limits the acquisition time to 3-4 minutes. Spatial coverage can be improved using either sequential multi-slice or simultaneous multi-slice imaging. With current technology balanced steady state precession cannot be used with either techniques. In our work, we developed an echo planar imaging (EPI) sequence that is compatible with sequential multi-slice imaging and optimized it for cardiac imaging. We then demonstrated the feasibility of using this EPI sequence to improve spatial coverage of ASL in the heart. ❧ ASL is a contrast free technique that poses no incremental risk to the patient which makes it ideal for screening high-risk but asymptomatic patients for early diagnosis of coronary artery disease. We explored the application of ASL for diagnosis of coronary endothelial function (CED) and demonstrated the feasibility of ASL to diagnose CED.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Improving sensitivity and spatial coverage of myocardial arterial spin labeling

PDF

Improved myocardial arterial spin labeled perfusion imaging

PDF

Assessment of myocardial blood flow in humans using arterial spin labeled MRI

PDF

Radiofrequency pulse performance for myocardial ASL

PDF

Dynamic cardiovascular magnetic resonance imaging for improved assessment of ischemic heart disease

PDF

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

PDF

Mapping water exchange rate across the blood-brain barrier

PDF

Improving arterial spin labeling in clinical application with deep learning

PDF

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

PDF

Functional connectivity analysis and network identification in the human brain

PDF

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

PDF

One-dimensional nanomaterials for electronic and sensing applications

Asset Metadata

Creator Javed, Ahsan (author)

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

School Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Electrical Engineering

Publication Date 12/17/2019

Defense Date 10/23/2019

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

Tag arterial spin labeling,cardiac magnetic resonance imaging,coronary artery disease,echo planar imaging,multi-slice cardiac imaging,myocardial blood flow,myocardial perfusion,OAI-PMH Harvest,steady pulsed arterial spin labeling

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Nayak, Krishna S. (committee chair), Haldar, Justin P. (committee member), Wang, Danny J. (committee member), Wong, Eric C. (committee member)

Creator Email ahsan.javed@zohomail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-256011

Unique identifier UC11673226

Identifier etd-JavedAhsan-8083.pdf (filename),usctheses-c89-256011 (legacy record id)

Legacy Identifier etd-JavedAhsan-8083.pdf

Dmrecord 256011

Document Type Dissertation

Rights Javed, Ahsan

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