Download fast cardiovascular magnetic resonance imaging

FAST CARDIOVASCULAR
MAGNETIC RESONANCE IMAGING
a dissertation
submitted to the department of electrical engineering
and the committee on graduate studies
of stanford university
in partial fulfillment of the requirements
for the degree of
doctor of philosophy
Krishna Shrinivas Nayak
January 2001
c Copyright by Krishna Shrinivas Nayak 2001
All Rights Reserved
ii
I certify that I have read this dissertation and that in
my opinion it is fully adequate, in scope and quality, as
a dissertation for the degree of Doctor of Philosophy.
Dwight G. Nishimura
(Principal Advisor)
I certify that I have read this dissertation and that in
my opinion it is fully adequate, in scope and quality, as
a dissertation for the degree of Doctor of Philosophy.
Bob S. Hu
I certify that I have read this dissertation and that in
my opinion it is fully adequate, in scope and quality, as
a dissertation for the degree of Doctor of Philosophy.
John M. Pauly
Approved for the University Committee on Graduate
Studies:
iii
Abstract
is rapidly emerging as a powerful tool for cardiovascular imaging. While respiratory and cardiac motion
have complicated traditionally long MR scans, recently developed real-time interactive MRI techniques appear robust. Entire image acquisitions are completed within
a fraction of a cardiac cycle (usually < 150 ms), with minimal motion artifacts. An
important part of cardiovascular MRI is the development of imaging sequences that
provide dierent types of contrast and information in real time. In this thesis, I will
present four real-time interactive MRI sequences that are aimed at imaging dierent aspects of cardiovascular disease. In addition, I will present a new technique for
correcting one important type of MR image artifact.
First a method is presented for imaging cardiac blood ow as well as anatomy
in real time. Spin velocity and spin density information are acquired and displayed
simultaneously, much like color ow ultrasound. This tool is useful for evaluating
regurgitant valves and provides great scan exibility compared to ultrasound.
Secondly, a method is presented for improving contrast and reducing artifacts in
cardiac images that are dominated by blood signal. Real-time suppression of owing
blood improves blood-myocardium contrast and eliminates ow artifacts, enabling
the fast imaging of heart and vessel wall.
Next, there will be a discussion of two applications that require high spatial
and high temporal resolution. First is multislice imaging of the left ventricle, which
agnetic resonance imaging (MRI)
M
iv
involves acquiring upwards of 50 images per second to provide multiple slice information under conditions of cardiac stress. Second is the interactive screening of
coronary arteries, which requires sub-millimeter image resolution. When designing
imaging sequences, one must make decisions about tradeo between temporal resolution, spatial resolution and signal strength. In both of these cases the limits of
temporal and spatial resolution are discussed.
Finally, a new method is presented for estimating and correcting o-resonance artifacts. These artifacts, mainly caused by magnetic eld inhomogeneity and chemical
shift, cause image blurring in radial-based acquisitions. The new technique corrects
for these artifacts and saves scan time compared to existing techniques.
v
Acknowledgments
I have been incredibly fortunate to work closely with not one but three advisers here
at Stanford. My principal adviser, Prof. Dwight Nishimura, has been a constant
guiding force during my graduate career. He was involved in my admission, advised
me during my Masters degree year, and then welcomed me into his group. He has
the rare ability to advise while giving autonomy to his students. I appreciate his
incredible intuition and his subtle style of advising. For me, he is a role model.
I have also been fortunate to work with Prof. Bob Hu. Not only has he motivated
and guided me through four projects, but he is undeniably the best volunteer subject
in cardiac MRI. I thank him for bailing me out of multiple last-minute conference
situations, as well as for oering sensitive clinical advice on some of my family's
health problems. But what I appreciate most are our discussions, which are always
full of new and interesting ideas.
My third adviser has been Dr. John Pauly, who has provided a majority of the
technical backing for this work. I thank him for helping me work on the real-time
imaging project, for providing thorough answers to all of my MRI questions, and
for teaching me to always approach problems with a positive attitude.
On the clinical side, I have had the opportunity to work with many fantastic
collaborators. Particularly I thank Dr. Pedro Rivas, Dr. Michael McConnell, and
Dr. Phillip Yang for their involvement in the color ow, intravascular, and coronary
imaging projects. I also thank Dr. Garry Gold, Dr. Shuichiro Kaji, and Dr. Frandics
vi
Chan. On the engineering side, I would also like to thank Dr. Steve Conolly and
Dr. Craig Meyer for being very generous with their time.
I have thoroughly enjoyed spending time at the Magnetic Resonance Systems
Research Laboratory. The student atmosphere is incredible, and for that, I must
thank all the \labguys" past and present. Particularly, I thank Brian Hargreaves,
for being my (fraternal) twin in the lab, Karla Miller for being hip and musical,
Bob Schaer for loving football and being entertaining with all his impressions,
Julie Sabataitis for always being so sweet, and Bill Overall for all the games. Also,
I thank past students such as Gerry Luk-Pat, Sanjay Mani, and Adam Kerr who
helped me get started when I rst joined the lab.
I gratefully acknowledge the nancial support of a Fannie and John Hertz Foundation Graduate Fellowship, as well as grant support from GE Medical Systems, the
National Institutes of Health, and the Whittaker Foundation.
For me, these years at Stanford have been a highlight of my life. Outside of
research, I have been fortunate to meet many wonderful people, become involved in
their lives, and grow as a result. The positive energy, love, and support that has
come from numerous close friends has really kept me going. I appreciate all of you,
and your places in my life: Vinita Jain, Brad Johanson, Rudy Alarcon, Hari Nair,
Tony Chen, Aparajita Sohoni, Yaneeka Huq, and everyone else.
Finally, I wish to thank my immediate family for their support and guidance.
As a child, my parents ingrained in their children the highest value for education.
Throughout all adversity, mom and dad were always there to provide us with love
and unlimited support for our academic aspirations. While making our way through
an unconventional grade school and undergraduate path, my older sister, Asha,
blazed the trail which I simply had to follow. I thank her for setting such a good
example, and for looking out for her little brother.
Krishna Shrinivas Nayak
Stanford University
January 2001
vii
Contents
Abstract
iv
Acknowledgments
vi
List of Tables
xii
List of Figures
xiii
1 Introduction
1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Magnetic Resonance Imaging
5
2.1 Nuclear Magnetic Spins . . . .
2.2 Components of an Experiment .
2.3 Stages in MRI . . . . . . . . . .
2.3.1 Polarization . . . . . . .
2.3.2 Excitation . . . . . . . .
2.3.3 Acquisition . . . . . . .
.
.
.
.
.
.
viii
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5
8
9
9
9
10
2.3.4 Image Reconstruction .
2.4 Imaging Goals . . . . . . . . .
2.4.1 Contrast . . . . . . . .
2.4.2 Speed and SNR . . . .
2.4.3 Artifact Management .
2.5 Experimental Setup . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3 Real-Time Color Flow MRI
3.1 Method . . . . . . . . . .
3.1.1 Pulse Sequence . .
3.1.2 Acquisition Timing
3.1.3 Reconstruction . .
3.1.4 Display . . . . . .
3.2 Results . . . . . . . . . . .
3.2.1 Phantom Studies .
3.2.2 In Vivo Studies . .
3.3 Discussion . . . . . . . . .
14
15
15
16
16
18
20
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4 Real-Time Black Blood MRI
4.1 Methods . . . . . . . . . . . .
4.1.1 Pulse Sequence . . . .
4.1.2 Experimental Methods
4.2 Results . . . . . . . . . . . . .
4.2.1 Wall Motion . . . . . .
4.2.2 Intravascular Imaging .
21
21
24
25
26
28
29
30
38
40
.
.
.
.
.
.
.
.
.
.
.
.
ix
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
41
41
43
45
46
47
4.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5 Limits of Resolution
53
5.1 Real-time Multi-slice Stress Imaging . . . . .
5.1.1 Method . . . . . . . . . . . . . . . . .
5.1.2 Results . . . . . . . . . . . . . . . . . .
5.1.3 Summary of Multi-slice Stress Imaging
5.2 Real-time Coronary Imaging . . . . . . . . . .
5.2.1 Methods . . . . . . . . . . . . . . . . .
5.2.2 Results . . . . . . . . . . . . . . . . . .
5.2.3 Summary of Coronary Imaging . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6 EÆcient O-resonance Correction
6.1 Theory . . . . . . . . . . . . . . . . . . . .
6.2 General Methods . . . . . . . . . . . . . .
6.3 Application to Projection Reconstruction
Imaging . . . . . . . . . . . . . . . . . . .
6.3.1 Method . . . . . . . . . . . . . . .
6.3.2 Results . . . . . . . . . . . . . . . .
6.4 Application to Spiral Imaging . . . . . . .
6.4.1 Method . . . . . . . . . . . . . . .
6.4.2 Results . . . . . . . . . . . . . . . .
6.5 Discussion . . . . . . . . . . . . . . . . . .
7 Summary and Recommendations
x
54
56
59
62
65
66
71
74
76
. . . . . . . . . . . . . . . 77
. . . . . . . . . . . . . . . 78
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
80
80
81
84
87
88
91
93
7.1 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
A Artifacts from Fast Flow
97
A.1 Through-Plane Flow and Spectral-Spatial
Excitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
A.2 In-Plane Flow and Spiral Acquisition . . . . . . . . . . . . . . . . . . 100
Bibliography
103
xi
List of Tables
2.1 Tissue chemical shift and relaxation parameters at 1.5 T . . . . . . . 15
3.1 Scan parameters used in color ow MRI. . . . . . . . . . . . . . . . . 33
5.1 Scan parameters for RTI multi-slice studies . . . . . . . . . . . . . . . 58
5.2 Temporal and spatial resolution tradeos for high-resolution real-time
imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
xii
List of Figures
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
Polarization of magnetic spins . . . . . . . . . . . . . .
Relaxation of magnetic spins . . . . . . . . . . . . . . .
Magnetic elds used in MRI . . . . . . . . . . . . . . .
Trajectory of magnetization during excitation. . . . . .
Slice selective excitation . . . . . . . . . . . . . . . . .
Fourier relationship between k-space and object space.
Popular k-space trajectories. . . . . . . . . . . . . . . .
Stanford RTI system block diagram . . . . . . . . . . .
Stanford RTI system interface . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6
7
8
11
11
13
14
19
19
3.1
3.2
3.3
3.4
3.5
3.6
3.7
3.8
Spiral phase contrast pulse sequence for color ow MRI . . . . . . .
Sliding window reconstruction for color ow MRI . . . . . . . . . .
Image reconstruction block diagram for color ow MRI . . . . . . .
Display colormaps for color ow MRI . . . . . . . . . . . . . . . . .
Articial velocity aliasing in color ow MRI . . . . . . . . . . . . .
Validation of measured velocities . . . . . . . . . . . . . . . . . . .
Validation of measured real-time velocity waveforms . . . . . . . . .
Comparison of color ow ultrasound and color ow MRI in patients
.
.
.
.
.
.
.
.
22
25
26
27
28
31
32
34
xiii
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3.9
3.10
3.11
3.12
3.13
Color ow MRI sequences from patients with valvular regurgitation
Images of carotid ow . . . . . . . . . . . . . . . . . . . . . . . . .
Images of ow in the iliac aorta . . . . . . . . . . . . . . . . . . . .
Images of ow in the popliteal artery . . . . . . . . . . . . . . . . .
Images of coronary ow . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
35
36
36
37
37
4.1
4.2
4.3
4.4
4.5
Pulse sequence for black blood imaging using spatial presaturation . .
Out-of-slice suppression proles from simulations and measurements .
Short axis images with and without blood suppression . . . . . . . . .
Intravascular coil ow phantom experiment . . . . . . . . . . . . . . .
Intravascular coil rabbit aorta experiment . . . . . . . . . . . . . . .
42
44
48
49
50
5.1
5.2
5.3
5.4
5.5
5.6
5.7
5.8
Acquisition timing for multi-slice stress MRI . . . . . . . . . . . . . .
Resolution tradeo for high frame rate real-time imaging . . . . . . .
Multi-slice cardiac image sequence . . . . . . . . . . . . . . . . . . . .
Multi-slice cardiac images under resting and stress conditions . . . . .
First pass perfusion observed in three slices by multi-slice MRI . . . .
First pass perfusion signal time course . . . . . . . . . . . . . . . . .
High-resolution k-space trajectories . . . . . . . . . . . . . . . . . . .
Temporal and spatial resolution tradeos for high-resolution real-time
imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Comparison of EPI and spiral high-resolution real-time acquisitions .
Real-time coronary images . . . . . . . . . . . . . . . . . . . . . . . .
Comparison of real-time and gated coronary images . . . . . . . . . .
Contrast enhanced real-time images of the left coronary tree . . . . .
57
58
61
62
63
64
67
5.9
5.10
5.11
5.12
xiv
69
70
72
73
75
6.1
6.2
6.3
6.4
6.5
6.6
6.7
6.8
6.9
ORC-PR pulse sequence . . . . . . . . . . . . . . . . . . . . . . . .
ORC-PR k-space coverage . . . . . . . . . . . . . . . . . . . . . . .
ORC-PR phantom images . . . . . . . . . . . . . . . . . . . . . . .
ORC-PR single slice peripheral images . . . . . . . . . . . . . . . .
ORC-PR targeted maximum intensity projection images . . . . . .
ORC-VDS k-space trajectory . . . . . . . . . . . . . . . . . . . . .
Performance comparison of ORC-VDS with conventional techniques
ORC-VDS resolution phantom images . . . . . . . . . . . . . . . .
ORC-VDS coronary images . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
81
82
83
85
86
87
88
89
90
A.1 Eects of through-plane ow during spectral-spatial excitation. . . . . 99
A.2 Eects of in-plane ow during spiral acquisition. . . . . . . . . . . . . 102
xv
Chapter 1
Introduction
Magnetic resonance imaging (MRI) has experienced rapid growth over the past
twenty years. Compared to other imaging modalities such as X-ray or ultrasound,
MRI has the unique ability to acquire several dierent types of soft tissue image
contrast as well as ow, spectral, and other information.
The phenomenon of nuclear magnetic resonance (NMR) was rst discovered simultaneously by Bloch [1] and Purcell [2] in 1945. For many years it was mainly
used by chemists and physicists as a tool for identifying chemical compositions based
on dierences in atomic resonant frequencies, also known as NMR spectroscopy [3].
In 1972, Lauterbur discovered that the application of linear magnetic eld gradients
spatially encoded the NMR signal [4], spawning the eld of magnetic resonance imaging (MRI). Since then, MRI has evolved into a premier medical imaging modality for
imaging soft tissue, particularly in parts of the body that are relatively stationary.
Early research focused on achieving greater understanding of the MR signal and
developing hardware that would enable in vivo experimentation. Fundamental advances included the development of spatial encoding [4], slice-selective excitation [5],
and k-space analysis [6,7], each of which are described in Chapter 2.
Since then, MR imaging breakthroughs have come from the design of improved
hardware, novel pulse sequences for faster imaging, and new contrast mechanisms.
1
Chapter 1. Introduction
2
These advances have been instrumental in bringing MRI to routine clinical use.
Much of MRI research today is geared towards enabling new clinical applications.
1.1
Motivation
In the United States, cardiovascular disease is the leading cause of death among
adults. Diagnostic imaging plays a key role in disease monitoring and management.
Several dierent imaging modalities have been deemed appropriate for evaluating
dierent aspects of the disease. Currently, echocardiography is the gold standard
for imaging heart function and valvular regurgitation due to its high frame rates
and the availability of ow information. In addition, X-ray angiography is the gold
standard for imaging coronary arteries due to its snapshot imaging capability and
excellent image resolution.
Advantages of MRI over ultrasound include the ability to image arbitrary scan
planes and the ability to study ow in any direction. Ultrasound is restricted by
physical acoustic windows and is limited to imaging ow towards or away from
the transducer. The main advantage of MRI over X-ray techniques is that harmful
ionizing radiation is not used. In addition, MRI is capable of studying a wide variety
of processes such as myocardial perfusion, wall motion, valvular ow, and coronary
arteries{all in a single examination.
1.2
Outline
In this thesis, we present several techniques for fast cardiac and vascular imaging
with the specic goal of evaluating cardiovascular disease. We address the imaging of valvular regurgitation, cardiovascular ow, ventricular function, wall-motion
and perfusion, and coronary artery imaging. In addition we present a time saving
technique for correcting image blurring caused by o-resonance.
The structure of this thesis is as follows:
Chapter 1. Introduction
3
Chapter 2: Magnetic Resonance Imaging
This chapter contains a basic overview of MR imaging concepts, with emphasis
on fast imaging and the eects of ow and o-resonance. The experimental
setup used in the rest of this thesis is also described.
Chapter 3: Real-Time Color Flow MRI
This chapter describes the design and implementation of real-time color ow
MRI, which involves the simultaneous acquisition and display of anatomical
and ow information in a way that resembles color ow ultrasound. Pulse
sequence design, user interface, and patient scanning issues are described.
Prior publication of this work includes [8].
Chapter 4: Real-Time Black Blood MRI
This chapter describes a method for suppressing blood ow and ow-related
artifacts in real time. A discussion of steady state blood suppression is followed
by its application to ventricular assessment and intravascular imaging. Prior
publication of this work includes [9].
Chapter 5: Limits of Resolution in Real-Time MRI
The design of fast imaging sequences requires a compromise between temporal resolution, spatial resolution, and SNR. This chapter characterizes these
tradeos for popular acquisition schemes and current hardware limitations.
Example applications requiring extremes of this tradeo are described: 1) fast
multiple-slice real-time ventricular assessment under stress (high temporal resolution), and 2) real-time coronary localization (high spatial resolution). Prior
publications of this work include [10{12].
Chapter 6: EÆcient O-resonance Correction
This chapter describes a technique for estimating and correcting o-resonance
artifacts in projection reconstruction (PR) and spiral imaging, with little or
no cost in scan time. Prior publications of this work include [13,14].
Chapter 1. Introduction
Chapter 7: Summary & Future Work
4
This chapter summarizes the engineering contributions presented in this thesis,
and outlines areas for future work.
Chapter 2
Magnetic Resonance Imaging
This chapter contains an introduction to the magnetic resonance imaging process
which can also be found in many popular textbooks [15{18]. A classical description
of MRI physics is followed by a discussion of the components of an MRI experiment,
and a discussion of imaging goals. Finally, we describe the experimental setup that
was used to acquire the data presented in future chapters.
2.1
Nuclear Magnetic Spins
Atoms with an odd number of protons and/or neutrons possess a nuclear property
known as spin angular momentum, often referred to as spin. These can be classically
described as charged spheres spinning about an axis. Because these spins have
charge, they act as tiny magnetic dipoles. Under normal conditions, these spin axes
are pointing in random directions such that the net magnetic moment M, the sum
of all spins, is zero.
In the presence of an external magnetic eld, however, these spins exhibit three
interesting properties.
5
6
Chapter 2. Magnetic Resonance Imaging
Bo
a) no external
magnetic field
b) polarized by
magnetic field
Figure 2.1: Polarization of magnetic spins in the presence of an external magnetic
eld. When there is no applied magnetic eld, (a) spins are oriented randomly. In
the presence of an external eld, (b) spins orient either parallel or antiparallel to
the external eld with probabilities dened by the Boltzmann distribution.
Spins tend to align either parallel or antiparallel to the applied eld,
with a slightly greater number in the parallel position, thus resulting in a
net magnetization or magnetic moment. This process is commonly known as
polarization or alignment, as shown in Figure 2.1.
The ratio of parallel n" to antiparallel n# spins is given by the Boltzmann
distribution:
Alignment:
n"
n#
= e kTE ;
(2.1)
From this we see that the net magnetization is related to the the parallel/antiparallel energy dierence E (which is linearly related to the strength of the
applied eld), and inversely related to the absolute temperature. This net
magnetization is commonly termed M0 .
Resonance: The nuclear spins also possess a resonant frequency proportional to
the magnitude of the applied magnetic eld. This frequency is known as the
Larmor frequency:
7
Chapter 2. Magnetic Resonance Imaging
Mz
Mxy
Mo
Mo exp(-t/T2)
Mo (1-exp(-t/T1))
t
t
Figure 2.2: Relaxation of magnetic spins in the presence of an external magnetic
eld. From an initially perturbed position, spins relax longitudinally (in the direction of the applied eld) based on T1 and relax in the transverse plane based on
T2
f
= 2 B
(2.2)
which is the product of the gyromagnetic ratio (an atomic property) and
the magnetic eld strength B. This means that when spins are perturbed from
their preferred axis of alignment, they rotate about this axis at the Larmor
frequency.
Relaxation: Finally, when perturbed from equilibrium alignment, the net magnetic
moment from a collection of spins relaxes towards its equilibrium value based
on two time constants (as shown in Fig. 2.2). The longitudinal relaxation rate
T1 species the relaxation rate in the direction of the applied eld (usually
denoted ^z), while the transverse relaxation rate T2 species the relaxation
rate in the perpendicular plane (usually denoted as the x^ y^ plane). Both
relaxation rates vary in dierent tissues, as discussed later in Section 2.4.1.
The above three properties of the net magnetization vector M are concisely
characterized by the Bloch equation:
dM
dt
= (M B)
Mx x
^ + My y^
T2
(Mz
M0 )^
z
T1
;
(2.3)
8
Chapter 2. Magnetic Resonance Imaging
z
z
z
RF modulated
x,y
Bo
x,y
Bo: Polarizing Field
x,y
G: Linear Gradients
RF: radiofrequency field
Figure 2.3: Magnetic elds used in magnetic resonance imaging.
where where x^, y^, and ^z are unit vectors along the x, y, and z axes, M0 represents the
equilibrium magnetization and B represents the net magnetic eld vector. From this
equation, we can observe the main spin behaviors: 1) the equilibrium magnetization
M0^
z, 2) the precession of individual spins around the net magnetic eld vector B
at the Larmor frequency f = 2 jBj, and 3) the relaxation of spins towards the
equilibrium magnetization based on time constants T1 and T2.
2.2
Components of an Experiment
In a typical imaging setup, three specialized magnetic elds (shown in Figure 2.3)
are used:
B0 :
a strong homogeneous and constant magnetic eld in ^z, usually between 0.2
to 3 Tesla for clinical systems. This eld is used to polarize nuclear spins
(as shown in Figure 2.1b), and to provide the resonance condition during
excitation and reception.
B1 or RF: a radiofrequency (RF) eld in the x
^-y^ plane tuned to the Larmor frequency of the species of interest. This time varying eld is used during excitation to perturb magnetic spins out of their equilibrium aligned state.
Chapter 2. Magnetic Resonance Imaging
G:
9
three linear gradients in ^z that depend linearly on position in x^, y^, and ^z
respectively. These gradients are small in magnitude compared to B0. They
are usually on the order of 10 to 40 mT/m, and are used to spatially encode
position in spins' resonant frequency. i.e. f (r) = 2 (B0 + G r).
2.3
Stages in MRI
A typical MRI experiment consists of four stages: polarization, excitation, acquisition, and image reconstruction. Polarization, excitation, and acquisition require the
specialized hardware described in the previous section, while image reconstruction
is typically done on a conventional workstation.
2.3.1
Polarization
Polarization is the process of aligning spins along a preferred direction in order to
create a net magnetic moment from collections of spins. In conventional MRI, the
strong B0 eld is kept on during an entire scan to provide the necessary polarization.
In addition, this eld establishes a resonance condition that is important during
excitation and acquisition.
2.3.2
Excitation
Excitation is the process of disturbing spins from their equilibrium orientation along
B0 so that they can be imaged. Since spins resonate at a well dened frequency
(the Larmor frequency), a modulated magnetic eld in the transverse plane tuned to
that frequency eectively tips the magnetization vector out of its equilibrium. This
modulated magnetic eld is usually called B1 , or RF, because its frequency usually
falls within the radiofrequency band. The path of the magnetization M during such
an excitation is shown in Figure 2.4.
Chapter 2. Magnetic Resonance Imaging
10
For most applications, we are only interested in imaging a small region instead of
the whole body. This is usually done by selectively exciting just the region of interest.
In the case of single slices, this can be done by accompanying a band-limited RF
excitation with a linear gradient in the slice select direction (direction perpendicular
to the slice plane). As shown in Figure 2.5, the linear gradient creates a resonant
frequency that varies linearly in space, while the band-limited RF pulse excites a
set of frequencies that is conned to a slice. The shape of the RF pulse's frequency
response is identical to the excitation slice prole. Similar Fourier analogies can also
be used to excite arbitrary volumes, specic velocities, or spectral species [19,20].
Other than slice-selective excitations, the most relevant excitation to this thesis
is the spectral-spatial excitation. This excites a limited bandwidth within a slice,
and is commonly used to suppress fat in single slices. Such pulses are thoroughly
described by Meyer et al. [21].
2.3.3
Acquisition
Once the desired region is excited, spins within the region begin precess and relax
towards equilibrium. By reading the signal from an RF receiver coil, we sample the
following data, often called a free induction decay (FID):
S (t ) =
Z
r
M (r; t)dV
(2.4)
By using linear gradients during reception, a linear frequency shift as a function
of space can be applied, thus resulting in a linear phase that accumulates over time.
This results in the following signal equation:
S (t) =
Z
r
M (r; t)e
i
Rt
0 G( )rd dV;
(2.5)
11
Chapter 2. Magnetic Resonance Imaging
Figure 2.4: The trajectory of the locus of magnetization M during a 90o excitation.
position
Slice Selective
Excitation
linear relationship
established using
a constant linear
gradient G
resonant
frequency f
Bandlimited
RF Pulse
Slice Profile
Figure 2.5: Slice selective excitation.
12
Chapter 2. Magnetic Resonance Imaging
where G is the linear gradient term. For simplicity we now assume that the excited
region is a slice in the x^-y^ plane, and that M (r; t) is constant during an acquisition.
This encoding then reduces to Fourier encoding:
S (t) =
Z
x;y
M (x; y )e i2(kx (t)x+ky (t)y) dxdy;
(2.6)
where kx(t) = 2 R0t Gx( )d and ky (t) = 2 R0t Gy ( )d . Notice that data collected
at certain kx and ky values are the kx; ky spatial frequencies of the object M (x; y). In
eect, by playing carefully designed linear gradients G, we can sample the Fourier
transform of the object M (x; y). This Fourier domain is commonly known as kspace.
k-space
By designing appropriate gradients, k-space samples can be acquired and then inverse Fourier transformed to obtain an image of the magnetization M (x; y). k-space
must be suÆciently sampled according to the Nyquist criterion to avoid object domain aliasing. The extent of k-space coverage determines the image resolution.
Relationships between eld-of-view (FOV) and resolution (x) shown in Figure 2.6
are as follows:
FOV = F OVx = 1k
resolution = x = 2k1
max
(2.7)
(2.8)
To acquire images with a desired FOV and resolution, k-space must be suÆciently sampled based on the above constraints. However these requirements do not
restrict the particular order or trajectory in which samples are acquired. Figure 2.7
illustrates three common k-space sampling trajectories. The most common sampling
trajectory, called two-dimensional Fourier transform (2DFT), involves acquiring one
13
Chapter 2. Magnetic Resonance Imaging
k-space
object space
∆x = 1 / (2*kmax)
∆k
Fourier
transform
2 * kmax
FOV = 1 / ∆k
Figure 2.6: Illustration of the Fourier relationship between k-space and object space.
14
Chapter 2. Magnetic Resonance Imaging
2DFT
Echo Planar
Spiral
Figure 2.7: Popular k-space trajectories.
scan line of k-space per acquisition. Both echo planar imaging (EPI) [22] and spiral
imaging [23] trajectories can acquire the necessary k-space samples in fewer excitations. EPI acquires multiple k-space lines during each excitation, while spirals
traverse k-space in a spiral pattern.
2.3.4
Image Reconstruction
When acquiring k-space samples on a rectangular grid, image reconstruction can be
simply performed with an inverse fast-Fourier transform (FFT) [24] of the acquired
data, since the image M (x; y) is the inverse Fourier transform of the signal as described in Eqn. 2.6. Other non-Cartesian sampling schemes (such as spirals) can be
reconstructed using a modied transform. However, for computational eÆciency,
data is usually gridded [25,26] or interpolated to a regularly spaced Cartesian grid
and then inverse Fourier transformed. This is primarily due to the speed of the FFT
algorithm.
Taking advantage of the fact that the image should be real in the object domain,
it is also possible to use approximate hermitian symmetry in k-space to reduce
acquisition times. These methods are called partial k-space methods, which require
special reconstruction algorithms [27,28].
Chapter 2. Magnetic Resonance Imaging
2.4
15
Imaging Goals
MR imaging technology can be used to noninvasively image any non-magnetic and
non-metallic sample containing nuclear magnetic spins. The most common applications for MRI are in medicine, where the goal is to visualize and identify diagnostically signicant abnormalities in the body. Hydrogen (1H) is the most abundant
nuclear spin in the human body, and is the subject of all imaging presented in this
thesis.
2.4.1
Contrast
One primary imaging goal is to establish image contrast between dierent tissues.
For example, when trying to identify cancerous regions in an image there should
be some image contrast between cancerous and non-cancerous signal. MRI pulse
sequences are designed specically to provide image contrast based on a variety of
tissue parameters including proton density, chemical shift, T1, and T2.
Proton density is the most basic type of image contrast. As the signal levels are
proportional to the concentration of 1H, regions with higher density of proton spins
exhibit greater signal. Contrast based on T1 and T2 can also be achieved by adjusting the time spacing between excitation, acquisition, and subsequent acquisitions.
Contrast can also be created by exciting only certain species, or suppressing certain
spins.
f
T1
T2
muscle 0 Hz 850 ms
45 ms
blood 0 Hz 1000 ms 220 ms to 50 ms
fat
-220 Hz 250 ms
60 ms
Table 2.1: Approximate tissue chemical shift and relaxation parameters at 1.5 T.
Note that the T2 of blood varies with oxygenation, and for fully oxygenated blood
it is around 220 ms.
Chapter 2. Magnetic Resonance Imaging
16
In this thesis, contrast is predominantly created by inow or by contrast agents.
Inow contrast comes from the fact that spins that are stationary in the imaging
slice have a lower steady state signal level compared to owing spins. Blood pool
contrast agents increase signal in areas that they reach, therefore producing high
signal levels at places where blood delivers the contrast agent.
2.4.2
Speed and SNR
Two closely related concerns are imaging speed and signal-to-noise ratio (SNR).
Imaging speed and SNR both relate to spin relaxation parameters as well as the
design and timing of imaging sequences. SNR is also proportional to the square root
of total readout time. Therefore, an appropriate measure of signal is SNReÆciency
which is a combination of SNR and scan time:
SNReÆciency
2.4.3
=p
SNR
readout time
(2.9)
Artifact Management
Fast imaging should also be done with minimal artifacts. Common causes of MR
image artifacts include spin ow eects, B0 eld inhomogeneity, hardware eddy
currents, and spectral shifts of dierent magnetic species.
Two sources of artifacts of particular relevance to this dissertation are oresonance and ow. Both can be most easily described by their eects on the
signal equation:
Shifts in the resonant frequency of spins are often caused by B0
eld inhomogeneity, chemical shift, and susceptibility dierences at tissue borders. In general o-resonance can be represented by a shift f in the precession frequency that is a function of spatial position. This reects in the signal
equation as:
O-resonance:
17
Chapter 2. Magnetic Resonance Imaging
S (t) =
Z
x;y
M (x; y )e i2f (x;y)t e i2(kx (t)x+ky (t)y) dxdy
(2.10)
Assuming that all of k-space is acquired at the same time t, there would
be no artifacts in a reconstructed magnitude image. However, when using
long readouts, o-resonant phase accrues during each acquisition. Thus image
artifacts depend on the acquisition trajectory. In 2DFT imaging, o-resonant
spins appear slightly shifted in the readout direction. In EPI imaging, these
spins appear shifted in the phase encode direction, resulting in a geometric
distortion. In PR and spiral imaging, o-resonant spins appear blurred.
Flow: Flowing spins also accrue dierent phase as they move during excitation and
readout gradients, inheriting a dependence on gradient moments. This causes
signal phase shifts that are a function of velocity, acceleration, etc. and the
gradient waveforms up to that point in time. First, reconsider signal Equation
2.5:
S (t)
=
=
Z
Zr
r
Rt
(2.11)
( ) dV;
(2.12)
(2.13)
M (r; t)e
i
M (r; t)e
i r;t
0 G( )r( )d dV
where (r; t) = R0t G( ) r( )d . Notice that spin movement during a linear
gradient changes the received signal phase of that spin. Further expanding
(r; t) reveals:
(r; t)
=
=
Z t
2
(2.14)
2 a + d
i
(M0g (t) ro) + (M1g(t) v) + (M2g (t) a2 ) + ; (2.15)
h0
G( ) ro + v +
18
Chapter 2. Magnetic Resonance Imaging
where M0g (t), M1g (t), and M2g (t) represent the zeroth, rst, and second
moments of the gradients G(t).
S (t) =
Z
h
x;y
M (x; y )e
[(
i
( ) )+(M2g (t) a2 ) ] e i2(kx (t)x+ky (t)y) dxdy
i M1g t v
(2.16)
In addition to the zeroth moment of the gradients M0g (t) encoding position,
higher order gradient moments encode velocity, acceleration, and higher-order
motion into image phase. This results in additional artifacts depending on the
acquisition scheme. In general, to limit ow artifacts, trajectories are designed
such that all gradient moments are zero when the k-space origin is acquired.
This is because the central k-space region contains most of the image energy,
so it is most important to keep those data points undistorted. See Appendix A
for further details.
2.5
Experimental Setup
All experiments described in this thesis were conducted on a GE Signa 1.5 Tesla
CV/i scanner (General Electric, Milwaukee, WI) with gradients capable of 40 mT/m
magnitude and 150 mT/m/ms slew rate, and a receiver supporting 4 s sampling
(125 kHz).
The techniques discussed in Chapters 3, 4, and 5 were implemented using the
Stanford real-time interactive (RTI) imaging environment originally developed by
Kerr and Pauly et al. [29]. This system uses an external workstation for real-time
reconstruction and display of images, and provides an intuitive user interface for
manipulating the scan plane and modifying sequence parameters. It also supports
a variety of excitation pulses and k-space acquisition schemes. Figure 2.8 contains a
system block diagram of the Stanford RTI system, while Figure 2.9 shows the user
interface.
19
Chapter 2. Magnetic Resonance Imaging
Raw Data
X-windows
Workstation
Signa 1.5T
CV/i scanner
Pulse Sequence
Control
Reconstruction Servers
Figure 2.8: Block diagram of the Stanford RTI system. An external workstation
provides real-time parallel reconstruction, display, and scan control.
Figure 2.9: User interface of the Stanford RTI system. X-windows panels provide
interactive real-time control over the scan plane, eld of view, ip angle, and other
important scan parameters.
Chapter 3
Real-Time Color Flow MRI
The development of color ow mapping in ultrasound revolutionized the diagnosis
and assessment of valvular heart disease. Over time, ultrasound has proven to be
particularly eective at visualizing regurgitant jets and abnormal ow [30]. It is
desirable to have this capability in an MR exam because MR imaging has several
unique advantages over ultrasound. In MR, there are no acoustic window constraints, which gives the operator a great variety of achievable views. MR can also
be used to examine velocity in any direction including through-plane. Sequences
can be tuned to any desired velocity range, whereas ultrasound typically is limited
to 70 cm/s before aliasing interferes.
MR has traditionally suered from long scan times, making it diÆcult to image
in the presence of ow and motion. Cardiovascular imaging has been particularly
diÆcult due to cardiac and respiratory motion. To avoid motion artifacts, scans are
often gated, or require breath-holds. With improvements in gradient hardware, novel
k-space coverages, and new reconstruction techniques, real-time dynamic imaging,
or MR uoroscopy has developed into a popular and robust way to avoid motion
artifacts. MR uoroscopy was rst proposed by Riederer in 1988 [31,32], and has
recently been revisited by Kerr, Rache, Hardy, Gmitro, and others [29,33{36]. Image
rates of up to 30 frames per second have been reported, making dynamic MRI of the
heart and abdomen possible. Real-time MR has also demonstrated its diagnostic
20
Chapter 3. Real-Time Color Flow MRI
21
capability, particularly when examining cardiac wall motion and left ventricular
(LV) mass [37], and when tracking motion in the small bowel [38].
The color overlay of ow information over anatomical information has a long
history in ultrasound and other imaging modalities. It has been used in MR since
Klipstein et al. [39], and has received renewed attention as a powerful aid for visualizing ow information and anatomical information simultaneously [40,41]. Dynamic
color ow MRI was rst presented in 1990 by Riederer et al. [42], with frame rates
up to 1 image/s.
In this chapter, we describe a system that improves upon those previously reported by providing the following key features: 1) reduced ow artifacts due to spiral
k-space coverage, 2) interactivity with improved response time, and 3) higher temporal resolution, with image rates up to 9 images/s and reconstruction and display
rates up to 18 frames/s (suÆcient to resolve cardiac ow).
The system described here was implemented within the context of the Stanford
RTI system described in Section 2.5. We use a fast spiral phase contrast pulse sequence [43,44], real-time reconstruction to compute both velocity and density maps,
and intelligent color overlay. No breath holding or gating is required. Interactive
controls are provided for examining arbitrary scan planes, arbitrary ow directions,
and arbitrary velocity ranges. Results from phantom studies, as well as human
studies are presented.
3.1
3.1.1
Method
Pulse Sequence
We elected to use a simple excitation-recovery phase contrast sequence because it
results in short acquisition times and provides velocity as well as density information.
As shown in Figure 3.1, excitations are followed by ow-encoding gradients, spiral
22
Chapter 3. Real-Time Color Flow MRI
Water Selective Excitation
elocity Encoding (VENC)
RF
Slice Select Gradient
Crusher
Spiral Readout
Readout Gradients
Figure 3.1: Spiral phase contrast pulse sequence used in real-time color ow MRI: A
low ip-angle excitation (water-selective excitation is shown) is followed by velocityencoding bipolar gradients in x^, y^, and ^z which provide ow encoding along a
prescribed direction. Finally, spiral-interleaved readouts are used to acquire k-space,
and a gradient spoiler dephases any remaining signal.
readouts, and a gradient spoiler. For most applications, a water-selective spectralspatial excitation (shown in Figure 3.1) is used [21]. As noted by Fredrickson et
al. [45], if suÆcient gradient strength is available, a yback excitation pulse should
be used to help provide uniform excitation in the presence of through-plane ow.
In the design of a yback pulse, RF power is delivered only whenR the gradients are
ow-compensated in the slice-select direction (when M1z (t) = 0t Gz ( ) d = 0).
For those applications in which a shorter TR is required, we opt for a purely sliceselective excitation [46], which occupies on the order of 2 ms.
Spiral interleaved acquisitions are used to collect image data. Spiral trajectories are chosen because they cover k-space eÆciently [23], and have excellent ow
properties [47,48]. Specically, owing spins produce blur in the direction of ow,
instead of ghosting which is seen in other fast imaging trajectories such as echoplanar (EPI) [47]. The readouts themselves are generated to make optimal use of
available gradients [49,50].
23
Chapter 3. Real-Time Color Flow MRI
The pulse sequence ends with a gradient spoiler in the slice-select direction which
dephases any remaining transverse signal. For reference, magnitude images from this
sequence possess contrast similar to GRASS.
This sequence also includes a velocity-encoding (VENC) bipolar gradient that
occurs between the excitation and readout. The introduction of such a gradient
(with a nonzero rst moment M1 ) produces image domain phase osets proportional
to velocity [51]. As described in Section 2.4.3 the phase oset at the echo time is
linearly dependent on the component of velocity in the direction of gradient ow
encoding:
= vd (M1 )
M1
=
Z TE
0
Gd ( ) d ;
(3.1)
(3.2)
where vd is a voxel's velocity component in the ow encoding direction, is the
gyromagnetic ratio, and is that voxel's corresponding phase oset; Gd(t) is the
gradient in the ow encoding direction, and M1 is the rst moment of the gradient
waveforms in that same direction. Our velocity encoding gradient consists of two
trapezoids equal and opposite in area. In this case, M1 = A , where A is the area
under one trapezoid, and is the time between the centers of the two trapezoids.
Given this, the maximum velocity discernible from a particular VENC is the velocity
that would produce an overall phase shift of :
vmax
= M
1
(3.3)
This maps the velocity range [ vmax ; vmax ] to the phase range [ ; ]. True
velocities outside of the [-vmax,vmax ] range will result in aliased estimates. Signal
with true velocity 2 vmax will appear stationary, and signal with true velocity
1 41 vmax will appear to have velocity 34 vmax . To ensure the accuracy of velocity
measurements, the chosen velocity range should be large enough to avoid this eect.
Chapter 3. Real-Time Color Flow MRI
24
If the range chosen is too large however, phase noise will dominate the velocity
signal, and velocity estimates will be unreliable. For cardiac valve studies, we use
a vmax of anywhere from 2 to 10 m/s. This reasonably covers the range of normal
and abnormal velocities we expect to nd. Note that due to the inverse relationship
between M1 and vmax , a smaller vmax requires larger M1 and therefore a longer
VENC gradient waveform. A large M1 may also result in signal dropouts due to
dephasing within voxels.
The relative scale of the VENC on the x^, y^, and ^z gradients determines the
direction of ow sensitivity. The absolute scale of those waveforms determines the
sensitive velocity range. These properties may be controlled interactively by the
scan operator to provide the desired ow sensitivity.
3.1.2
Acquisition Timing
For each color frame reconstructed, two full images are acquired, one ow-compensated
image I0 acquired with VENC o, and one ow encoded image I1 acquired with
VENC on. This is typically referred to as asymmetric encoding. Alternatively,
symmetric encoding would involve both images having equal and opposite ow encodes. Symmetric encoding has the advantage that the same vmax can be achieved
with a shorter VENC waveform, which results in shorter echo times. However,
asymmetric encoding provides a ow compensated magnitude image jI0j that is free
of ow encoding artifacts. Flow encoding often produces undesirable magnitude
variations [52, 53]. Our primary motivation for using asymmetric encoding is to
guarantee that one magnitude image has minimal ow encoding artifacts.
Figure 3.2 illustrates the timing of acquisitions. Acquisitions for I0 and I1 are
interleaved, so that each spiral interleaf is acquired once without and once with ow
encoding, before moving on to the next interleaf. This is done to maintain temporal
coherence between data that will be compared. In addition, if the number of interleaves is four or more, the interleaves themselves can be executed in a shued or
bit-reversed order to maximize time between adjacent interleaves, thereby reducing
25
Chapter 3. Real-Time Color Flow MRI
Flow
Encoding
Gradient:
Spiral
Readout
Gradient:
Interleaf 1
Interleaf 2
Interleaf 3
Interleaf 1
Figure 3.2: Acquisition and sliding window reconstruction timing for real-time color
ow MRI: Interleaves with and without ow encoding are acquired continuously.
Each interleaf is acquired twice (without and with ow encoding). At any given
time, the most recent complete data set is used to reconstruct density and velocity
images.
motion induced spiral-shape artifacts [54]. The cycle of measurements is repeated
indenitely to permit a sliding window reconstruction [31].
3.1.3
Reconstruction
Flow compensated image I0 and ow encoded image I1 are formed from spiral raw
data via gridding reconstruction [26]. Velocity and density maps are computed from
these images as shown in Figure 3.3. The density map is taken as the magnitude of
the non-ow-encoded image I0, while the velocity map is computed from the phase
dierence [55] between the two images I0 and I1 .
D = jI0 j
V
(3.4)
(I1I0 )
= argM
1
(3.5)
26
Chapter 3. Real-Time Color Flow MRI
Image 0
Magnitude
non
flow encoded
Density
Image 1
Phase
Difference
flow encoded
Velocity
Color Overlay
Figure 3.3: Color ow MRI reconstruction: density and velocity maps are computed
from the acquired images. The density image is taken as the magnitude of the ow
compensated image I0 ; and the velocity image is inferred from the phase dierence
between the two images I0 and I1 . Density and velocity information is then combined
on a pixel-by-pixel basis to produce a color ow image. In this example of aortic
regurgitation, the color overlay image depicts the velocity prole of a regurgitant
jet while providing valuable anatomical landmarks.
Only I0 is used for the density map because asymmetric encoding results in I0
having fewer magnitude artifacts than I1. However, if symmetric encoding is used,
the magnitude images can be averaged to improve SNR [56].
3.1.4
Display
There are several methods for viewing real-time density and velocity images. Our
rst implementation involved displaying density and velocity video side-by-side,
however, this made it diÆcult to relate the positions of ow features in velocity
images to anatomical features in density images. We therefore employed a color
overlay that is similar to what is used in color ow ultrasound.
To produce a nal color overlaid image, each pixel's density and velocity is
quantized to one byte each and then passed though a colormap (see Figure 3.4). We
use an adapted version of an ultrasound colormap (provided by the Medical Products
Group of Hewlett Packard, Palo Alto, CA) in order to make overlay images look
familiar to physicians. Figure 3.3 demonstrates the pixel-by-pixel overlay.
Chapter 3. Real-Time Color Flow MRI
b
Velocity
a
27
Density
Figure 3.4: Colormaps. The horizontal axis represents density, and the vertical axis
represents the range of velocities. (a) an adapted ultrasound colormap (obtained
from the Medical Products Group, Hewlett Packard, Palo Alto, CA), and (b) clipped
version of that colormap incorporating minimum density and minimum velocity
criteria.
A major concern in the overlay process is the accuracy of velocity maps. When
there is little signal in a pixel, its phase is more prone to noise, and thus the velocity
estimate is less reliable. To prevent this from appearing in overlaid images, a minimum density criteria is used. Coloring is restricted to pixels which have suÆcient
density signal, and which therefore have reliable velocity estimates. In addition,
since we are usually interested in abnormal fast ow, we further limit the colored
areas of the displayed image using a minimum velocity criteria (see Figure 3.4b).
These clipping values may be controlled interactively by the scan operator, to optimize visualization of the pathology.
The scan operator also has the option of brightening the density map, or adding
articial aliasing to the velocity map, as shown in Figure 3.5. The option of articial aliasing was included mainly for the benet of clinicians accustomed to color
discontinuities (as in ultrasound). Since this is done in a reconstruction step, our
28
Chapter 3. Real-Time Color Flow MRI
a
b
c
Figure 3.5: Articial velocity aliasing can be added to an image to make fast ow
more visible. The same frame of data is reconstructed with (a) no articial velocity
aliasing, and (b-c) increasing amounts of articial velocity aliasing. The vmax at
acquisition was 240 cm/s but the eective vmax for coloring is 160 cm/s in (b) and
80 cm/s in (c).
articial aliasing preserves the direction of ow information, unlike true aliasing.
Therefore the colors are wrapped from sky blue to dark blue and yellow to dark red.
3.2
Results
Results from phantom and human studies are presented. All MR images were acquired on a 1.5 T GE Signa scanner with gradients supporting 40 mT/m magnitude
and 150 mT/m/ms slew rate. A 5-inch surface coil was used for signal reception,
and a body coil was used for transmission of RF. In addition, the color ow sequence
was run with a 30-ms TR, 30o ip angle, asymmetric encoding, 242-cm/s vmax , using a 7-ms conventional spectral-spatial excitation, 1-ms velocity encode, and 16-ms
spiral readouts. The sequence used 3-interleave spirals to achieve 2.4-mm isotropic
resolution over a 20-cm FOV, and acquire data for a complete color ow image every
180 ms (approximately 6 images/s), or used 2-interleave spirals to achieve 2.92-mm
resolution over a 20-cm FOV, every 120 ms (9 images/s).
Chapter 3. Real-Time Color Flow MRI
29
The primary application of interest for this technique is the visualization of cardiac and vascular ow. This typically involves orienting the scan plane such that
the ow of interest is either in-plane or through-plane, and then prescribing the
corresponding ow-encoding direction. Flow-encoding directions that are a combination of in-plane and through-plane are not typically used but are supported by
the system.
3.2.1
Phantom Studies
Two sets of ow phantom experiments were performed to evaluate the performance
of this technique at measuring through-plane and in-plane ow. In the rst experiment, velocity measurements were taken on a steady-ow phantom. This phantom
consisted of a straight tube passing through a tub of distilled water. Water doped
with manganese-chloride (T1 of 640 ms, T2 of 90 ms) was fed through the tube
by a constant-rate ow pump (Masterex model 7520-25; Cole-Parmer Instrument
Company, Chicago, IL). Reference velocities were measured using a standard phase
contrast sequence (2DFT, 5.4-ms TE, 30-ms TR, 30o ip angle, 250-cm/s VENC,
11 s imaging time). It has previously been shown that standard phase contrast is an
accurate reference for velocity measurement [57,58]. Velocities were also measured
with real-time color ow (30-ms TR, 30o ip, 242-cm/s vmax , 180-ms imaging time)
both when the ow was oriented in-plane and through-plane. Measured velocities
were compared over the range from 40 cm/s to 200 cm/s, and results are shown
in Figure 3.6. The plotted quantities are mean velocities over a cross-section of
the tube. Over the full range, real-time through-plane and in-plane measurements
were within 5% of reference velocities. At higher velocities, through-plane realtime measurements tended to underestimate the velocity. This could be due to fast
through-plane ow not experiencing the full excitation or the fact that a yback
spectral-spatial excitation was not used. The standard deviation of through-plane
real-time measurements was less than 2.5 cm/s at all measured pump setting. This
error is just larger than the quantization error of saved velocity data. Velocity information from -242 cm/s to 242 cm/s being quantized to one byte, results in a
Chapter 3. Real-Time Color Flow MRI
30
velocity resolution of 1.8 cm/s. In-plane ow measurements experienced more variability over time, with a standard deviation of up to 10 cm/s. These results indicate
that velocities measured using this system are accurate up to at least 2 m/s.
The second phantom experiment was conducted to evaluate the ability of this
technique to capture real-time velocity waveforms. The phantom consisted of a loose
tube in a closed loop driven by a pulsatile ow pump (Pulsatile Blood Pump, model
1421; Harvard Apparatus, South Natick, MA). The pump was set to 80 cycles per
minute with a 28 cc stroke volume. This phantom was rst lled with water and
Albunex (human albumin, sonicated; Molecular Biosystems, San Diego, CA), and
examined using 1D continuous-wave Doppler ultrasound (CWUS) on an HP Sonos
2500 (Hewlett Packard, Palo Alto, CA) using a 1.9 MHz probe. The measured
waveform is shown in Figure 3.7a. Following this, the phantom was lled with
manganese-chloride doped water, and scanned in real-time using MR with the same
scan parameters as listed above. The scan plane was oriented parallel to ow for inplane ow measurements, and was oriented perpendicular to ow for through-plane
ow measurements. Real-time measurements indicated plug ow was achieved in
the phantom. In a post-processing step, the average of a few pixels from the center
of the tube was chosen as representative, and plotted as a function of time to yield
the waveforms shown in Figure 3.7b-c. These results indicate that this system is able
to accurately capture the magnitude and shape of a real-time velocity waveform.
3.2.2
In Vivo Studies
Our primary application has been the imaging and visualization of cardiac valvular
regurgitation. We have qualitatively evaluated the performance of the real-time
color ow imaging sequence in several normal volunteers and patients with valvular
disease.
The interactive nature of this system enabled the quick localization of scan planes
of interest and quick visualization of relevant ow. The scan plane, eld of view,
31
Chapter 3. Real-Time Color Flow MRI
Real-Time Color Flow Velocity (cm/s)
240
200
160
120
80
through−plane
in−plane
40
40
80
120
160
200
240
Reference Velocity (cm/s)
Figure 3.6: Constant velocities produced by a steady-ow phantom and measured
using real-time color ow and reference phase contrast. Through-plane and in-plane
ow measurements are shown separately. (dotted line is the ideal y=x line)
Chapter 3. Real-Time Color Flow MRI
32
a
b
c
Figure 3.7: Pulsatile ow velocity waveform measured with (a) continuous-wave
Doppler ultrasound; and real-time color ow using (b) through-plane and (c) inplane measurements. The velocity window for all three is -60 to 60 cm/s. MR and
ultrasound experiments were conducted at dierent locations with the same phantom and same ow parameters. Note that the ultrasound data is one-dimensional,
while the MR ow data is two-dimensional with the plotted quantity describing the
average of measured velocities within the cross-section of ow imaged.
Chapter 3. Real-Time Color Flow MRI
33
slice thickness, ow encoding direction and magnitude were all interactively controlled. Although arbitrary scan planes could be reached, for comparison purposes,
we present mainly views commonly used in echocardiography. Figure 3.8 shows stillframe comparisons of aortic and mitral regurgitation visualized with real-time color
ow MR and color ow ultrasound. The aortic regurgitation is depicted in threechamber views (Figure 3.8a-b), and mitral regurgitation is depicted in four-chamber
views (Figure 3.8c-d). Both MR images were acquired with ow encoding in the vertical direction. Regurgitant jets are clearly visualized in both MR and ultrasound;
however, the jet is visualized closer to the valve in the ultrasound images.
Image sequences from a patient with aortic regurgitation and mitral regurgitation
are shown in Figure 3.9. The aortic regurgitation is visualized throughout diastole
in series A; and the eccentric mitral regurgitant jet in series B is visible during the
two frames of systole. The images shown are separated by 180 ms which is the true
image rate. The real-time display however operates at up to 18 frames/s using a
sliding window reconstruction as described earlier. In our patient studies to date,
this temporal resolution has been suÆcient to resolve normal cardiac ow as well as
regurgitant ow.
Real-time color ow can also be used in other areas of the body where dynamic
ow information is useful and important. Here, we demonstrate this system's ability
Cardiac
Carotid Popliteal Coronary
Receive Coil
5" surface 3" surface extremity 5" surface
Field of View (cm)
20
12
12
24
Slice Thickness (mm)
7
5
5
7
Resolution (mm)
2.4
1.46
1.50
2.4
Velocity vmax (cm/s)
242
100
50
100
Repetition Time T R (ms)
30
44
44
30
Number of Interleaves
3
3
3
3
Image Time (ms)
180
264
264
180
Table 3.1: Scan parameters used in color ow MRI.
Chapter 3. Real-Time Color Flow MRI
a
b
c
d
34
Figure 3.8: Comparison of ultrasound and real-time color ow MRI: Three chamber views showing aortic regurgitation imaged in a patient using (a) color ow
ultrasound, and (b) real-time color ow MR. Four chamber views showing mitral
regurgitation in a patient using (c) color ow ultrasound, and (d) real-time color
ow MR. Regurgitant jets are indicated by arrows, and the respective valve planes
are indicated by dashed lines.
Chapter 3. Real-Time Color Flow MRI
35
a
b
Figure 3.9: Image sequences from three-chamber views of a patient with (a) aortic insuÆciency, and a patient with (b) mitral regurgitation. These consecutive
frames are acquired in 180 ms each, which is the true frame rate. A sliding window
reconstruction and display operates at 18 frames/s. In series a aortic regurgitation is observed throughout diasole (frames 3-5). In series b mitral regurgitation is
observed during diastole (frames 2-3).
to examine carotid ow patterns, ow in the descending aorta and peripheral vessels, and coronary ow in real-time. Table 3.1 outlines scan parameters used while
exploring these other applications. Field of view and resolution are chosen based on
the size of the anatomy, while other parameters are primarily chosen based on the
expected pulsatility and peak velocity.
Rapid detection of ow disturbance in the carotids may be helpful in examining
stenoses of intermediate grade. Images of carotid ow in a normal volunteer are
presented in Figure 3.10. This single frame from a real-time sequence immediately
indicates ow patterns around the carotid bifurcation.
Flow disturbances in the iliac aorta are thought to contribute to aneurysm formation. Figure 3.11 shows color-ow images of the iliac aorta bifurcation in a normal
36
Chapter 3. Real-Time Color Flow MRI
a
b
c
Figure 3.10: In-plane ow through the carotid bifurcation in a normal volunteer. A
single frame during systole is shown by its (a) density image, (b) velocity map, and
(c) color overlay.
a
b
c
Figure 3.11: In-plane ow through the iliac aorta bifurcation in a normal volunteer.
A single frame during systole is shown by its (a) density image, (b) velocity map,
and (c) color overlay.
37
Chapter 3. Real-Time Color Flow MRI
b
a
Figure 3.12: Through-plane ow in the popliteal artery during systole. (a) popliteal
artery, and (b) popliteal artery below rst bifurcation.
a
b
c
d
e
f
Figure 3.13: Through-plane ow in coronaries seen during diastole. left coronary (a)
density image, (b) velocity map, and (c) overlay. right coronary (d) density image,
(e) velocity map, and (f) overlay. Arrows identify coronary ow in the overlay
images.
Chapter 3. Real-Time Color Flow MRI
38
volunteer. The normal ow is clearly visualized in an examination taking less than
10 minutes, including localization.
Images of through-plane ow in the lower leg are presented in Figure 3.12. These
are frames during systole taken above and below the rst bifurcation of the popliteal
artery. This technique can be used to rapidly measure blood ow in peripheral
vessels.
Coronary ow imaging is a challenge because of vessel motion and highly pulsatile
ow [59, 60]. However, rapid imaging of ow in coronaries may be a feasible way
to detect coronary occlusion. Coronary ow is imaged in the through-plane images
shown in Figure 3.13. Because coronary ow is much slower than cardiac ow, the
majority of color in these frames is from ow in the chambers. This distraction can
be removed with a simple region-of-interest window for the coloring.
In these studies, this system has demonstrated its ability to image cardiac and
vascular ow in real-time. Patient studies have indicated that this sequence is
useful for visualizing cardiac ow particularly in and around regurgitant valves.
The interactive nature of this system also makes it useful for rapidly imaging ow
in other areas of the body.
3.3
Discussion
Two areas for improvement are temporal resolution and the visualization of fast ow
(such as that seen in the core of regurgitant jets). Temporal resolution will improve
with hardware developments, allowing the acquisition of k-space in less time, but
can also be immediately improved by sacricing spatial resolution (see Chapter 5).
Robust imaging in the presence of fast ow is also an important issue because
spins moving at high speeds may either not experience the full excitation (if they are
moving through-plane) or move signicantly during a readout (if they are moving
in-plane). Both of these can potentially cause signal reduction in areas of ow. This
can be improved by designing shorter excitations and shorter readout gradients
Chapter 3. Real-Time Color Flow MRI
39
while increasing the number of interleaves. Shorter readouts will reduce in-plane
ow eects, and a shorter excitation will reduce through-plane ow eects. See
Appendix A for more details.
In summary, we have presented a system for real-time interactive color ow imaging, which is based on spiral phase contrast acquisitions, real-time reconstruction,
color overlay and display. Flow phantom results indicate that this technique accurately measures peak velocity, and accurately captures real-time velocity waveforms.
This technique is also demonstrated as a useful tool for the rapid examination of
cardiac and other vascular ow. This technique is particularly useful for imaging
valvular regurgitation. Arbitrary scan planes and arbitrary ow directions further
enable the examination of eccentric jets. For other vascular imaging, the interactive nature of this technique is most useful. Rapid localization and ow imaging
translates into signicantly reduced examination time.
Chapter 4
Real-Time Black Blood MRI
In real-time MRI, the visualization of myocardium and vessel wall is often obscured
by signal from owing blood. For applications where ow signal and ow artifacts
are obtrusive, images can be signicantly improved by using black-blood techniques.
In addition to improving myocardial border and vessel wall denition, black-blood
techniques suppress ow-related artifacts. Achieving good blood suppression in
real-time continues to be a challenge.
Currently, the most eective blood suppression techniques used in cardiovascular
imaging are based on double inversion [61,62]. Double inversion sequences suppress
blood based on its ow and T1 properties. A non-selective 180o pulse is immediately
followed by a slice-selective 180o pulse; then a delay before imaging is chosen in
order to null signal from relaxing blood. Double inversion provides excellent blood
suppression; however, the long inversion preparation time requires a long TR and is
not feasible for real-time imaging.
Another technique involves using spatial presaturation to suppress blood [63,64].
Spatial presaturation techniques suppress blood based only on ow. Immediately
before imaging, a volume or volumes upstream from the slice (usually thick slabs
on either side of the imaging slice) are excited and dephased. Since this technique rapidly suppresses blood, it is practical for continuous or real-time imaging.
40
Chapter 4. Real-Time Black Blood MRI
41
This is also a steady-state technique which, for stationary objects, results in consistent signal intensity between measurements. While eective in the presence of fast
through-plane ow, this technique produces reduced contrast in views containing
predominantly in-plane ow [64, 65]. In the context of real-time imaging, spatial
presaturation is a practical choice because it can be done quickly and maintains a
steady state.
In this chapter, we present a real-time black-blood sequence implementation
based on spatial presaturation. Sharp proled saturation pulses are used to provide
good blood suppression even in planes containing slow through-plane ow [66,67].
The sequence was implemented within the Stanford RTI framework described in
Section 2.5. Extensions to this system include the optional volume saturation of
one or two slabs before each imaging excitation, and interactive control over the
saturation slab thickness, placement, and ip angle. No breath holding or gating is
required. We applied this technique to real-time ventricular wall motion study and
to real-time intravascular imaging.
4.1
4.1.1
Methods
Pulse Sequence
Figure 4.1 illustrates the basic pulse sequence. Each TR consists of spatial presaturation, which prepares the black-blood contrast, followed by an imaging acquisition.
The presaturation portion (Fig. 4.1a) consists of two slab-selective excitations followed by a dephaser in the slice-select direction. This has the eect of nulling signal
from the slabs, particularly the blood in the slabs which will ow into the imaging
slice.
The design of sharp-prole slab excitations allows the saturated slabs to be placed
close to the imaging slice for better ow suppression, without saturating spins in
the imaging slice. As this technique depends on saturated blood spins owing into
the imaging slice, closer saturation slabs improve black-blood performance when
42
Chapter 4. Real-Time Black Blood MRI
90˚
90˚
15˚
RF
Slice Selective
RF Sat Pulses
Water Selective
Excitation
Slice Select Gradient
Dephaser
Readout
Gradients
a)
Spiral Readout
b)
10 ms
7 ms
16 ms
Figure 4.1: Real-time black blood MRI pulse sequence: (a) spatial presaturation
consists of two slab-selective excitations followed by a dephaser in the slice-select
direction and (b) imaging consists of an excitation (water-selective excitation is
shown) followed by imaging readouts (interleaved spirals are shown) and a gradient
spoiler in the slice-select direction.
through-plane ow is slow. Our slab excitations were designed using RF design
tools by Pauly et al. [46, 68] based on the Shinnar-LeRoux technique. Simulated
proles for 2-ms and 4-ms least squares RF saturation pulses with a design bandwidth of 2 kHz are shown in Figure 4.2a, illustrating how longer pulses can produce
theoretically sharper slice proles. Figure 4.2b contains experimentally measured
proles of these pulses obtained on a static phantom. The longer saturation pulses
predictably produce a sharper prole; however, as shown by the arrow in Fig. 4.2c
(with longer pulses) the delay between the two slab excitations allows for some T1
recovery in the rst slab. The T1 recovery experienced by blood or myocardium (T1
> 800 ms) is much smaller than what is observed in this phantom (T1 90 ms).
One way to compensate for this is to use a ip angle slightly greater than 90o for
the rst slab. Another option would be to simultaneously excite both slabs with
Chapter 4. Real-Time Black Blood MRI
43
a single modulated RF pulse. This would result in the equal suppression of both
slabs, but would make it diÆcult to adjust the slab separation and slab thickness
independently and in real-time.
The imaging portion of the pulse sequence (Fig. 4.1b) consists of a water-selective
slice excitation followed by a spiral interleaved readout and a gradient spoiler in
the slice-select direction. Water-selective excitations are used to suppress signal
from lipids and interleaved spiral acquisitions provide eÆcient k-space coverage [23]
with suppressed motion artifacts [47]. While we made these excitation and readout choices for this study, the real-time interactive system supports many types of
excitations (such as water-selective, slice-selective or velocity-selective) and many
k-space acquisition schemes (such as spiral, echo planar, projection reconstruction
or 2DFT).
To maintain steady-state, the spatial presaturation is a part of every imaging
TR, thus constituting a xed cost of about 7 to 11 ms per TR. From a scan eÆciency
perspective, it is therefore benecial to use long readouts. In many cases, aggressive
readout trajectories and excitations require some pulse sequence dead-time to stay
within gradient duty cycle limits and amplier heating limits. In such cases the
spatial presaturation comes at a reduced cost because it does not use much gradient
power and can be done during some of the dead-time.
4.1.2
Experimental Methods
Experiments were conducted on a GE Signa 1.5 T CV/i scanner (General Electric
Inc., Milwaukee, WI). The pulse sequence was designed for gradients capable of
40 mT=m magnitude and 150 mT=m=ms slew rate, with a receiver capable of 4 s
sampling (125 kHz). A body coil was used for RF transmission. For signal reception, a 5-inch surface coil was used in cardiac studies and a custom designed exible
twin-lead coil [69,70] was used in intravascular studies. The black-blood sequence
was run with a 40-ms TR. The imaging portion used a 7-ms conventional spectralspatial excitation and 16.384-ms spiral acquisitions. For cardiac studies, we used an
44
Chapter 4. Real-Time Black Blood MRI
a) simulated profiles
4 ms SAT
pulses
2 ms SAT
pulses
b) measured profiles
4 ms SAT
pulses
2 ms SAT
pulses
Presaturated Slab #1
Presaturated Slab #2
Figure 4.2: Simulated and measured proles for 2-ms and 4-ms RF saturation pulses.
Pulse pairs are used to saturate spins in two slabs on either side of the imaging
slice. (a) Simulations show that more precise pulses permit the saturation slabs to
be placed closer to the imaging slice, potentially improving the suppression of slow
through-plane ow. (b) Measured proles also demonstrate that 4-ms pulses allow
saturation slabs to be placed closer to the imaging slice. However, using longer pulse
durations creates time for some T1 recovery in the rst presaturated slab (see arrow).
Using saturation ip angles greater than 90o for the rst slab can compensate for
this eect. Alternatively the two slabs can be excited using a single modulated RF
pulse.
Chapter 4. Real-Time Black Blood MRI
45
imaging ip angle of 15o which is the Ernst angle for myocardium (T1 of 800 ms) at
a 40-ms TR, and for intravascular studies, we used ip angles between 20o and 45o.
The imaging slice thickness was 5 to 7 mm in all studies. The spatial presaturation
(included in the TR) consisted of two 4-ms slab-selective excitations and a dephaser
in the slice-select direction; occupying 10 ms total. For cardiac studies, we used
2-interleave spirals to achieve 2.35-mm resolution over a 20-cm FOV, at 80 ms per
image. For intravascular studies, we used 6-interleave spirals for 500-m resolution
over a 2.4-cm FOV every 240 ms, or 20-interleave spirals for 360-m resolution over
a 3-cm FOV every 800 ms.
An external workstation was used to provide interactive control over scan plane
and imaging parameters and to provide real-time image reconstruction and display [29]. While image acquisition was at rates of 1 to 12.5 images/s, a sliding
window reconstruction [31] enabled display rates of up to 25 images/s.
The interactive nature of this system enabled the quick localization of standard
views such as short-axis and four-chamber in cardiac studies. On the user interface,
the scan plane, eld of view, slice thickness, saturation band placement and ip
angle are all interactively controlled. Note that while the saturation band thickness
and separation were controlled by the operator, certain default settings worked well
for most cases. In cardiac short-axis views, we used a default slab thickness of 6 cm
and default gap of 1.5 cm. In intravascular studies, we used a default slab thickness
of 4 cm and default gap of 1.5 cm. Slab thickness and gap spacing are dened based
on the prole half-max of the slab excitations.
4.2
Results
We have examined the utility of real-time interactive black-blood MRI in two different applications: ventricular wall motion assessment and intravascular imaging.
Chapter 4. Real-Time Black Blood MRI
4.2.1
46
Wall Motion
One potential application for this technique is in the evaluation of left ventricular function. Real-time white-blood techniques have proven their ability to detect
wall motion abnormalities in large patient populations [37] without gating or breath
holding. In a typical real-time wall-motion study, short-axis movies are acquired
at various levels of the left ventricle (from apex to base), providing coverage of
all relevant wall segments. In addition, the left ventricle may be segmented to
accurately estimate end-diastolic and end-systolic volumes from the multiple slice
data [71]. One documented problem area for white-blood techniques is the often
blurred blood-myocardium border during systole [37]. Complex ow can present
reduced blood signal, and therefore reduced contrast at the endocardial border.
One way to eliminate this artifact is to employ black-blood techniques to image
the myocardium directly while suppressing blood and its resulting artifacts. Since
short-axis views involve a signicant amount of through-plane ow, spatial presaturation methods may be particularly eective. Using 4-ms saturation pulses, and
two shot spiral acquisitions, we evaluated this black-blood technique in normal volunteers. Nine healthy subjects were scanned using white-blood and black-blood
real-time MRI, with representative short-axis images shown in Figure 4.3. White
blood images were acquired with an imaging TR of 30 ms and ip angle of 30o,
while black blood images were acquired with an imaging TR of 40 ms (longer because of presaturation) and ip angle of 15o. From these images, one can observe
the receiver coil sensitivity prole as well as the regional contrast improvement from
using the black-blood technique. In Fig. 4.3a-b, arrows identify improved endocardial denition around the posterior lateral wall during systole. This wall segment
is particularly diÆcult because it moves the most and is furthest from the surface
receiver coil. In Fig. 4.3c, the arrow identies poor white-blood contrast during
systole due to signicant myocardial through-plane motion. This depiction is also
improved with the velocity sensitive black-blood technique. Note that in-plane myocardial motion may also cause blurring at the border and reduced contrast, and
Chapter 4. Real-Time Black Blood MRI
47
this can be improved by shortening image acquisition times, which typically comes
at the cost of spatial resolution.
A contrast-to-noise ratio (CNR) analysis [72] revealed that blood-myocardium
contrast in these problem areas was signicantly higher in black-blood studies. CNR
myocardiumj using regions of interest around the posterior mewas calculated as jSblood Snoise
dial wall segment which was not biased by the coil sensitivity pattern. Problematic
white-blood studies had an average blood-myocardium CNR of 1.02, while blackblood studies had an average blood-myocardium CNR of 5.27 in the same regions.
In our single-coil studies, the added bias of the coil sensitivity pattern resulted in
additional black-blood CNR improvements in anterior wall segments, and CNR reductions in posterior lateral segments. Other coil arrangements may be used to
change this spatial variance. We also observe that while the contrast improvements
can be observed in single frames (as shown in Fig. 4.3), the qualitative improvements
are more dramatic when images are viewed in real time or in a CINE loop.
4.2.2
Intravascular Imaging
Intravascular imaging is another potential application for this technique. When
imaging with intravascular coils, vessel/coil motion and blood ow cause signicant
image artifacts with prolonged scan times. For this reason, attempts have been
made at real-time intravascular imaging [69]. Typically in these studies, vessel cross
sectional images are acquired using an intravascular coil placed within a blood vessel.
Artifacts are often disruptive to the point that vessel occlusion is necessary for wall
imaging [73]. By suppressing blood (which is the dominant signal in white-blood
images) there may be reduced artifacts and greater signal dynamic range available
for the vessel wall.
A ow phantom experiment was performed to evaluate the quality of ow suppression this technique could provide in a controlled setting. Real-time images were
acquired using a constant rate ow phantom. The phantom consisted of a straight
plastic tube with an outside layer of silicone rubber (to provide signal). Water doped
48
Chapter 4. Real-Time Black Blood MRI
Diastole
mid-Systole
end-Systole
a
b
c
Figure 4.3: Short axis comparison of black-blood and white-blood techniques during
end-diastole, mid-systole, and end-systole. (a-b) In these volunteers, the white arrow
identies improved endocardial denition at the posterior lateral LV wall. Also
notice the coil sensitivity pattern. (c) In this large volunteer, notice the high noise
level and signal drop-o from the large eld of view. The white arrow identies
poor white-blood contrast during systole due to signicant myocardial throughplane motion.
49
Chapter 4. Real-Time Black Blood MRI
silicone
trapped
fluid
lumen
plastic
a
b
Figure 4.4: Intravascular coil ow phantom experiment. Doped water is owing at
a constant rate through a plastic tube coated with a layer of silicone. Images were
acquired (a) without ow (b) with ow and no ow suppression and (c) with water
ow and ow suppression on. The relative positions of lumen, plastic tube, and
silicone are identied for this phantom in a. Notice the elimination of ow signal
and ow artifacts, and the preserved silicone signal in c. The remaining central
signal was from uid trapped within the coil.
with manganese-chloride (T1 of 640 ms, T2 of 90 ms) was fed through the tube by a
constant rate ow pump (Masterex model 7520-25; Cole-Parmer Instrument Company, Chicago, IL). Real-time cross sectional images were acquired using a exible
twin-lead coil inserted within a 9-french catheter inside the phantom. Real-time images acquired using a 2.4-cm FOV and 6-interleave spiral acquisitions are contained
in Figure 4.4. The coil sensitivity pattern can be inferred from the image with no
ow (Fig. 4.4a) which shows the relative positions of the lumen, plastic tube, and
silicone outer layer. In the presence of ow, white-blood images contained significant artifacts (Fig. 4.4b). The strength of ow artifact signal is due both to the
undisturbed Mz of owing spins and the sensitivity of the coil being the highest near
the coil center. Real-time ow suppression using spatial presaturation (Fig. 4.4c)
resulted in the near elimination of ow signal and images containing only the static
objects of interest, in this ow phantom. The only remaining central signal was a
small amount of uid trapped within the coil.
Two rabbit experiments were then performed to evaluate the quality of blood
suppression in vivo. In each experiment a exible twin-lead intravascular coil within
50
Chapter 4. Real-Time Black Blood MRI
a
b
c
d
e
Figure 4.5: Intravascular coil rabbit aorta experiment. Vessel cross section images
taken with an intravascular coil in a rabbit's descending aorta. 6-interleave images
are shown (a) without ow suppression, and (b) with ow suppression. 20-interleave
images are shown (c) without ow suppression, (d) with ow suppression, and (e)
lipid image with ow suppression. The lipid image was achieved by setting the scan
center frequency to lipids.
a 6-french catheter was steered to the descending aorta of a sedated rabbit. Realtime vessel cross-sectional images acquired using a 2.4-cm FOV and 6-interleave
spirals and using a 3.0-cm FOV and 20-interleave spirals are shown in Figure 4.5.
Conventional real-time images show signicant artifacts due to pulsatile ow and
respiratory motion (Fig. 4.5a,c), while ow suppressed images of water and fat
(Fig. 4.5b,d,e) show signicantly improved wall depiction and the elimination of
disruptive ow artifacts.
Chapter 4. Real-Time Black Blood MRI
51
Note that for this application, where the region of coil sensitivity only covers one
direction of through-plane ow, the sequence would be more time eÆcient with just
one saturation slab, placed upstream from the imaging slice.
4.3
Discussion
Spatial presaturation is a practical method for acquiring black-blood images in real
time. By devoting a portion of each TR to saturating out-of-slice spins, ow is suppressed and black-blood contrast is achieved in real-time while maintaining steadystate conditions for static objects. Initial studies have indicated that this technique
may supplement real-time cardiac wall motion study and improve non-occluded intravascular imaging.
In this technique, the quality of ow suppression is directly related to the
through-plane velocity component of owing spins and to the saturation prole of
suppression pulses. To be robust in the presence of slow through-plane ow, sharply
proled suppression pulses can be used at the cost of scan time (increased TR).
Alternatively, if through-plane velocities are known to be large, similar black-blood
contrast can be achieved with less precise saturation pulses and minimal increases
in scan time.
Technical limitations of this sequence include: 1) slow ow, which may not be
fully suppressed using spatial presaturation, 2) noise, which fundamentally limits
real-time studies, and 3) frame rates, which are reduced because the presaturation
requires additional scan time. Issues of slow ow may be addressed with improved
saturation pulse design. Noise also may be addressed with the development of newer
more targeted receiver coils or coil arrays.
Real-time white-blood MRI has become a popular tool for the evaluation of LV
function, while often suering from reduced contrast in certain segments due to
Chapter 4. Real-Time Black Blood MRI
52
ow and through-plane myocardial motion. Recently developed refocused SSFP sequences have shown improved endocardial border denition, however rapid throughplane motion continues to degrade image contrast due to the lack of a steady state.
We have demonstrated an interactive scanning system, where the ability to switch
to a black-blood mode for the evaluation of these segments is benecial. In addition,
real-time black-blood MRI enables intravascular imaging in the presence of regular
ow and motion, by suppressing ow and its related artifacts. With improved resolution and suppression pulse design, applications such as real-time valve imaging [74]
may also be possible.
Chapter 5
Limits of Resolution
When designing MRI pulse sequences, one must negotiate temporal resolution, spatial resolution, and signal-to-noise ratio (SNR) based on the desired applications.
As introduced in Section 2.3.3, the development of novel k-space sampling schemes
has enabled fast, high resolution imaging at the cost of SNR and some image artifacts. Each acquisition scheme is capable of achieving a range of temporal and
spatial resolutions. While SNR is mainly dependent on imaging time, it can also be
improved by other methods such as using contrast agents or smaller localized RF
coils.
In this chapter, we will characterize the temporal and spatial resolution limits
currently experienced in MRI, and illustrate design decisions and results for two
cardiac applications. A rst application is real-time multi-slice stress imaging of left
ventricular (LV) function and perfusion. This requires very high temporal resolution
(about 48 images/s) and limited spatial resolution (about 3 3 mm2). A second
application is real-time coronary imaging which requires sub-millimeter resolution
but acquisition times can be as long as the shortest stable diastolic window (<
150 ms).
For both applications, the sequence motivation is followed by discussion of the
temporal-spatial resolution tradeo and early in vivo results.
53
Chapter 5. Limits of Resolution
5.1
54
Real-time Multi-slice Stress Imaging
The detection of myocardial perfusion decits and contractile dysfunction are the
most frequent clinical assessments made in ischemic heart disease. Annually, more
than 3 million stress nuclear and echocardiography studies are performed in the
United States alone [75]. Magnetic resonance determination of stress induced wall
motion changes and perfusion have already been demonstrated [76{78]. However,
there are signicant limitations related to total scan time, scan plane localization,
arrhythmia, and image registration. Recent studies suggest that, for wall motion
assessment, imaging time may be as long as 1 hour using current techniques [76] with
good quality images obtained in only 80% of patients [77]. Furthermore, up to 35%
of patients develop signicant ventricular or atrial extra-systoles (> 5/min) during
pharmacologic stress [79]. This clinically tolerable rate of arrhythmia signicantly
complicates sequences that require gating.
Myocardial perfusion imaging has been investigated using several MR techniques.
Early studies used T1 or T2 agents and gradient recalled echo sequences to demonstrate the ability to detect myocardial perfusion defects in humans [80{82]. Recently, using echo-planar imaging (EPI) acquisitions, multiphase, multi-level perfusion imaging has also been reported [83]. Initial studies providing quantitation of
myocardial perfusion using the ventricular input function and a temporal resolution of 160{235 ms have been performed [82,84]. However, these clinical perfusion
studies are not done under true stress conditions and may be infeasible given the
prolonged image acquisition and contrast bolus timing.
Current perfusion exam protocols rely on vasodilation rather than true ischemic
stress. While these established techniques can acquire images in about 100 ms/slice,
they are suitable only at rest. Under the conditions of true stress and induced
ischemia, where the diastolic window can be less than 180 ms, we believe faster
volumetric real-time techniques are necessary.
Chapter 5. Limits of Resolution
55
Real-time wall-motion imaging using single-slice techniques have already shown
that rapid acquisitions can provide accurate wall-motion information and quantitative measures of function during free breathing [37,71]. Using slice interleaving,
real-time images of up to two slices have also been previously demonstrated with
limited interactive control [85,86].
An ideal cardiac wall motion and perfusion study would combine the high spatial
resolution, image quality, and variable scan plane features of MR imaging with the
real-time temporal resolution and control of echocardiography. Multi-slice coverage is an extremely important for the thorough evaluation of myocardial ischemia.
The current standard requires imaging 16 myocardial segments as dened by the
American Society of Echocardiography (ASE) [87,88]. This must be done quickly
and accurately during stress and must correspond to the segments imaged at rest.
Current real-time imaging systems are able to evaluate left ventricular (LV) function
and wall motion in single slices, but multi-slice coverage and imaging under stress
would be useful additions to such a system.
In summary, the limitations of current methods include: 1) insuÆcient temporal
resolution preventing volumetric perfusion studies under conditions of true induced
ischemia (tachycardia), 2) dependence on cardiac gating, which excludes patients
with arrhythmia (at rest or induced by stress), 3) multiple breath holds, which
requires patient cooperation and cannot be expected of patients experiencing true
ischemia, 4) lengthy examinations due to the need for multiple scout images and
static planning scans, and 5) the lack of simultaneous volumetric coverage, requiring
registration of multiple slices during stress function studies.
To address these issues, we have developed a real-time interactive (RTI) multislice MR imaging system that is capable of volumetric LV evaluation under resting
and stress conditions. Three parallel slices are continuously imaged in an interleaved
fashion. By choosing these slices to contain apical, mid and base short-axis views,
all 16 myocardial segments (as dened by the American Society for Echocardiography) [87, 88] can be visualized simultaneously. Initial studies using this system
indicate that in healthy subjects resting and stress LV function can be evaluated in
Chapter 5. Limits of Resolution
56
examinations requiring about 5 minutes each (including pre-scan calibration); and
when used with contrast agents, can provide perfusion information in all slices.
5.1.1
Method
Our real time interactive (RTI) multi-slice sequence was implemented within the
Stanford RTI system [29] described in Section 2.5, which provides the framework
for real-time reconstruction and interactive adjustment of imaging parameters. In
this system, the operator is provided with interactive control over the scan plane,
and parameters including eld of view, slice thickness, and ip angle.
The pulse sequence consists of a slice-selective excitation followed by a readout
and a gradient spoiler in the slice-select direction. We did not use spectral-spatial
excitations because the target application requires short TRs, and therefore short
excitations. The slice-selective excitations were designed using the Shinnar-LeRoux
algorithm [19,46]. Initial studies used a 2-ms excitation with a sharp prole and
minimum slice thickness of 3 mm. For studies requiring shorter TR, we used a
640-s excitation with minimum slice thickness of 5 mm.
Readouts used the partial k-space circular EPI (CEPI) trajectory [89] shown in
Figure 5.1a, and were designed to achieve 3.12-mm resolution over a 20-cm eld of
view (FOV). Circular EPI is a variation of the EPI trajectory that has a circular
k-space footprint and therefore a circular image FOV. Utilizing a partial k-space
approach [27,28] only 34 of 64 k-space lines are acquired for each 64 64 image
(53% coverage). Each slice is imaged using one or two excitations. In the singleshot case, the readout duration is 14.8 ms, while the two-shot case has a readout
duration of 7.6 ms. Frame rates of 48-50 complete images/s could thus be acquired.
With the same hardware and acquisition method, one could also tradeo spatial
resolution for temporal resolution to achieve 3.64-mm resolution at 72 images/s or
5-mm resolution at 100 images/s.
57
Chapter 5. Limits of Resolution
21 ms
Slice 1
b) single shot imaging
Slice 2
Slice 3
Slice 1
Slice 2
...
Display Set
c) two shot imaging
a) partial k-space
CEPI trajectory
Slice 1 Slice 2 Slice 3 Slice 1 Slice 2 Slice 3 Slice 1 Slice 2 Slice 3 Slice 1
Intl 1 Intl 1 Intl 1 Intl 2 Intl 2 Intl 2 Intl 1 Intl 1 Intl 1 Intl 2
...
Display Set
Figure 5.1: Acquisition and display timing. Three slices are imaged using (a) partial
k-space circular echo planar trajectories. Depending on the application, images are
acquired using a single-shot, or two-shots. Acquisition ordering is shown for (b)
single-shot, and (c) two-shot imaging with two interleaved acquisitions. All three
slices are imaged continuously and are displayed together.
Specic parameters from each of our studies are summarized in Table 5.1. We
initially used single-shot CEPI in a resting wall motion study, but found that twoshot CEPI resulted in reduced ow and o-resonance image artifacts. During stress
and perfusion studies, we used two-shot CEPI and a shorter excitation pulse to
maintain the same frame rate. Using the shorter excitation, all but 2 ms of each TR
was occupied by readout. Under these conditions, the resolution versus image time
tradeo is shown in Figure 5.2 for dierent numbers of interleaves. In general, dividing the acquisition into more interleaves increases the imaging time, but shortens
readout durations, resulting in reduced ow and o-resonance artifacts [47,90].
The ordering of acquisitions is summarized in Figure 5.1b-c. All three slices
are imaged continuously and are reconstructed and displayed asynchronously at
the highest rate possible. In the two-shot case, each CEPI readout is acquired for
all slices before moving to the next readout. This maintains a constant TR for
each slice, and therefore consistent steady state signal levels in each slice over all
58
Chapter 5. Limits of Resolution
Field of View (FOV)
Resolution
Slice Thickness
Flip Angle
Number of Interleaves
Excitation Duration
Readout Duration
Repetition Time (TR)
Frame Rate (three slices)
Rest Study Stress Study Perfusion Study
20 cm
3.12 mm
3 mm
30o
1
2.048 ms
14.8 ms
21 ms
15.87 fps
20 cm
3.12 mm
5 mm
30o
2
640 s
7.6 ms
10 ms
16.7 fps
20 cm
3.12 mm
5 mm
80o
2
640 s
7.6 ms
10 ms
16.7 fps
Table 5.1: Scan parameters for RTI multi-slice studies
Spatial Resolution (mm)
1
2
3
4
5
# Interleaves
5 mm
4 mm
3 mm
2 mm
10 ms
20 ms
30 ms
40 ms
Time per Frame (ms)
Figure 5.2: Resolution tradeo for high frame rate real-time imaging using the
partial k-space CEPI trajectory. The resolution curves for one- to ve-interleaf
CEPI acquisitions are shown based on a 20-cm FOV and 2 ms per TR for the
excitation and gradient spoiler.
Chapter 5. Limits of Resolution
59
excitations. Note that while imaging with three slices provides adequate coverage
for the left ventricle, the method applies to any number of slices.
Real-time reconstruction and display is managed by an external workstation [29].
For each display set, the three slices are reconstructed and displayed for immediate
feedback. Image triplets are reconstructed and displayed at the rate of 16 frames/s
with a display latency of less than 1 s. If desired, in the multi-shot case, higher
frame rates could be achieved by using a sliding window reconstruction [31].
Our MR studies were conducted on a 1.5 T GE Signa CV/i scanner (General
Electric, Milwaukee, WI). The scanner was equipped with gradients supporting
40 mT/m magnitude and 150 mT/m/ms max slew rate and a receiver capable of
4 s sampling (125 kHz). A body coil was used for RF transmission, with a 5inch surface coil used for signal reception. The multi-slice sequence was designed to
image three slices using imaging parameters summarized in Table 5.1.
5.1.2
Results
Ventricular Function
Our initial multi-slice real-time system was validated in a study of healthy volunteers. Twenty healthy subjects were scanned to evaluate the visibility of LV wall
segments at rest and during free breathing. For each subject, apical, mid and base
short-axis views were localized, and 10 to 15 seconds of video was stored. Each exam
took no longer than 5 minutes (including pre-scan calibration, interactive localization, and video acquisition). Figure 5.3 contains a representative image sequence
acquired during one such study (with every fourth frame shown). In this subject,
seventeen frames of each slice were acquired within one cardiac cycle at a heart rate
of 54 beats per minute. Six of seventeen frames occurred during the systolic period.
LV wall segments are clearly visualized, while LV blood volume experiences some
signal voids due to the movement of saturated spins (from the other slices). In all
twenty subjects studied, all sixteen wall segments were adequately visualized. Five
Chapter 5. Limits of Resolution
60
of the studies showed reduced signal in the posterior lateral wall segments due to coil
sensitivity fallo, and three studies showed some susceptibility artifacts. However,
neither artifact prohibited the evaluation of wall motion.
Due to the high temporal resolution of this sequence, it may also be used to
assess wall motion during stress. Figure 5.4 contains systolic and diastolic images of
a normal subject studied at rest and under physiological stress. In this study, physiological stress was induced by having the subject run for 10 minutes immediately
before scanning. This subject had a resting heart rate of 86 beats per minute, and
a stress heart rate of 176 beats per minute. Even under highly stressed conditions,
wall motion is well resolved in all three slices.
Myocardial Perfusion
When used in conjunction with contrast agents, multi-slice uoroscopy provides
perfusion information as well as wall motion information. The course of a bolus
contrast can be tracked in real-time with eectively volumetric LV coverage. Figure 5.5 illustrates the blood and myocardium signal enhancement observed after a
bolus injection of Gadolinium-DTPA (10 ml Magnevist, Berlex Laboratories Inc.,
Wayne, NJ). Figure 5.6 illustrates the signal levels in the right ventricle, left ventricle, and myocardium for 45 seconds following the introduction of contrast. In
this healthy subject at rest (heart rate of 83 beats per minute), enhancement is
seen in the right ventricle, left ventricle, and nally in the myocardium. Due to the
volumetric coverage and real-time nature of this sequence, only a single contrast
injection (without the need for careful bolus timing) is needed to study perfusion
throughout the left ventricular wall. Note that a nulling pre-pulse was not used in
this acquisition, but can be included at a small cost in imaging time.
Left ventricular contractile function and blood ow information is simultaneously
obtained. Furthermore, the arterial input function is sampled evenly throughout
the cardiac cycle compared to current multiple slice acquisitions [84]. This avoids
61
Chapter 5. Limits of Resolution
apex
middle
base
time
Figure 5.3: Multi-slice image sequence from a normal volunteer. Real-time imaging
of three slices allows visualization of the entire LV during systole and diastole. In
this subject, seventeen frames of each slice were acquired within one cardiac cycle at
a heart rate of 54 beats per minute (every fourth frame is shown). Six of seventeen
frames occurred during the systolic period.
62
Chapter 5. Limits of Resolution
86 bpm
b) stress
176 bpm
systole
diastole
a) rest
Figure 5.4: Resting versus stress. Three-level images from a normal volunteer
scanned under (a) resting conditions (heart rate of 86 beats per minute), and under
(b) physiological stress (heart rate of 176 beats per minute). Physiological stress in
the subject was induced by running for 10 minutes immediately before the scan.
the need for further image registration prior to interpretation and should improve
quantitative analysis.
5.1.3
Summary of Multi-slice Stress Imaging
Real-time interactive multi-slice MRI is a practical tool for evaluating LV function.
It can provide volumetric coverage of 16 wall segments in real-time and at frame
rates suÆcient for combined stress wall motion and perfusion study. The interactive
nature enables rapid localization and real-time scan plane adjustment, while rapid
acquisitions eliminate the need for gating or breath holding. Our initial studies
indicate that a LV wall motion study can be completed in roughly ve minutes,
while stress and perfusion studies are also possible with similar scan times.
Changes in cardiac position during stress studies are common due to changes
in hemodynamics and patient motion especially when patient discomfort is present
when true ischemia is induced. An extremely important component of stress imaging
is the ability for the operator to compensate for these changes. Current MR stress
63
Chapter 5. Limits of Resolution
a
0s
b
7s
c
19 s
d
27 s
e
64 s
Figure 5.5: First pass perfusion images. Time series of images from a normal subject
after bolus injection of Gadolinium-DTPA. Images show signal in three slices (a)
before injection, (b) with RV signal elevation, (c) with LV signal elevation, (d) with
myocardium signal elevation, and nally (e) wash out. The subject's heart rate was
83 beats per minute during this study.
64
Chapter 5. Limits of Resolution
RV
Signal Level
LV
Myocardium
0
15
30
45
Time (seconds)
Figure 5.6: First pass perfusion signal time course. The graph shows right ventricle (RV), left ventricle (LV), and myocardium signal levels for the rst 45 seconds
after a bolus Gadolinium-DTPA injection. The plotted quantities are signal intensities measured during late diastole, and were computed by calculating mean
signal in manually selected regions of interest in the RV, LV, and anterior LV wall
(myocardium).
studies do not address this issue of registration since the current process of slice
prescription is both time-consuming and not amenable to rapid adjustment. The
same limitation applies to perfusion imaging.
Although no denitive data exist on the optimal frame rate needed for diagnostic wall motion studies, there are suggestions that frame rates beyond 24 frames/s
may not be required. At peak stress, systolic ejection is typically completed only
150{180 ms after the onset of ventricular excitation. The broad clinical experience
in assessing ventricular function under stress using echocardiography indicates the
need for 4 to 6 frames over the systolic period [91]. This translates into a temporal
resolution of about 30{40 ms. The necessary high frame rate of 24 frames/s and
available imaging time at peak stress of 60{90 seconds together pose stringent constraints on real-time stress wall motion imaging sequences. Furthermore, prolonged
Chapter 5. Limits of Resolution
65
acquisitions for perfusion [82] during diastole are not feasible during true pharmacologic stress when the entire cardiac cycle could be as short as 350 ms. Many of the
available tradeos depend on the hardware performance, software reconstruction
techniques, and clinical constraints. We have demonstrated simultaneous real-time
images of 3 slice locations at 16 complete frames/slice/s.
Faster gradients will also improve both the readout and excitation performance of
the multi-slice perfusion sequence. Readout duration can be reduced by increasing
gradient area at the cost of some SNR. Increased bandwidth would also improve
multi-slice implementations by allowing more precise slice proles to minimize the
eect of partial saturation from slice movement during the cardiac cycle.
Recently, the development of multiple-coil based imaging sequences, such as
SMASH [92] and SENSE [85,93], have led to signicant advances in reducing the data
set needed for image reconstruction. These methods are particularly suitable for
large FOV studies where greater temporal resolution is desired. In our experience,
however, the 16-24 cm FOV necessary for cardiac studies can be adequately covered
by a single coil or two-coil array; and currently available gradients appear capable of
achieving reasonable temporal resolution. However, with improved reconstruction
speed, SENSE and SMASH techniques may be useful for further improvements in
temporal resolution with large FOV multi-slice imaging.
5.2
Real-time Coronary Imaging
The rst successful coronary imaging trials using MRI operated at between 1.1 and
1.5-mm spatial resolution and were acquired over acquisition windows on the order of
150 to 250 ms per cardiac cycle. At these rates, coronary evaluation with sensitivities
and specicities on the order of 63-90% were reported [94{96]. With hardware and
pulse sequence improvements that have occurred over the past ten years, scanners
are now capable of acquiring superior spatial and temporal resolution in real-time.
In this section, we discuss the resolution capabilities of common fast imaging k-space
Chapter 5. Limits of Resolution
66
trajectories on a conventional scanner, and present a real-time interactive system
for rapid coronary screening.
Real-time interactive (RTI) imaging is a popular and robust cardiac imaging
technique because it does not require gating or breath holding, enables the observation of dynamic motion and enables the rapid localization of scan planes containing
desired cardiac views. However, it is a challenge to achieve high spatial and temporal
resolution simultaneously in real time while maintaining suÆcient SNR. Real-time
coronary artery imaging is particularly diÆcult due to the small vessel size and the
signicant motion that they experience. One advantage of real-time acquisition is
the availability of multiple images of the same slice. This can be used to selectively
average images for improved SNR, as demonstrated by Hardy et al. [97,98]. Alternatively, multiple images, acquired with asymmetric resolution and high resolution
in dierent directions, can be combined in some intelligent way [11]. A conventional
use for real-time imaging is as a scan plane localizer for high resolution 2D coronary
imaging sequences [23].
We present a similar high-resolution RTI imaging approach that achieves submillimeter resolution in short acquisition windows. Preliminary results in normal
volunteers indicate this is useful as a high quality 2D coronary localizer, and may
be used to screen the coronary artery tree.
5.2.1
Methods
Experiments were conducted on a GE Signa 1.5 T CV/i scanner (General Electric, Milwaukee, WI) equipped with gradients supporting 40 mT/m magnitude and
150 mT/m/ms slew rate and receiver capable of 4 s sampling ( 125kHz). A body
coil was used for RF transmission, and a 5 inch surface coil for signal reception.
This pulse sequence was implemented within the Stanford real-time imaging
(RTI) environment described in Section 2.5, which provided a framework for interactive control of scan plane and imaging parameters, and for continuous real-time
reconstruction and display.
67
Chapter 5. Limits of Resolution
a
c
Spiral
b
pkCEPI
2:1 pkCEPI
Resolution
Elements
Figure 5.7: K-space trajectories for high-resolution real-time imaging. (a) interleaved spirals, (b) partial k-space CEPI and (c) 2:1 asymmetric partial k-space
CEPI. Cropped images demonstrate the asymmetric resolution of c When the elliptical resolution elements of the asymmetric image are oriented parallel to comb
bristles, they appear well dened, however, when they are oriented perpendicular to
the bristles, the features are blurred away.
The pulse sequence consisted of a water-selective excitation followed by interleaved readouts and a gradient spoiler in the slice select direction. To meet the
demands of high spatial and temporal resolution, three readout trajectories shown
in Figure 5.7 were explored; specically interleaved spirals, partial k-space circular
EPI (pkCEPI), and asymmetric pkCEPI. Circular EPI is a variation of the EPI
trajectory [22] that has a circular k-space footprint and therefore a circular image
FOV. Partial k-space CEPI utilizes the conjugate symmetry of k-space and requires
the acquisition of slightly more than half of the k-space lines and uses a modied
reconstruction [27, 28]. Furthermore, asymmetric pkCEPI stretches the pkCEPI
trajectory to achieve higher resolution in the readout direction while sacricing resolution in the phase encode direction. We particularly explored using 2:1 k-space
Chapter 5. Limits of Resolution
68
asymmetry to achieve twice the resolution in one direction as in the other. However,
any arbitrary resolution ratio is achievable and is supported by this system.
Cropped images of a resolution phantom are included in Fig. 5.7c to demonstrate
the eect of asymmetric resolution. When the elliptical resolution elements are
oriented parallel to the comb bristles, they appear well dened, however, when they
are oriented perpendicular to the bristles, the features are blurred away.
In real-time coronary studies, the direction of high-resolution can be interactively
placed perpendicular to the vessel or vessel segment of interest. This enables the
quick scouting of even tortuous vessels, and takes advantage of the multiple image
capability of real-time MRI.
Figure 5.8 and Table 5.2 characterize the temporal and spatial resolution tradeo for each of these three trajectories based on our hardware conguration. All
comparisons involved optimal trajectory design [89, 99], and are based on gradients supporting 40 mT/m magnitude and 150 mT/m/ms slew rate, and receiver
supporting 4 s sampling (125kHz), 20-cm FOV, 16.384-ms readouts, 30-ms TR,
and roughly 55% coverage by partial k-space. These are all typical parameters for
single-coil real-time cardiac imaging. All three trajectories are capable of achieving
sub-millimeter resolution; however, spiral and pkCEPI trajectories require longer
imaging times, making them more susceptible to motion artifacts. Figure 5.9 contains images acquired using 2:1 pkCEPI and spirals of comparable voxel size. Both
images achieve high resolution for a real-time system, while exhibiting dierent
types of artifacts. The pkCEPI image is sharper due to its shorter acquisition and
relative insensitivity to o-resonance [100], but suers from ow ghosting in the
phase-encode direction. The spiral image exhibits minimally disruptive swirling artifacts from ow, but shows blurring due to o-resonance and the longer acquisition
window [47]. The spiral image also has higher SNR because it was acquired over a
longer acquisition window.
In our coronary studies, we elected to use the 2:1 asymmetric pkCEPI trajectory
with 5 interleaves achieving 0.8 1.6 mm2 resolution in 135 ms, and spiral trajectories with 7 interleaves achieving 1.13 1.13 mm2 resolution in 189 ms. These two
69
Chapter 5. Limits of Resolution
4 mm
Spatial Resolution
Spiral
pkCEPI
3 mm
2:1 pkCEPI
2 mm
1 mm
27 ms
54 ms
81 ms
108 ms
Imaging Time
135 ms
162 ms
189 ms
216 ms
27 ms TR
Figure 5.8: Temporal and spatial resolution tradeos for fast imaging k-space trajectories.
Imaging
Image Resolution (mm)
Time Spirals pkCEPI 2:1 pkCEPI
27 ms
54 ms
81 ms
108 ms
135 ms
152 ms
189 ms
216 ms
3.7
2.3
1.8
1.5
1.3
1.2
1.13
1.0
3.2
2.1
1.7
1.4
1.2
1.0
0.94
0.9
1.85 3.70
1.33 2.66
1.01 2.02
0.89 1.78
0.80 1.60
0.70 1.40
0.64 1.28
0.60 1.20
Table 5.2: Temporal and spatial resolution tradeos for fast imaging k-space trajectories. Acquisitions used in future studies are shown in bold.
70
Chapter 5. Limits of Resolution
2:1 pkCEPI
a
Spiral
b
0.8 x 1.6 mm, 135 ms
1.13 x 1.13 mm, 189 ms
Figure 5.9: Comparison of 2:1 pkCEPI and spiral images. The right coronary from
a normal volunteer is shown using (a) 2:1 pkCEPI with the ellipse indicating the
orientation of resolution elements and (b) interleaved spirals. The pkCEPI image is
sharper due to its shorter acquisition and insensitivity to o-resonance, but suers
from ow ghosting in the phase encode direction. The spiral image exhibits minimally disruptive swirling artifacts from ow, but shows blurring due to o-resonance
and the longer acquisition window. The spiral image also has higher SNR because
it was acquired over a longer acquisition window.
trajectories were specically chosen because they have roughly the same voxel size
(for comparison purposes). Both used a 20-cm FOV and 5-mm slice thickness. The
pulse sequence consisted of a 7-ms water-selective excitation followed by interleaved
readouts, and a gradient spoiler. For non-contrast studies, a ip angle of 30o was
used, and for contrast enhanced studies, a ip angle of 90o was used. Using a sliding
window reconstruction [31], we were able to achieve image display rates of up to 33
images/s.
Chapter 5. Limits of Resolution
5.2.2
71
Results
Our initial study consisted of nine normal volunteers scanned using the 2:1 pkCEPI
acquisition. Each volunteer was scanned for roughly twenty minutes with the goal
of visualizing segments of both right and left coronary trees. As a result, the right
coronary artery (RCA) was visualized in all nine, and some part of the left coronary
system was visualized in ve of the nine. We believe that greater diÆculty was
experienced viewing the left coronary system because of the smaller vessel sizes, and
lower SNR due to distance from the surface receiver coil. Figure 5.10 contains single
frames captured from real-time video acquired with this sequence. Figure 5.10(a
and b) depict the right coronary and origin of the left main in one volunteer, (c)
depicts a tortuous RCA in another volunteer, and (d) depicts another RCA in a third
volunteer. While images here show vessel segments at single instances, a longer vessel
length is observed in a real time video due to segments passing through the imaging
slice in dierent time frames. In addition, while viewing video, noise temporally
averages down.
One immediate application of this technique is as a high quality localizer for 2D
sequences. Figure 5.11 contains views of an RCA from the same volunteer during
the same scan, rst imaged with the real-time system, and then with a gated breath
held 2D spiral technique [23]. The gated image clearly has higher SNR and achieves
better resolution due to the longer integration time, however the real-time image
provides reasonable resolution and can be used to achieve very accurate localization.
As SNR is a major limiting factor in real-time acquisition, this system could
greatly benet from the use of T1 shortening contrast agents. In addition to improving SNR, blood-ow-induced artifacts will be reduced due to the more consistent
blood pool signal. This should signicantly improve both EPI and spiral based acquisitions. In a preliminary trial, two normal volunteers were scanned with both
acquisition schemes following bolus injections of Gadolinium-DTPA (10 cc Magnevist, Berlex Laboratories Inc., Wayne, NJ). Figure 5.9 contains right coronary
images acquired 3 minutes after contrast injection. Figure 5.12 contains images of
72
Chapter 5. Limits of Resolution
a
b
c
d
Figure 5.10: Real-time images of the (a) right coronary and (b) left coronary in one
volunteer, and the (c,d) right coronaries of two other volunteers, using 5-interleave
2:1 pkCEPI acquisitions. Each image is a single still frame taken from a real-time
video sequence. The high-resolution readout direction is the horizontal direction.
73
Chapter 5. Limits of Resolution
real time 2:1 pkCEPI
a
0.8 x 1.6 mm 2
gated spirals
b
135 ms
0.75 x 0.75 mm 2
14 beats
Figure 5.11: RCA images acquired in (a) real-time and (b) using a gated spiral
technique on the same healthy volunteer. The SNR improvement using gated techniques is signicant, however the resolution achieved in real-time is suÆcient for
visualization.
Chapter 5. Limits of Resolution
74
the left anterior descending (LAD) and some smaller left diagonal branches (indicated by arrows), which became visible with contrast. In this preliminary study,
smaller vessels such as the diagonal branches became visible with the SNR enhancement provided by contrast agents.
5.2.3
Summary of Coronary Imaging
We have demonstrated sub-millimeter resolution real-time interactive 2D coronary
imaging with currently available scanner hardware. Novel k-space trajectories such
as asymmetric pkCEPI can be used in conjunction with a RTI system to provide
better resolution in one preferred direction. In vivo studies show that real-time
images can achieve sub-millimeter resolution, but have low SNR compared to gated
images. However, preliminary studies suggest that contrast agents may signicantly
improve SNR. This sequence can be immediately used for accurate 2D coronary
localization, and may be useful for the rapid initial screening of coronary lesions
and for the guidance of high resolution scans.
75
Chapter 5. Limits of Resolution
2:1 pkCEPI
a
0.8 x 1.6 mm 2
Spirals
b
135 ms
1.13 x 1.13 mm 2
189 ms
Figure 5.12: Real-time images of the left coronary tree 3 minutes after bolus injection
of Gadolinium. As shown by white arrows, both (a) pkCEPI and (b) spiral sequences
begin to show small vessels such as the left diagonal branches.
Chapter 6
EÆcient O-resonance Correction
Radial sampling schemes such as projection reconstruction (PR) and spirals are
popular MR imaging techniques because of their ability to achieve short echo times
and to produce images with minimal ow artifacts. These techniques are often used
in the imaging of short T2 species [101] and in rapid (or real-time) imaging [29,33].
Inherent in radial acquisitions is an oversampling around the k-space origin. This
is considered to be a strong advantage because it is well known that the low spatial
frequencies contain most of the energy in images. This oversampling can however be
exploited to improve imaging in other ways. For example, image eld of view (FOV)
is often calculated based on an inner area of k-space, tolerating spoke artifacts from
insuÆciently sampling the outer regions. Two schemes that use this approach are
undersampled PR [102{105] and variable density spirals [106].
One major drawback of radial sampling is that o-resonant spins cause image
blurring. The most common way to compensate for this blurring involves acquiring
a eld map with an additional scan and using a frequency-sensitive reconstruction
to compensate for o-resonance [107].
We present a technique that corrects for o-resonance without a separate eld
map acquisition. This technique incorporates a eld map acquisition within imaging
excitations by oversampling a region around the k-space origin and interleaving
76
Chapter 6. EÆcient O-resonance Correction
77
acquisitions with two dierent echo times (TEs). The acquired data is then used
both to generate a low-resolution eld map, and to reconstruct the nal image
with o-resonance correction. In projection reconstruction imaging, the acquired kspace is unchanged, while in spiral imaging, a modied (variable density) trajectory
is required. In this chapter, a theoretical introduction and general methodology is
followed by details of PR and spiral implementations. Results from phantom studies
and in vivo studies are presented.
6.1
Theory
In MR imaging, the Fourier relationship between image domain and k-space yields
the following signal equation.
S (t) =
Z
x;y
M (x; y )e i2(kx (t)x+ky (t)y) dxdy;
(6.1)
where M (x; y) is the imaging slice and kx(t) and ky (t) are Rthe k-space trajectories
t
determined
by
the
linear
gradients
elds;
i.e.,
kx(t) = 2 0 Gx ( )d and ky (t) =
Rt
2 0 Gy ( )d . In the presence of eld inhomogeneity or o-resonance the signal
equation translates into the relationship shown in Eq. 2.10:
S (t) =
Z
x;y
M (x; y )e i2f (x;y)t e i2(kx (t)x+ky (t)y) dxdy;
(6.2)
where f (x; y) is the shift in the precession frequency at point (x; y) in object space.
If all of k-space were acquired at the same instant (at the same time t), o-resonance
would not aect a reconstructed magnitude image. However, acquisition windows
have durations on the order of 2 to 16 ms, which introduces o-resonance artifacts.
Note that techniques that use longer readouts experience increased artifact.
One way to compensate for o-resonance is to acquire a eld map to estimate
f (x; y) and to use this information to better reconstruct the nal image. The eld
78
Chapter 6. EÆcient O-resonance Correction
map or o-resonance map fm (x; y), can be computed by acquiring two images at
dierent echo times and comparing the phase images at each location. These complex images would represent M (x; y)e i2f (x;y)T E1 and M (x; y)e i2f (x;y)T E2 . The
o-resonance term f (x; y) is then simply found by computing the phase dierence
between these two images divided by the dierence in echo times (TE2 TE1).
6.2
General Methods
In the proposed technique, radial trajectories are designed such that a small circular
region around the k-space origin is oversampled by a factor of two. Every alternate
acquisition is then delayed by a small amount labeled TE. The 2 oversampled
region is used to generate two lower resolution images with dierent echo times|one
from the early-TE interleaves and one from the late-TE interleaves. Because of the
oversampling in the central k-space region, both low resolution images will have the
same eld of view (FOV) as an image from the full data set.
A eld map fm (x; y) is then computed in a standard way, using the phase
dierence between the two low-resolution images, Ilate T E and Iearly T E :
E Iearly
fm (x; y) = arg(Ilate2TTE
TE
)
(6.3)
The measurable range of this eld map (in Hz) is notably limited by the oresonance that will cause a phase shift of after TE:
1
jfm (x; y)j < 2TE
(6.4)
Since eld map estimates are reliable in areas of suÆcient signal, a smooth polynomial approximation fp(x; y) [108] is computed for use during reconstruction.
This polynomial eld map fp(x; y) is found using a weighted least-squares algorithm which minimizes:
Chapter 6. EÆcient O-resonance Correction
X
i;j
wij [fp (xi ; yj )
fm (xi; yj )]2;
79
(6.5)
where wij measures the importance of minimizing the error at that voxel. The
weighting used is proportional to the square magnitude of the voxel signal. Such
a weighting may overvalue voxels that are close to the receiver coil therefore having
higher signal intensity, however this is accepted because the accuracy of eld map
estimates is directly related to this signal intensity. Another reasonable weighting
strategy is to equally weight all pixels with signal intensity above some threshold,
while excluding all pixels below that threshold. In the nal reconstruction step, we
use the smooth polynomial tted eld map.
Multifrequency image reconstruction is then used to simultaneously compensate
for o-resonance and correct for the dierence in echo times [107,109]. In multifrequency reconstruction, a small nite set of frequency samples ff ng are selected
such that they span the full range of o-resonant frequencies. For each f n, an
image is reconstructed based on that frequency of precession. This is done by modulating the raw data of each readout by e+i2f n t and phase-aligning the early and
late echo data by multiplying the raw data from late-TE acquisitions by e+i2f n T E .
An image is then generated via gridding reconstruction [25,26] with the modulated
and aligned data from both early-TE and late-TE acquisitions. For the nal image,
each voxel is estimated by interpolating between the images based on the closest
ff ng frequency samples to f (x; y) [107] or by taking a linear combination of all
the ff ng images [109].
Resulting images are formed without separate eld map acquisitions and with
only a moderate increase in reconstruction time. This combination of image and eld
map acquisitions provides for greater scan eÆciency, while the ability to specify the
oversampled region provides exibility in trading o eld map resolution for image
resolution.
Results were collected on a GE Signa 1.5T CV/i scanner (General Electric, Milwaukee, WI) equipped with gradients capable of 40 mT/m magnitude and 150 T/m/s
Chapter 6. EÆcient O-resonance Correction
80
slew rate and a receiver capable of 4 s sampling ( 125 kHz). For phantom studies, a head coil was used. For peripheral angiography studies, an extremity coil was
used. And for coronary imaging studies, a body coil was used for RF transmission
and 5-inch surface coil used for signal reception. Unless specied, all ORC-PR and
ORC-VDS images used a T E of 1 ms.
6.3
Application to Projection Reconstruction
Imaging
In projection reconstruction (PR) imaging, the central k-space data is inherently
oversampled, thus yielding the simplest application of this interleaved echo technique. This section describes the application of the automatic o-resonance correction technique to PR, termed ORC-PR.
6.3.1
Method
Figure 6.1 illustrates the ORC-PR pulse sequence. Flow-compensated excitations
are followed by radial spoke readouts and dephasing pulses in the slice-select direction. Every alternate spoke acquisition is delayed by a small amount labeled TE.
Notice that this pulse sequence is identical to conventional PR except for the echo
delay between early-TE and late-TE acquisitions.
As described in Section 6.2, early-TE and late-TE spokes are acquired at dierent
echo times. A low-resolution eld map is computed using the oversampled central
region of k-space. Finally, multifrequency reconstruction is used to correct for oresonance. This o-resonance correction is implemented with no additional scan
time and a modest increase in reconstruction time.
81
Chapter 6. EÆcient O-resonance Correction
RF
Gz
early-TE spokes
Gxy
DAQ
∆TE
late-TE spokes
Gxy
DAQ
Figure 6.1: O-resonance corrected ORC-PR pulse sequence: A small-tip excitation
(ow-compensated excitation is shown) is followed by a refocused readout which
acquires one radial spoke, and a gradient spoiler to dephase remaining signal. LateTE acquisitions have a delayed readout and echo, as shown.
6.3.2
Results
PR acquisitions consisting of 512 radial spokes were used to achieve an image resolution of 163 163 pixels (0.98-mm resolution over a 16-cm circular FOV, or 0.74-mm
resolution over a 12-cm circular FOV). Unless specied, each spoke readout lasted
8.192 ms, providing 256 samples. ORC-PR images used a 1-ms TE, 12th order
polynomial t (including cross terms) for the eld map, and 10 frequency samples
during multifrequency reconstruction. A 1-ms TE yields an o-resonance range of
500 Hz.
Results of this technique applied to a resolution phantom are shown in Figure 6.3.
These images were acquired using a head coil, 16-cm FOV, 3-mm slice thickness, 300ms TR, 4.5 ms TE, and 90o ip angle. Figure 6.3a is the result of a conventional PR
scan (without TE or o-resonance correction). Figures 6.3b-d show corresponding
segments of the estimated eld map, polynomial t eld map, and the corrected
image using our new scan and reconstruction technique. The measured o-resonance
was entirely within the range of [-38.3 to 19.4] Hz. The corrected image shows
82
Chapter 6. EÆcient O-resonance Correction
all acquisitions
early-TE (low res)
late-TE (low res)
Figure 6.2: K-space coverage. Of the full coverage (left), solid lines represent earlyTE acquisitions and dotted lines represent late-TE acquisitions. Early-TE and lateTE low resolution images (middle and right) are each computed from half of the
data in the central portion of k-space. These images have the same FOV as the nal
image, and have one-quarter the resolution{one-half in both x and y.
improved denition in the comb structure which is visible in cropped images (see
Figures 6.3e-f). This improvement is greater for sequences with long readouts, but
is apparent even when imaging with shorter readouts.
To demonstrate this method in vivo, we applied it to peripheral angiography
using 2D time of ight (TOF). Images were acquired using an extremity coil, 12 cm
FOV, 3-mm slice thickness, 33-ms TR, 4.5-ms TE, and 60o ip angle. In this application, the presence of lipids and their chemical-shift causes sharp discontinuities
in the acquired eld map. To correct for smooth o-resonance variations (mainly
B0 inhomogeneity), lipid voxels are excluded from the polynomial tting. These
lipid voxels are excluded by masking voxels with a measured fm of at least 110
Hz below the center frequency of water. Figure 6.4 illustrates a conventional PR
slice and the equivalent slice using our new acquisition and reconstruction scheme.
When reconstructing these data sets, the eld map estimate can be customized to
the application. When using a polynomial tted eld map, with fat excluded from
the tting, slice images show improved vessel denition (see Figures 6.4b-c). When
using a hybrid eld map consisting of the measured o-resonance in areas of high
signal, and the smooth estimate elsewhere, resulting slice images show improved
Chapter 6. EÆcient O-resonance Correction
a
b
c
d
83
e
f
Figure 6.3: Resolution phantom images using 8.192-ms readouts. Imaged with (a)
conventional PR, and with ORC-PR resulting in images of the (b) acquired eld
map, (c) smooth polynomial t eld map, and (d) nal reconstruction. The range
of the eld map images is -38.3 to 19.4 Hz. Close-ups of the comb structure in a
and d are shown in e and f.
Chapter 6. EÆcient O-resonance Correction
84
denition in vessels and areas of fat, while introducing some edge enhancement due
to eld map discontinuity (see Figures 6.4d-e).
Figure 6.5 illustrates targeted maximum intensity projections (MIPs) of peripheral vasculature imaged with and without this technique. These images were acquired using an echo time of 6.3 ms, which places water and fat out-of-phase, and
using a saturation pulse to suppress signal from veins. The MIPs constructed with
o-resonance correction show improved vessel denition throughout the popliteal
trifurcation and decreased blurring.
6.4
Application to Spiral Imaging
Spiral imaging is also used for a variety of applications due to its eÆcient k-space
coverage [23] and excellent ow properties [47]. In spiral imaging, o-resonance also
can result in image domain blurring [110]. Conventional approaches to spiral oresonance correction involve acquiring a eld map using extra acquisitions. Typically
two low resolution single-shot spiral images, taken at dierent echo times, are used
to compute a eld map. This map may then be used for linear eld map correction
or a frequency sensitive reconstruction [107,109]. Disadvantages of this approach
are that separate eld map acquisitions are required, magnitude information from
the eld map images are unused, and the eld map resolution is not very exible.
In this section, we apply the interleaved echo technique to spiral imaging, termed
ORC-VDS (o-resonance correction using variable density spirals) that combines the
image and eld map acquisitions using a variable density spiral (VDS) trajectory
that over-samples the k-space origin. This technique has greater scan time eÆciency
relative to conventional techniques, and enables the exible tradeo of eld map
resolution for image resolution.
Compared to the PR case, in spiral sampling, oversampling the k-space origin
requires trajectory modication, but the longer readouts make o-resonance blurring
a more signicant obstacle that requires accurate correction.
Chapter 6. EÆcient O-resonance Correction
85
a
b
c
d
e
Figure 6.4: In vivo example: An axial slice of the lower leg imaged with (a) conventional PR and (b-e) ORC-PR. For imaging vasculature, (b) fat is ignored during
eld map approximation fp and (c) the resulting corrected image shows improved
vessel denition. Alternatively, (d) the computed eld map fm can be used directly during reconstruction, resulting in (e) improved denition in areas of fat with
added edge enhancement due to eld map discontinuity.
Chapter 6. EÆcient O-resonance Correction
a
86
b
Figure 6.5: Targeted maximum intensity projections of the popliteal trifurcation
using multi-slice 2D TOF and venous saturation. Imaged using (a) conventional PR
and (b) ORC-PR. Notice the improved vessel denition and de-blurring in areas of
o-resonance, as indicated by arrows. Both images are windowed identically.
87
Chapter 6. EÆcient O-resonance Correction
a
b
c
∆TE
interleaves acquired at
two echo times
Figure 6.6: ORC-VDS k-space trajectories: (a) VDS trajectories are designed to
oversample a circular region around the k-space origin, (b) interleaves are acquired
at two dierent echo times, and (c) two low resolution images formed from the oversampled region are used to compute a eld map. Note that the two low resolution
images have the same eld of view as the full VDS data set.
6.4.1
Method
In the proposed technique, variable density spiral (VDS) trajectories (see Figure 6.6)
are designed such that a small circular region around the k-space origin is oversampled by a factor of two. As described in Section 6.2, every alternate spiral acquisition
is delayed by a small amount labeled TE. The 2 oversampled region is used to
generate two lower resolution images with dierent echo times{one from the earlyTE interleaves and one from the late-TE interleaves. These images are used to
compute a eld map which is used to correct for o-resonance in the complete data
set. Notice that the low resolution images have the same eld of view (FOV) as an
image from the full data set.
Final images are formed without separate eld map acquisitions and with only
a moderate increase in reconstruction time. This combination of image and eld
map acquisitions provides for greater scan eÆciency, while the ability to specify the
oversampled region provides exibility in trading o eld map resolution for image
resolution.
88
Chapter 6. EÆcient O-resonance Correction
320
6 interleave ORC-VDS
Image Resolution (pixels)
Image Resolution (pixels)
320
4 interleave spirals
with 2-shot field map
240
160
80
0
160
80
6
4
8
10
12
14
16
18
20
Number of
Excitations
0
0
a
240
40
80
120
Field Map Resolution (pixels)
160
0
b
40
80
120
160
Field Map Resolution (pixels)
Figure 6.7: Performance comparison of ORC-VDS and conventional techniques:
(a) Resolution comparison of 6 interleave ORC-VDS, and conventional 4 interleave
spirals with a 2 excitation eld map. (b) similar comparisons for dierent numbers
of excitations. Comparisons are based on a 20-cm FOV, 16.384-ms readouts with
4 s sampling, and gradients supporting 40 mT/m magnitude and 150 T/m/s slew
rate.
6.4.2
Results
Performance Evaluation
A rst comparison of ORC-VDS against the conventional (separate eld map) technique is illustrated in Figure 6.7. Optimal spiral and VDS trajectories [50,99,106]
were designed based on a 20-cm FOV, 16.384-ms readouts, and the full gradient and
receiver capability of our scanner.
In Figure 6.7a, ORC-VDS and the conventional technique are compared using
six excitations. The bullet represents the eld map resolution and image resolution
achievable using four interleaves for the image and two eld map acquisitions. The
curved line represents the full range of eld map and image resolutions that can
be achieved with ORC-VDS simply by adjusting the size of the oversampled kspace region. Figure 6.7b contains identical comparisons for dierent numbers of
excitations.
89
Chapter 6. EÆcient O-resonance Correction
a
b
c
Figure 6.8: Resolution phantom: Cropped images of a resolution phantom scanned
with (a) 22 interleave spirals and no inhomogeneity correction, (b) 22 interleave
spirals with a 2 excitation eld map and multifrequency correction, achieving 6.78mm eld map resolution, and (c) 24 interleave ORC-VDS achieving 5.2-mm eld
map resolution. The ORC-VDS image is sharper than the conventional corrected
spiral image acquired in the same scan time. White bars indicate eld map spatial
resolution in b and c.
In all cases, the ORC-VDS technique achieves higher image and eld map resolution because all acquired data is used for both the image and the eld map. In
addition, the proposed technique enables greater exibility in choosing eld map
resolution based on the expected o-resonance. For example, in the 6-excitation
case, ORC-VDS can achieve either 10% higher image resolution with the same eld
map resolution, or 20% higher eld map resolution with the same image resolution;
or, eld map resolution can be traded o for up to a 40% increase in image resolution
without additional scan time.
Experimental Results
To demonstrate this technique it was applied to the imaging of a resolution phantom
with high-order o-resonance, and to 2D coronary imaging with mostly linear oresonance. Both experiments were conducted in a well-shimmed magnet with oresonance within 35 Hz.
Figure 6.8 contains cropped images of a resolution phantom rst scanned with
conventional spirals (22 interleaves for the image and 2 excitations for the eld
90
Chapter 6. EÆcient O-resonance Correction
a
b
Figure 6.9: Coronary images: normal volunteer scanned with (a) conventional spirals
with 12 image interleaves, 2 excitation eld map, and multifrequency correction,
achieving 0.72-mm image resolution, and (b) ORC-VDS with 14 interleaves trading
o eld map resolution for 0.68-mm image resolution. The higher resolution ORCVDS image shows improved vessel denition (indicated by white arrows).
map) and then ORC-VDS with the 24 interleaves designed to achieve higher eld
map resolution (with the same spatial resolution). All three images have 0.74-mm
in-plane resolution over a 16-cm FOV. Both corrected images used multifrequency
reconstruction using the eld map; the conventional spiral image achieved 6.78mm eld-map resolution, while the ORC-VDS image achieved 5.2-mm eld-map
resolution with the same scan time. The ORC-VDS technique provided a 68%
improvement in eld map resolution (voxel area) which is apparent by the reduced
blurring of the comb bristles even in this well-shimmed phantom.
Figure 6.9 contains 2D coronary images from ECG triggered breath-held studies
on a normal volunteer. The volunteer was rst scanned with conventional spirals
(12 interleaves for the image and 2 excitations for the eld map) and then ORCVDS with the 14 interleaves designed to achieve higher image resolution (sacricing
Chapter 6. EÆcient O-resonance Correction
91
eld map resolution). Over a 20-cm FOV, the conventional spiral image has 0.72mm image resolution and 3.25-mm eld map resolution, while the ORC-VDS image
achieves 0.68-mm image resolution and 4-mm eld map resolution. By permitting
very low resolution eld map correction, ORC-VDS enabled the same scan time to
be used to achieve higher image resolution. Notice the improved vessel denition in
Fig. 6.9b (shown by white arrows). In such cases, higher image resolution may be
more valuable than eld map resolution.
6.5
Discussion
In summary, PR and spiral imaging sequences can be adapted to automatically
acquire a eld map. In PR, delayed acquisitions are simply interleaved with nondelayed acquisitions. In VDS imaging trajectories are designed to oversample the kspace origin, in addition to interleaving delayed and non-delayed acquisitions. A eld
map is computed from the central portion of k-space, and is used by a multifrequency
reconstruction to compensate for o-resonance artifacts in nal images. Phantom
and in vivo studies indicate this is a practical method for reducing o-resonance
induced blurring.
VDS simulations indicate this technique achieves higher image and/or eld map
resolution compared to acquiring a eld map in separate acquisitions. In addition
ORC-VDS has the ability to freely trade o image and eld map resolution. Field
maps acquired using this technique also show reduced o-resonance artifacts, because the eld map acquisition is spread out over all interleaves and is acquired over
a smaller time window.
For imaging ultra short T2 species, there is an added consideration of mixed
contrast. Since images are formed from acquisitions at two echo times, dierences
in contrast between those two echoes will produce streaking or swirling pattern
artifacts. This artifacts are similar to what is seen with azimuthal undersampled
Chapter 6. EÆcient O-resonance Correction
92
because both echo images are angularly undersampled during the nal image reconstruction. In general, this o-resonance correction technique is intended for areas
where a small TE does not signicantly alter image contrast.
This technique is best suited for applications that require short echo times, good
ow and motion properties, and that suer from o-resonance artifacts. While
we present this technique within the context of 2D PR and spiral imaging, simple
implementations exist for many 3D trajectories such as 3DPR, 3D stacks of spirals,
and 3D cones.
Chapter 7
Summary and Recommendations
This dissertation presents a series of MR engineering developments aimed at performing more eÆcient and useful cardiovascular examination using MRI. Within the
real-time interactive imaging paradigm which involves fast acquisition reconstruction and display, we have explored ow imaging, black blood contrast, and high
frame rate and high resolution imaging.
The following list summarizes key contributions:
The development of a system for interactive color ow imaging. Using previ-
ously established ow encoding techniques (phase contrast), fast acquisitions
(spirals), and color overlay (ultrasound colormaps), we have designed and
implemented a practical system for evaluating cardiac and vascular ow in
real-time. Phantom studies demonstrate the accuracy of measured velocities
and real-time waveforms, and in vivo studies indicate that many types of cardiovascular ow can be evaluated. Recent blinded clinical trials using this
system suggest that it can identify clinically signicant valvular regurgitation
comparable to echocardiography [111].
The design and implementation of a system for steady state blood suppression
during real-time scanning. Real-time spatial presaturation is used to improve
endocardial border denition in LV wall motion studies, and eliminate ow
93
Chapter 7. Summary and Recommendations
94
artifacts in real-time intravascular studies. Recent vessel wall imaging studies
have used this system to test and image with intravascular coils [112].
The development of an interactive multi-slice imaging system which achieves
frame rates suÆcient for whole-LV functional assessment under conditions of
stress. Using this system, resting LV function of 16 wall segments can be
evaluated in exams taking less than ve minutes. In addition, it is able to
image rst pass perfusion, and image function under stress conditions.
The development of an interactive real-time coronary artery imaging sequence,
with submillimeter resolution and short acquisition windows.
The conception, design and implementation of eÆcient o-resonance correction
algorithms for spiral and PR imaging. By slightly oversampling the k-space
origin, a eld map is acquired with little or no cost in scan time. This is used to
correct o-resonance in both PR and spiral imaging, and in the case of spirals,
can be used to freely tradeo eld map resolution for spatial resolution.
7.1
Future Work
One of the major benets of MRI is the ability to explore dierent aspects of cardiovascular disease in a single examination. Several of the projects described are
aimed at application areas currently dominated by other modalities. For example,
LV function and valve assessment are almost always done using ultrasound, and
coronary imaging is done using X-ray angiography. The evaluation of these in addition to myocardial perfusion, and other things, could potentially be completed in
a single comprehensive cardiac examination using MRI. One important area for future work is the integration of several MRI based techniques into a single sequence,
examination, or imaging protocol. In real-time imaging this would mean developing
ways by which an operator could quickly switch between dierent imaging sequences
while staying focused on anatomical landmarks (like the current scan plane).
Chapter 7. Summary and Recommendations
95
The following list contains more specic possibilities for future work, and is
organized by chapter:
Real-time Color Flow Imaging
Better understand artifacts caused by fast ow, and verify the causes of ow
signal dropout at the center of regurgitant jets.
Explore the quantitativeness of ow measurement using real-time spiral phase
contrast. Alternative methods besides phase dierence may be used to improve
volume ow calculation.
Use the sequence in combination with a 1D velocity mapping option, similar
to M-mode ultrasound. In other words, develop a thin beam 1D ow imaging
sequence that has high velocity resolution and high temporal resolution.
Explore other clinical applications, such as the assessment of ow in the portal
vein, or mesenteric artery and vein.
Real-time Black Blood Imaging
Use a single modulated or single upstream RF pulse to reduce the imaging
TR.
For intravascular studies, explore some variable density k-space trajectories,
so that higher resolution can be achieved in the short acquisition windows.
Real-time Multi-slice Stress Imaging
Incorporate fat suppression to avoid the geometric artifacts caused by fat
chemical shift, and to avoid having to time the acquisition of the k-space
origin with fat-water in-phase echoes.
Incorporate myocardium nulling prepulse for use in perfusion studies.
Chapter 7. Summary and Recommendations
96
Explore a four-slice orientation that fully captures the 17-segment LV model
from the American Society for Echocardiography, which includes a new apical
segment. This could be adequately covered with three short-axis views and
one long-axis view, or three long-axis views, and a single mid short-axis view.
Since these would not all be parallel slices, signal dropouts in areas of slice
overlap would have to be considered.
Explore other points along the temporal and spatial resolution curves. One
main goal would be 24 frames/s in all slices, which the literature suggests is
suÆcient for imaging under ischemic stress conditions.
Conduct further clinical evaluation of this sequence.
Real-time Coronary Imaging
Explore other points along the resolution curve, as well as other ratios of
asymmetry for the asymmetric pkCEPI acquisitions.
Thoroughly compare real-time images from high-resolution pkCEPI and spiral
acquisitions.
Further explore the use of long-duration intravascular contrast agents, with
the goal of rapid coronary screening.
O-resonance Correction for Radial Imaging
Conduct further in-vivo studies to assess the eectiveness of the artifact re-
duction.
Implement this technique with 3D trajectories such as 3D cones or 3DPR [113].
In addition to the above areas, the new MR techniques we present need to be
tested in patient populations to determine if they are indeed accurate alternatives
to current standard protocols.
Appendix A
Artifacts from Fast Flow
In 2D imaging, there is often the assumption of no through-plane motion during an
excitation, and no in-plane motion during a readout. Therefore, spin motion that
is not accounted for causes image artifacts.
In cardiovascular imaging, normal ow can be up to 1 m/s and abnormal ow
can be up to 5 or 10 m/s resulting in signicant movement during both excitation
and readout. In this appendix, we review artifacts caused by fast ow, and examine
specically those experienced by the spiral phase contrast imaging sequence from
Chapter 3.
As demonstrated by the signal equation, Equation 2.16, when linear gradient
are used, spins incur a phase shift related to their position, velocity, higher order
movement and the moments of the gradient waveform:
a
(A.1)
2 + );
where Mng represents the n -th moment of the gradient waveform, and ro, v, and
a represent initial position, velocity, and acceleration. Most sequences assume that
the phase contributions from the higher order terms (Eqn. A.1 in parenthesis)
= M0g ro + (M1g v + M2g 97
Appendix A. Artifacts from Fast Flow
98
are negligible. Here, we examine just the contributions of the velocity phase term
M1g v, and discuss ways to minimize its resulting artifacts.
Simulations were conducted using a Bloch simulator written in Matlab.
A.1
Through-Plane Flow and Spectral-Spatial
Excitation
During slice-selective excitation, an RF pulse is accompanied by a linear gradient
pulse in the slice-selective direction (direction perpendicular to the slice). This
gradient is used to vary the resonant frequency of spins at dierent positions while
spins are excited (see Figure 2.5). Static spins will accrue phase at a constant rate
based on their position, and are selectively excited if their frequency of precession
is within the pass-band of the RF excitation. Spins owing through the imaging
plane instead have a changing frequency of precession, and therefore may be inside
or outside the pass-band of the RF pulse at dierent times during the excitation
itself. Spins moving through-plane therefore experience a modied excitation tip
angle, phase, and slice prole.
In this thesis, we often use a spectral-spatial excitation [21] to excite only water
in a slice. This pulse is particularly vulnerable to through-plane ow because it
requires a longer pulse duration (during which greater movement can occur) and
delivers a large portion of the RF pulse energy when M1g is nonzero. In this
case, owing spins accrue an unexpected phase related to M1g v and experience a
dierent excitation compared to static spins. Figure A.1a illustrates the problematic
slice prole and phase accrual of moving spins. As noted by Fredrickson et al. [45],
yback excitations can be designed such that the majority of RF energy is deposited
when M1g is zero. In this case spins with constant through-plane velocity in the slice
also experience the expected excitation. Figure A.1b illustrates the more accurate
slice prole experienced when using the yback excitation.
99
Appendix A. Artifacts from Fast Flow
conventional
spectral-spatial
flyback
spectral-spatial
7 ms
8 ms
static
(water)
static
(fat)
velocity
50 cm/s
velocity
100 cm/s
Figure A.1: Simulations of through-plane ow during conventional and yback
spectral-spatial excitations. Simulated slice proles are shown for static water, static
fat, and water moving at 50 cm/s and 100 cm/s. Black lines indicate magnitude
proles jMxy j, while gray lines indicate phase proles 6 Mxy . For moving spin simulations, the horizontal axis represents spin position at the midpoint of the RF
excitation. Fat was simulated by applying a shift of -220 Hz to the spin resonant
frequency.
Appendix A. Artifacts from Fast Flow
100
In general, through-plane ow artifacts can be minimized by reducing the total
duration of RF excitation, and by delivering RF energy while higher order gradient
moments are close to zero.
A.2
In-Plane Flow and Spiral Acquisition
Similar phase errors related to higher order gradient moments occur during image
readouts. Since most of the energy in images is in the low spatial frequencies acquisitions are typically designed such that the central k-space is acquired when high
order moments are zero. However even with this provision, fast ow can cause signicant image artifacts because of movement and phase errors in the rest of k-space.
Nishimura et al. [47] and Gatehouse et al. [48] have used velocity k-space and velocity point spread function (PSF) analyses to show that in-plane ow during spiral
readouts causes a smoothly varying added phase in k-space, and an object domain
blurring in the direction of ow.
As shown in Figure A.2, the velocity PSF carries a varying phase in k-space,
and results in blurring and some ringing in the object domain. The images shown
are based on simulations of a 50 cm/s spin imaged using the spiral phase contrast
sequence from Chapter 3 which used 16 ms readouts. In the case of single spins,
signal is spread in the direction of ow, and its phase cycles with distance from its
original position (position at the start of readouts, identied by white circles). In
the case of extended objects such as a gaussian pulse of ow, much of the spreading
cancels out to yield moderate blurring in the direction of ow, shown in Figure A.2c.
Simulations of single spins and pulses of spins conrmed that taking the phase
dierence of images acquired with dierent ow encodes (as is done in phase contrast) results in accurate velocity measure throughout the spread or blurred pixels
showing signal. However, in a real setting with several pixels of ow at dierent
velocities and dierent locations, there may be destructive interference with other
owing pixels sending signal to the same image location. It is desirable to design
acquisitions that minimize the extent and phase of the velocity PSF.
Appendix A. Artifacts from Fast Flow
101
In general, in-plane ow artifacts can be reduced by shortening the duration of
readouts, and acquiring data while high order gradient moments are zero.
102
Appendix A. Artifacts from Fast Flow
a) k-space phase
b) object point-spread
phase
50 cm/s
magnitude
[-π,π]
c) extended object spread
magnitude
phase
[-π,π]
10 cm
Figure A.2: Flow eects during spiral acquisition. A single point spin moving at a
constant velocity causes a (a) smoothly varying phase in k-space, and (b) a spread
signal in the image. In the case of an extended object such as a gaussian pulse of ow
in the ow direction, the image is (c) blurred in the direction of ow. In the object
domain images, the center of the white circles/ellipses represent the position of the
point spins or extended objects at the start of spiral readouts. These simulation
images used a velocity of 50 cm/s and the spiral readouts from Chapter 3.
Bibliography
[1] F. Bloch, W. Hansen, and M. Packard, \Nuclear induction," Phys. Rev.,
vol. 69, p. 127, 1946.
[2] E. Purcell, H. Torrey, and R. Pound, \Resonance absorption by nuclear magnetic moments in a solid," Phys. Rev., vol. 69, p. 37, 1946.
[3] W. Proctor and F. Yu, \The dependence of a nuclear magnetic resonance
frequency upon chemical compound," Phys. Rev., vol. 77, p. 717, 1950.
[4] P. Lauterbur, \Image formation by induced local interactions: Examples employing nuclear magnetic resonance," Nature, vol. 242, p. 190, 1973.
[5] J. Mallard, J. Hutchison, W. Edelstein, C. Ling, M. Foster, and G. Johnson,
\In vivo NMR imaging in medicine: the Aberdeen approach, both physical
and biological," Philos. Trans. R. Soc. Lond. B. Biol. Sci., vol. 289, no. 1037,
pp. 519{533, 1980.
[6] S. Ljunggren, \A simple graphical representation of Fourier-based imaging
methods," J. Magn. Reson., vol. 54, no. 2, pp. 338{343, 1983.
[7] D. Twieg, \The k-trajectory formulation of the NMR imaging process with
applications in analysis and synthesis of imaging methods," Med. Phys.,
vol. 10, no. 5, pp. 610{621, 1983.
[8] K. Nayak, J. Pauly, A. Kerr, B. Hu, and D. Nishimura, \Real-time color ow
MRI," Magn. Reson. Med., vol. 43, pp. 251{258, 2000.
103
BIBLIOGRAPHY
104
[9] K. Nayak, P. Rivas, J. Pauly, G. Scott, A. Kerr, B. Hu, and D. Nishimura,
\Real-time black-blood MRI using spatial presaturation," J. Magn. Reson.
Imaging. in press.
[10] K. Nayak, J. Pauly, A. Kerr, D. Nishimura, and B. Hu, \Rapid ventricular
assessment using real-time interactive multislice MRI," Magn. Reson. Med. in
press.
[11] K. Nayak, J. Pauly, P. Yang, A. Kerr, C. Meyer, B. Hu, and D. Nishimura,
\Real-time coronary MRA," J. Cardiovasc. Magn. Reson., vol. 1, no. 4, p.
337, 2000.
[12] K. Nayak, J. Pauly, B. Hu, P. Yang, C. Meyer, A. Kerr, and D. Nishimura,
\Real-time interactive coronary MR angiography," Circulation, vol. 102, no. 18
supplement, pp. I{398, 2000.
[13] K. Nayak and D. Nishimura, \Automatic eld map generation and oresonance correction for PR imaging," Magn. Reson. Med., vol. 43, pp. 151{
154, 2000.
[14] K. Nayak, C.-M. Tsai, C. Meyer, and D. Nishimura, \EÆcient o-resonance
correction for spiral imaging," Magn. Reson. Med. in press.
[15] D. Nishimura. Principles of magnetic resonance imaging. Stanford University
EE369B Course Notes, 1996.
[16] P. Morris, Nuclear Magnetic Resonance Imaging in Medicine and Biology.
Oxford University Press, 1986.
[17] Z.-P. Liang and P. Lauterbur, Principles of Magnetic Resonance Imaging: A
Signal Processing Perspective. IEEE Press Series in Biomedical Engineering,
1999.
[18] E. Haacke, R. Brown, M. Thompson, and R. Venkatesan, Magnetic Resonance
Imaging: Physical Principles and Sequence Design. John Wiley and Sons,
1999.
BIBLIOGRAPHY
105
[19] J. Pauly, D. Nishimura, and A. Macovski, \A k-space analysis of small tip
excitation," J. Magn. Reson., vol. 81, pp. 43{56, 1989.
[20] D. Hoult, \The solution of the Bloch equations in the presence of a varying
B1 eld: An approach to selective pulse analysis," J. Magn. Reson., vol. 35,
pp. 66{86, 1979.
[21] C. H. Meyer, J. M. Pauly, A. Macovski, and D. Nishimura, \Simultaneous
spatial and spectral selective excitation," Magn. Reson. Med., vol. 15, pp.
287{304, 1990.
[22] P. Manseld, \Multi-planar image formation using NMR spin echoes," J. Phys.
C, vol. 10, pp. 580{594, 1977.
[23] C. Meyer, B. Hu, D. Nishimura, and A. Macovski, \Fast spiral coronary artery
imaging," Magn. Reson. Med., vol. 28, no. 2, pp. 202{213, 1992.
[24] J. Cooley and J. Tukey, \An algorithm for the machine calculation of complex
Fourier series," Math. Comput., vol. 19, pp. 297{301, 1965.
[25] J. O'Sullivan, \A fast sinc function gridding algorighm for Fourier inversion
in computer tomography," IEEE Trans. Med. Imaging, vol. 4, pp. 200{207,
1985.
[26] J. I. Jackson, C. H. Meyer, and A. M. Dwight G. Nishimura, \Selection of a
convolution function for Fourier inversion using gridding," IEEE Trans. Med.
Imaging, vol. 10, no. 3, pp. 473{478, September 1991.
[27] G. McGibney, M. R. Smith, S. T. Nichols, and A. Crawley, \Quantitative
evaluation of several partial Fourier reconstruction algorithms used in MRI,"
Magn. Reson. Med., vol. 30, pp. 51{59, 1993.
[28] D. C. Noll, D. G. Nishimura, and A. Macovski, \Homodyne detection in magnetic resonance imaging," IEEE Trans. Med. Imaging, vol. 10, no. 2, pp. 154{
163, June 1991.
BIBLIOGRAPHY
106
[29] A. B. Kerr, J. M. Pauly, B. S. Hu, K. C. Li, C. J. Hardy, C. H. Meyer, A. Macovski, and D. G. Nishimura, \Real-time interactive MRI on a conventional
scanner," Magn. Reson. Med., vol. 38, pp. 355{367, 1997.
[30] C. Otto and A. Pearlman, Textbook of Clinical Echocardiography. Philadelphia:
WB Saunders, 1994.
[31] S. Riederer, T. Tasciyan, F. Farzaneh, J. Lee, R. Wright, and R. Herfkens,
\MR uoroscopy: Technical feasibility," Magn. Reson. Med., vol. 8, no. 1, pp.
1{15, 1988.
[32] A. Holsinger, R. Wright, S. Riederer, F. Farzaneh, R. Grimm, and J. Maier,
\Real-time interactive magnetic resonance imaging," Magn. Reson. Med.,
vol. 14, no. 3, pp. 547{553, 1990.
[33] V. Rasche, R. de Boer, D. Holz, and R. Proksa, \Continuous radial data
acquisition for dynamic MRI," Magn. Reson. Med., vol. 34, no. 5, pp. 754{
761, 1995.
[34] C. Hardy, R. Darrow, E. Nieters, P. Roemer, R. Watkins, W. Adams, N. Hattes, and J. Maier, \Real-time acquisition, display, and interactive graphic control of NMR cardiac proles and images," Magn. Reson. Med., vol. 29, no. 5,
pp. 667{673, 1993.
[35] A. Gmitro, A. Ehsani, T. Berchem, and R. Snell, \A real-time reconstruction
system for magnetic resonance imaging," Magn. Reson. Med., vol. 35, no. 5,
pp. 734{740, 1996.
[36] T. Haishi and K. Kose, \Real-time image reconstruction and display for MRI
using a high-speed personal computer," J. Magn. Reson., vol. 134, no. 1, pp.
138{141, 1998.
[37] P. C. Yang, A. B. Kerr, A. C. Liu, D. H. Liang, C. J. Hardy, C. H. Meyer,
A. Macovski, J. M. Pauly, and B. S. Hu, \New real-time interactive magnetic
BIBLIOGRAPHY
107
resonance imaging complements echocardiography," J. Amer. Coll. Cardiol.,
vol. 32, pp. 2049{2056, 1998.
[38] F. Chan, K. Li, R. Shifrin, J. Pauly, A. Kerr, G. Gold, E. Hansch, R. Mindelzun, M. Nino-Murcia, E. Olcott, and T. Ha, \Real-time interactive MR
imaging of the small bowel," RSNA Electronic Journal, vol. 1, 1997. See
http://ej.rsna.org/.
[39] R. H. Klipstein, D. N. Firmin, S. R. Underwood, G. L. Nayler, R. S. Rees,
and D. B. Longmore, \Colour display of quantitative blood ow and cardiac anatomy in a single magnetic resonance cine loop," Br. J. Radiol.,
vol. 60, no. 710, pp. 105{111, 1987.
[40] T. Tasciyan, D. Mitchell, and C. Spritzer, \Color ow-encoded MR imaging,"
J. Magn. Reson. Imaging, vol. 1, pp. 715{720, 1991.
[41] S. Fischer, S. Wickline, and C. Lorenz, \Virtual transducer color ow MRI of
septal defects and valvular disease," in Proc., ISMRM, 7th Annual Meeting
(Philadelphia), p. 1291, 1999.
[42] S. Riederer, R. Wright, R. Ehman, P. Rossman, A. Holsinger-Bampton,
N. Hangiangdreu, and R. Grimm, \Real-time interactive color ow MR imaging," Radiology, vol. 181, pp. 33{39, 1991.
[43] P. Gatehouse, D. Firmin, S. Collins, and D. Longmore, \Real time blood ow
imaging by spiral scan phase velocity mapping," Magn. Reson. Med., vol. 31,
pp. 504{512, 1994.
[44] G. Pike, C. Meyer, T. Brosnan, and N. Pelc, \Magnetic resonance velocity
imaging using a fast spiral phase contrast sequence.," Magn. Reson. Med.,
vol. 32, no. 4, pp. 476{483, 1994.
[45] J. Fredrickson, C. Meyer, and N. Pelc, \Flow eects of spectral spatial excitation," in Proc., ISMRM, 5th Annual Meeting (Vancouver), p. 113, 1997.
BIBLIOGRAPHY
108
[46] J. M. Pauly, P. L. Roux, D. G. Nishimura, and A. Macovski, \Parameter
relations for the Shinnar-Le Roux selective excitation pulse design algorithm,"
IEEE Trans. Med. Imaging, vol. 10, no. 1, pp. 53{65, 1991.
[47] D. G. Nishimura, P. Irarrazabal, and C. H. Meyer, \A velocity k-space analysis of ow eects in echo-planar and spiral imaging," Magn. Reson. Med.,
vol. 33, no. 4, pp. 549{556, 1995.
[48] P. Gatehouse and D. Firmin, \Flow distortion and signal loss in spiral imaging," Magn. Reson. Med., vol. 41, pp. 1023{1031, 1999.
[49] C. Meyer, J. Pauly, and A. Macovski, \A rapid, graphical method for optimal
spiral gradient design," in Proc., ISMRM, 4th Annual Meeting (New York),
p. 392, 1996.
[50] K. King, T. Foo, and C. Crawford, \Optimized gradient waveforms for spiral
scanning," Magn. Reson. Med., vol. 34, no. 2, pp. 156{160, 1995.
[51] E. Hahn, \Detection of sea-water motion by nuclear precession," J. Geophys.
Res., vol. 65, no. 2, pp. 776{777, 1960.
[52] K. S. Nayak, P. A. Rivas, A. B. Kerr, J. M. Pauly, B. S. Hu, and D. G.
Nishimura, \Applications of real-time MRI with color ow mapping: Fast and
slow ow," in Proceedings of Society for Cardivascular Magnetic Resonance
(Atlanta), p. 54, 1999.
[53] N. Hangiandreou, P. Rossman, and S. Riederer, \Analysis of MR phasecontrast measurements of pulsatile velocity waveforms," J. Magn. Reson.
Imaging, vol. 3, pp. 387{394, 1993.
[54] A. Kerr, J. Pauly, C. Meyer, and D. Nishimura, \New strategies in spiral MR
uoroscopy," in Proc., SMR, 3rd Annual Meeting (Nice), p. 99, 1995.
[55] G. Nayler, D. Firmin, and D. Longmore, \Blood ow imaging by cine magnetic
resonance," J. Comput. Assist. Tomogr., vol. 10, pp. 715{722, 1986.
BIBLIOGRAPHY
109
[56] N. Pelc, R. Herfkens, A. Shimakawa, and D. Enzmann, \Phase-contrast cine
magnetic resonance imaging," Magn. Reson. Q., vol. 7, pp. 229{254, 1991.
[57] D. N. Firmin, G. L. Nayler, R. H. Klipstein, S. R. Underwood, R. S. Rees,
and D. B. Longmore, \In vivo validation of MR velocity imaging," J. Comput.
Assist. Tomogr., vol. 11, no. 5, pp. 751{756, 1987.
[58] L. R. Pelc, N. J.Pelc, S. C. Rayhill, L. J. Castro, G. H. Glover, R. J. Herfkens,
D. C. Miller, and R. B. Jerey, \Arterial and venous blood ow: noninvasive
quantitation with MR imaging," Radiology, vol. 185, no. 3, pp. 809{812, 1992.
[59] M. Hofman, A. van Rossum, M. Sprenger, and N. Westerhof, \Assessment of
ow in the right human coronary artery by magnetic resonance phase contrast velocity measurement: eects of cardiac and respiratory motion," Magn.
Reson. Med., vol. 35, pp. 521{531, 1996.
[60] M. Hofman, S. Wickline, and C. Lorenz, \Quantication of in-plane motion
of the coronary arteries during the cardiac cycle: implications for acquisition
window duration for MR ow quantication," J. Magn. Reson. Imaging, vol. 8,
pp. 568{576, 1998.
[61] R. Edelman, D. Chien, and D. Kim, \Fast selective black blood MR imaging,"
Radiology, vol. 181, pp. 655{660, 1991.
[62] J. Debbins, S. Riederer, P. Rossman, R. Grimm, J. Felmlee, J. Breen, and
R. Ehman, \Cardiac magnetic resonance uoroscopy," Magn. Reson. Med.,
vol. 36, pp. 588{595, 1996.
[63] R. Edelman, H. Mattle, B. Wallner, R. Bajakian, J. Kleeeld, C. Kent, J. Skillman, J. Mendel, and D. Atkinson, \Extracranial carotid arteries: evaluation
with black blood MR angiography," Radiology, vol. 177, pp. 45{50, 1990.
[64] J. Felmlee and R. Ehman, \Spatial presaturation: a method for suppressing
ow artifacts and improving depiction of vascular anatomy in MR imaging,"
Radiology, vol. 164, no. 2, pp. 559{564, 1987.
BIBLIOGRAPHY
110
[65] K. S. Nayak, J. M. Pauly, B. S. Hu, A. B. Kerr, and D. G. Nishimura, \Realtime black-blood cardiac MRI," Circulation, vol. 100, no. 18 supplement, pp.
I{162, 1999.
[66] J. Ridgway, A. Kassner, U. Sivanathan, and M. Smith, \Multiple-phase, blackblood imaging of the heart using real-time acquisition: Initial results," in
Proc., ISMRM, 6th Annual Meeting (Sydney), p. 807, 1998.
[67] K. S. Nayak, J. M. Pauly, A. B. Kerr, and D. G. Nishimura, \Real-time blackblood MRI," in Proc., ISMRM, 7th Annual Meeting (Philadelphia), p. 1638,
1999.
[68] J. Pauly, D. Nishimura, and A. Macovski, \A linear class of large-tip-angle
selective excitation pulses," J. Magn. Reson., vol. 82, pp. 571{587, 1989.
[69] P. Rivas, M. McConnell, K. Nayak, G. Scott, C. Meyer, J. Pauly, D. Nishimura,
A. Macovski, and B. Hu, \Real-time intravascular magnetic resonance receiver
probe: in vivo observations in the rabbit aorta," in Proc., ISMRM, 7th Annual
Meeting (Philadelphia), p. 82, 1999.
[70] E. Atalar, P. Bottomley, O. Ocali, L. Correia, D. Kelemen, J. Lima, and
E. Zerhouni, \High resolution intravascular MRI and MRS by using a catheter
receiver coil," Magn. Reson. Med., vol. 36, pp. 596{605, 1996.
[71] E. Nagel, U. Schneider, S. Schalla, T. Ibrahim, B. Schnackenburg, A. Bornstedt, C. Klein, H. Lehmkuhl, and E. Fleck, \Magnetic resonance real-time
imaging for the evaluation of left ventricular function," J. Cardiovasc. Magn.
Reson., vol. 2, no. 1, pp. 7{14, 2000.
[72] S. Wolfe and R. Balaban, \Assessing contrast on MR images," Radiology,
vol. 202, pp. 25{29, 1997.
[73] A. Martin and R. Henkelman, \Intravascular MR imaging in a porcine animal
model," Magn. Reson. Med., vol. 32, no. 2, pp. 224{229, Aug. 1994.
BIBLIOGRAPHY
111
[74] A. Arai, F. Epstein, K. Bove, and S. Wol, \Visualization of aortic valve
leaets using black blood MRI," J. Magn. Reson. Imaging, vol. 10, no. 5, pp.
771{777, 1999.
[75] Heart and stroke 1998 statistical update. American Heart Association, 1998.
[76] E. Nagel, H. Lehmkuhl, C. Klein, U. Schneider, E. Frantz, A. Ellmer,
W. Bocksch, S. Dreysse, and E. Fleck, \Inuence of image quality on the
diagnostic accuracy of dobutamine stress magnetic resonance imaging in comparison with dobutamine stress echocardiography for the noninvasive detection
of myocardial ischemia.," Z. Kardiol., vol. 88, no. 9, pp. 622{30, Sep 1999.
[77] W. Hundley, C. Hamilton, M. Thomas, D. Herrington, T. Salido, D. Kitzman,
W. Little, and K. Link, \Utility of fast cine magnetic resonance imaging and
display for the detection of myocardial ischemia in patients not well suited for
second harmonicstress echocardiography.," Circulation, vol. 100, no. 16, pp.
1697{702, Oct 1999.
[78] F. Epstein, S. Wol, and A. Arai, \Segmented k-space fast cardiac imaging
using an echo-train readout," Magn. Reson. Med., vol. 41, no. 3, pp. 609{613,
Mar 1999.
[79] H. Mertes, S. Sawada, T. Ryan, D. Segar, R. Kovacs, J. Foltz, and H. Feigenbaum, \Symptoms, adverse eects and complications associated with dobutamine stress echocardiography: experience in 1118 patients," Circulation,
vol. 88, pp. 15{19, 1993.
[80] F. van Rugge, J. Boreel, E. van der Wall, P. van Dijkman, A. van der Laarse,
J. Doornbos, A. de Roos, J. de Boer, A. Bruschke, and A. van Voorthuisen,
\Cardiac rst pass and myocardial perfusion in normal subjects assessed by
sub-second Gd-DTPA enhanced MR imaging," J. Comput. Assist. Tomogr.,
vol. 15, pp. 959{65, 1991.
[81] W. Manning, D. Atkinson, W. Grossman, S. Paulin, and R. Edelman, \Firstpass nuclear magnetic resonance imaging studies using gadolinium-DTPA in
BIBLIOGRAPHY
112
patients with coronary artery disease," J. Amer. Coll. Cardiol., vol. 18, pp.
959{65, 1991.
[82] N. Wilke, K. Kroll, H. Merkle, Y. Wang, Y. Ishibashi, Y. Xu, J. Zhang,
M. Jerosch-Herold, A. Muhler, and A. Stillman, \Regional myocardial blood
volume and ow: rst-pass MR imaging with polylysine-Gd-DTPA," J. Magn.
Reson. Imaging, vol. 5, pp. 227{37, 1995.
[83] J. Schwitter, D. Leung, J. Debatin, G. von Schulthess, and G. McKinnon,
\Evaluation of myocardial perfusion: T1-weighted multislice multiphase spinecho EPI," Eur. J. Radiol., vol. 5, pp. s37{42, 1995.
[84] N. Wilke, M. Jerosch-Herold, Y. Wang, Y. Huang, B. Christensen, A. Stillman,
K. Ugurbil, K. McDonald, and R. Wilson, \Myocardial perfusion reserve: assessment with multisection, quantitative, rst-pass MR imaging," Radiology,
vol. 204, pp. 373{84, 1997.
[85] M. Weiger, K. Pruessmann, and P. Boesiger, \Cardiac real-time imaging using
SENSE," Magn. Reson. Med., vol. 43, pp. 177{184, 2000.
[86] M. Stuber, K. Kissinger, M. Chen, P. Danias, and W. Manning, \Non-ECG
triggered multi-slice real-time cardiac dobutamine stress imaging," in Proc.,
ISMRM, 8th Annual Meeting (Denver), p. 194, 2000.
[87] N. Schiller, P. Shah, M. Crawford, A. DeMaria, R. Devereux, H. Feigenbaum,
H. Gutgesell, N. Reichek, D. Sahn, and I. Schnittger, \Recommendations for
quantitation of the left ventricle by two-dimensional echocardiography," J.
Am. Soc. Echo., vol. 2, pp. 358{367, 1989. American Society of Echocardiography Committee on Standards, Subcommittee on Quantitation of TwoDimensional Echocardiograms.
[88] R. Homan, H. Lethen, T. Marwick, M. Arnese, P. Fioretti, A. Pingitore,
E. Picano, T. Buck, R. Erbel, F. Flachskampf, and P. Hanrath, \Analysis of
interinstitutional observer agreement in interpretation of dobutamine stress
echocardiograms," J. Amer. Coll. Cardiol., vol. 27, no. 2, pp. 330{336, 1996.
BIBLIOGRAPHY
113
[89] A. Kerr, J. Pauly, and D. Nishimura, \Partial k-space acquisition for real-time
MR imaging," in Proc., ISMRM, 6th Annual Meeting (Sydney), p. 1945, 1998.
[90] G. Luk Pat, C. Meyer, J. Pauly, and D. Nishimura, \Reducing ow artifacts
in echo-planar imaging," Magn. Reson. Med., vol. 37, pp. 436{447, 1997.
[91] K. Bjornstad, S. Aakhus, and H. Torp, \How does computer-assisted digital
wall motion analysis inuence observer agreement and diagnostic accuracy
during stress echocardiography?," Int. J. Card. Imaging, vol. 13, pp. 105{14,
1997.
[92] D. Sodickson and W. Manning, \Simultaneous acquisition of spatial harmonics
(SMASH): Fast imaging with radiofrequency coil arrays," Magn. Reson. Med.,
vol. 38, pp. 591{603, 1997.
[93] K. Pruessmann, M. Weiger, M. Scheidegger, and P. Boesiger, \SENSE: Sensitivity encoding for fast MRI," Magn. Reson. Med., vol. 42, pp. 952{962,
1999.
[94] A. Duerinckx and M. Urman, \Two-dimensional coronary MR angiography:
analysis of initial clinical results," Radiology, vol. 193, no. 3, pp. 731{738,
1994.
[95] W. Manning, W. Li, and R. Edelman, \A preliminary report comparing magnetic resonance coronary angiography with conventional angiography," N.
Engl. J. Med., vol. 328, no. 12, pp. 828{832, 1993.
[96] C. Paschal, E. Haacke, and L. Adler, \Three-dimensional MR imaging of the
coronary arteries: preliminary clinical experience," J. Magn. Reson. Imaging,
vol. 3, no. 3, pp. 491{500, 1993.
[97] C. Hardy, R. Darrow, J. Pauly, A. Kerr, C. Dumoulin, B. Hu, and K. Martin,
\Interactive coronary MRI," Magn. Reson. Med., vol. 40, pp. 105{111, 1998.
BIBLIOGRAPHY
114
[98] C. Hardy, M. Saranathan, Y. Zhu, and R. Darrow, \Coronary angiography
by real-time MRI with adaptive averaging," Magn. Reson. Med., vol. 44, pp.
940{946, 2000.
[99] C. Meyer and J. Pauly. A rapid method of optimal gradient waveform design
for k-space scanning in MRI. U.S. Patent Number 6,020,739, 2000.
[100] G. Johnson and J. Hutchison, \The limitations of NMR recalled-echo imaging
techniques," J. Magn. Reson., vol. 63, pp. 14{30, 1985.
[101] G. Gold, J. Pauly, A. Macovski, and R. Herfkens, \MR spectroscopic imaging
of collagen: Tendons and knee menisci," Magn. Reson. Med., vol. 34, pp.
647{654, 1995.
[102] P. Smith, T. Peters, and R. Bates, \Image reconstruction from nite numbers
of projections," J. Phys. A: Math, Nucl. Gen., vol. 6, pp. 361{383, 1973.
[103] D. Peters, C. Mistretta, F. Korosec, J. Holden, F. Kelcz, K. Wedding, and
T. Grist, \Using projection reconstruction with a limited number of projections to increase image resolution or acquisition speed," in Proc., ISMRM, 6th
Annual Meeting (Sydney), p. 182, 1998.
[104] K. Scheer and J. Hennig, \Reduced circular eld-of-view imaging," Magn.
Reson. Med., vol. 40, no. 3, pp. 474{480, 1998.
[105] V. Rasche, D. Holz, J. Kohler, and P. Roschmann, \Catheter tracking using
continuous radial MRI," Magn. Reson. Med., vol. 37, p. 963, 1997.
[106] C.-M. Tsai and D. Nishimura, \Reduced aliasing artifacts using variabledensity k-space sampling trajectories," Magn. Reson. Med., vol. 43, pp. 452{
458, 2000.
[107] D. Noll, C. Meyer, J. Pauly, D. Nishimura, and A. Macovski, \A homogeneity
correction method for magnetic resonance imaging with time-varying gradients," IEEE Trans. Med. Imag., vol. 10, no. 4, pp. 629{637, 1991.
BIBLIOGRAPHY
115
[108] G. Luk Pat, A. Kerr, and D. Nishimura, \Inhomogeneity correction for echoplanar imaging with a polynomial estimate of the eld map," in Proc., SMR,
3rd Annual Meeting (Nice), p. 617, Aug. 1995.
[109] L.-C. Man, J. Pauly, and A. Macovski, \Multifrequency interpolation for fast
o-resonance correction," Magn. Reson. Med., vol. 37, pp. 785{792, 1997.
[110] E. Yudilevich and H. Stark, \Spiral sampling in magnetic resonance imaging{
the eect of inhomogenaeities," IEEE Trans. Med. Imaging, vol. 6, pp. 337{
345, 1987.
[111] P. Rivas, K. Nayak, M. McConnell, A. Kerr, P. Yang, D. Nishimura, J. Pauly,
and B. Hu, \Real-time color ow MRI assessment of valvular regurgitation,"
J. Cardiovasc. Magn. Reson. manuscript in preparation.
[112] P. Rivas, K. Nayak, G. Scott, , M. McConnell, A. Kerr, D. Nishimura, J. Pauly,
and B. Hu, \In vivo real-time intravascular MRI," J. Cardiovasc. Magn. Reson. submitted.
[113] P. Irarrazabal and D. Nishimura, \Fast three-dimensional magnetic resonance
imaging," Magn. Reson. Med., vol. 33, no. 5, pp. 656{662, 1995.