Quantitative Vascular Elastography:

Stiffness and Stress Estimation for

Identifying Rupture-Prone Plaques
by

Steven J. Huntzicker

Submitted in Partial Fulfillment

of the

Requirements for the Degree

Doctor of Philosophy

Supervised by

Professor Marvin M. Doyley

Department of Electrical and Computer Engineering

Arts, Sciences and Engineering
Edmund A. Hajim School of Engineering and Applied Sciences

University of Rochester
Rochester, New York

2015
 ii

Dedicated to my family
 iii

Biographical Sketch

Steven Huntzicker grew up in Ann Arbor, MI. He attended Harvey Mudd College
in Claremont, CA after graduating from Ann Arbor Huron High School in 2005.
At Harvey Mudd, he studied General Engineering, with a focus on Electrical
Engineering. He graduated with a Bachelor of Science degree in the spring of 2009.
That fall he enrolled in the doctoral program at the University of Rochester Hajim
School of Engineering and Applied Science in the department of Electrical and
Computer Engineering. In 2010 Steven joined the Parametric Imaging Research
Laboratory (PIRL), under the direction of Dr. Marvin Doyley. His research has
been focused on ultrasound-based vascular elastography.

Journal Publications
Steven Huntzicker, Rohit Nayak, and Marvin M. Doyley. Quantitative
sparse array vascular elastography: the impact of tissue attenuation and modulus
contrast on performance. Journal of Medical Imaging 1, no. 2 (2014): 027001-
027001.

Steven Huntzicker, Himanshu Shekhar, and Marvin M. Doyley. Contrast-

enhanced quantitative intravascular elastography: The impact of microvasculature
on model based elatography. Ultrasound in Medicine and Biology, under review.

Steven Huntzicker and Marvin M. Doyley, Nonlinear tissue reconstruc-

tion in quantitative vascular elastography. Physics in Medicine and Biology, in
preparation.
 iv

Steven Huntzicker, Rohit Nayak, and Marvin M. Doyley, Clinical applica-

tions of quantitative vascular elastography. Ultrasound in Medicine and Biology,
in preparation.

Conference Abstracts
Steven Huntzicker, Sanghamithra Korukonda, and Marvin M. Doyley. Re-
constructing the mechanical properties of coronary arteries from displacements
measured with a synthetic aperture ultrasound imaging system. In SPIE Medi-
cal Imaging, pp. 867511-867511. International Society for Optics and Photonics,
2013.

Steven Huntzicker and Marvin M. Doyley. Can quantitative synthetic

aperture vascular elastography predict the stress distribution within the fibrous
cap non-invasively. The Journal of the Acoustical Society of America 135, no. 4
(2014): 2372-2372.
 v

Acknowledgments

Obtaining my PhD has been a long road, not all of which has been paved, with
the odd pothole here and there to keep me alert. Of course, there have also been
stretches of pastoral brilliance, as I wended my way through the landscape of
Western New York. I certainly could not have completed this journey without the
assistance and support of the many people who have helped me along the way.

I first have to thank my advisor Dr. Marvin Doyley, without whose guidance
and faith in my abilities I surely would not be here today. He taught me how
to become a better researcher and how to balance engineering techniques with
scientific principles. These foundations will help me through my entire career.

I next have to acknowledge the members of my committee: Dr. Akhtar Khan,

Dr. Kevin Parker, and Dr. Giovanni Schifitto. The feedback from their varied
backgrounds helped to craft my thesis, and contextualize my research from a
broader perspective.

I would like to acknowledge my fellow members of the Parametric Imaging

Research Laboratory, both past and present, who have provided assistance, who
have been insightful sounding boards for my ideas, and who have been my friends:
Dr. Jiang Yao, Dr. Ramin Eslami, Dr. Sanghamithra Korukonda, Dr. Michael
Richards, Dr. Himanshu Shekhar, Rohit Nayak, Prashant Verma, Shayin Jing,
Xing Sun, and Oscar Osapoetra. I must also thank the greater RCBU community
for their input and providing me with a greater understanding of the myriad
therapeutic and imaging applications of biomedical ultrasound.
 vi

Behind the scenes, many staff members in the departments of Electrical and
Computer Engineering as well as the Department of Biomedical Engineering have
diligently helped me to achieve my goals. Linda Weidman has made the lab run
smoothly, and has proofread my thesis as well as countless other documents. Paul
Osborne machined many of the phantom molds and various apparati used in my
experiments; Art Salo provided many hours of his time assisting me in conducting
mechanical testing experiments, Barbara Dick and Michelle Foster have assisted
with administrative details while Ive been at the University of Rochester and
insured that my thesis was processed and approved; and finally, John Strong, John
Simonson, Jim Prescot, and Will DiGrazio have maintained all of the computer
systems that I have used at the University of Rochester.

One of my most pleasurable activities while at the University of Rochester is

having the opportunity to be a member of the University of Rochester Symphony
Orchestra. It has been a wonderful creative outlet and has kept me centered over
the past six years. I therefore have to thank the conductor, Dr. David Harman,
as well as my fellow clarinetists.

Finally, I give the utmost thanks to my family: To my parents, Fred and Debra
Huntzicker, who have provided immeasurable moral support (and cookies), and
to my sister, Kathleen Huntzicker, who never fails to make me laugh.
 vii

Abstract

Every 40 seconds, someone in the United States dies of cardiovascular disease,

with many of these deaths occurring from the rupture of atherosclerotic plaques
in the carotid and coronary arteries. These events cause thombi to form in the
arteries, which prevent blood flow. Clots in the carotid artery may lead to stroke,
while clots in the coronary artery may lead to myocardial infarction. Therefore,
there is a need for an imaging technique that can identify rupture-prone plaques.
One proposed method for this is ultrasound-based elastography, which maps the
stiffness of the vessel based on the extent of observed tissue motion. Plaques have
been shown to rupture when the internal stresses exceed 300 kPa, and conventional
elastography measures strain, which is proportional to stress. Absolute values
of stress are needed to determine the likelihood of plaque rupture. To obtain
these stresses, one must measure Youngs modulus and all components of strain.
Computing these using ultrasound is challenging due to the poor lateral resolution.

The objective of this thesis was to investigate the feasibility of using a model-
based approach to reconstruct the modulus and stresses within arteries using clin-
ically available ultrasound systems. To achieve this goal the following objectives
had to be satisfied: (1) Develop accurate inversion schemes for estimating the
modulus distribution within vessels; (2) assess the impact of microvessels on the
performance of the techniques developed in (1) and develop methods to overcome
its limitations; (3) investigate the effects of material nonlinearity on the ability to
 viii

visualize the stress distribution within vascular tissues; (4) assess the feasibility
of producing clinically useful images.

The results of these studies showed that the proposed reconstruction method
could accurately compute modulus and stress elastograms. In all simulation stud-
ies, peak stresses could be recovered with under 15% error using the proposed
reconstruction method. Phantom and in vivo studies were also able to accurately
reconstruct these parameters despite the presence of measurement noise, atten-
uation, and sub-resolution features. This research may be useful in the further
advancement of clinical vascular imaging.
 ix

Contributors and Funding Sources

This work was overseen by my thesis committee, consisting of Dr. Marvin Doyley
(advisor), Prof. Akhtar Khan (Center for Applied and Computational Mathe-
matics - Rochester Institute of Technology), Prof. Kevin Parker (Department of
Electrical and Computer Engineering), and Dr. Giovanni Schifitto (University of
Rochester Medical Center).

The Field II ultrasound simulation codes were developed by Dr. Sanghamithra

Korukonda and Rohit Nayak. The cross-correlation and registration-based dis-
placement tracking routine used in this paper are based on algorithms developed
by Dr. Sanghamithra Korukona, and Dr. Michael Richards.

The work in Chapter 4 of this dissertation is based on results published in the

Journal of Medical Imaging in 2014 with Dr. Marvin Doyley and Rohit Nayak
as co-authors. The work in Chapter 5 of this dissertation is based on results
submitted to the journal Ultrasound in Medicine and Biology, with Dr. Marvin
Doyley and Dr. Himanshu Shekhar as co-authors.

The work presented in this thesis was supported by the National Heart and
Lung Research Grant R01 HL088523.
 x

Table of Contents

Biographical Sketch iii

Acknowledgments v

Abstract vii

Contributors and Funding Sources ix

List of Tables xiii

List of Figures xiv

1 General Introduction 1

1.1 Pathophysiology of Plaques . . . . . . . . . . . . . . . . . . . . . 1

1.2 Imaging Techniques for Atherosclerosis and Plaque Vulnerability . 3

1.3 Thesis Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.4 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2 Background 14

2.1 Review of Vascular Elasticity . . . . . . . . . . . . . . . . . . . . 15

2.2 Review of Direct Elastography . . . . . . . . . . . . . . . . . . . . 22

 xi

2.3 Review of Model-Based Elastography . . . . . . . . . . . . . . . . 28

2.4 Current Limitations of Quantitative Elastography . . . . . . . . . 36

3 General Methods 38

3.1 Finite Element Modeling . . . . . . . . . . . . . . . . . . . . . . . 38

3.2 RF Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3.3 Displacement Estimation . . . . . . . . . . . . . . . . . . . . . . . 42

3.4 Strain and Stress Estimation . . . . . . . . . . . . . . . . . . . . . 43

3.5 Soft-Priors Reconstruction . . . . . . . . . . . . . . . . . . . . . . 44

3.6 Phantom Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . 47

3.7 Mechanical Testing . . . . . . . . . . . . . . . . . . . . . . . . . . 48

4 Quantitative Vascular Elastography 49

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

4.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

4.3 Simulation Study . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

4.4 Phantom Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

4.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

4.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

5 The Effects of Microvessels on Quantitative Elastography 73

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

5.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

5.3 Simulation Study . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

5.4 Phantom Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

 xii

5.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

5.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

5.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

6 Material Nonlinearity in Vascular Elastography 99

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

6.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

6.3 Simulation Study . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

6.4 Phantom Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

6.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

6.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

6.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

7 In Vivo Quantitative Elastography for Routine Screening 126

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

7.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

7.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

7.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

7.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

8 Conclusion and Future Work 144

8.1 Summary of Results . . . . . . . . . . . . . . . . . . . . . . . . . 144

8.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

Bibliography 149
 xiii

List of Tables

4.1 Composition of materials used to fabricate vessel phantoms . . . . 57

4.2 Mean and standard deviation of Youngs modulus of the plaques

and vessel walls. These values were estimated from the recovered
elastograms recovered with the three imaging approaches. We also
report the values from independent mechanical testing of represen-
tative samples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

7.1 Normalized Mean strain, regularization parameter, RMSD, EYM

and maximum principal Cauchy stress values for each volunteer and
acquisition/reconstruction combination . . . . . . . . . . . . . . . 140
 xiv

List of Figures

1.1 Atherosclerotic plaque progression and rupture in the coronary

artery adapted from Libby et al. . . . . . . . . . . . . . . . . . . . 2

1.2 Vascular angiography image of an artery with stenosis [Akins et al.,

1998] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.3 Computed tomography image of an artery with stenosis [Vale et al.,

2012] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.4 Magnetic resonance imaging scan of artery with stenosis [Corti and
Fuster, 2011] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.5 Doppler ultrasound image of artery with stenosis [Gaitini and

Soudack, 2005] . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.6 Intravascular ultrasound image with virtual histology highlighting

an atherosclerotic plaque [Calvert et al., 2011] . . . . . . . . . . . 9

1.7 Intracoronary optical coherence tomography of a vessel with atheroscle-

rotic plaque [Prati et al., 2010] . . . . . . . . . . . . . . . . . . . . 10

2.1 a: An object modeled using plane stress that is thick in the z

direction. b: An object modeled using plane strain that is thin in
the z direction. Image adapted from [Carpinteri, 2002]. . . . . . . 16

2.2 Linear and hyperelastic stress-stretch curves, with = 2(K1 + K2 ) 20

 xv

2.3 Measured stress-stretch curves for different vascular tissues in the

circumferential and longitudinal directions [Holzapfel et al., 2000] 22

2.4 a & b: Measured axial and lateral displacement images, respec-

tively, obtained from cross-correlation speckle tracking from two
simulated RF images using linear array transducers. c & d: Ra-
dial and circumferential strain elastograms, respectively, computed
from the displacements in a & b. Adapted from [Korukonda and
Doyley, 2011b]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.5 (a) b-mode and (b) sonoleastography images of a thermal lesion

within a porcine liver [Parker et al., 2011]. . . . . . . . . . . . . . 26

2.6 Reconstructed stress elastogram in a simulated phantom with two

inclusions using a genetic algorithm approach [Karimi et al., 2008]. 32

2.7 Reconstructed modulus elastograms from simulated IVUS im-

ages a) with and b) without spatial information. Adapted from
[Richards and Doyley, 2011]. . . . . . . . . . . . . . . . . . . . . . 35

2.8 Reconstructed and ground truth shear modulus elastograms recon-

structed using the MOLS method on noise-free simulated displace-
ments. Adapted from [Jadamba et al., 2014]. . . . . . . . . . . . . 36

3.1 Sample finite element mesh created in Comsol using 1124 CST el-
ements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.2 Diagram of cross-correlation displacement tracking algorithm: The

blue boxes are the pre- and post-compression images (RF1 and
RF2). The orange box represents the reference window, the red
box represents the search window, and the shaded lavender boxes
represent the cross-correlation map within the search window. . . 42

3.3 Diagram of reconstruction flow . . . . . . . . . . . . . . . . . . . 45

 xvi

3.4 Sample PVAc phantom, where the soft inclusion is gray due to the
addition of silicon carbide (SiC) . . . . . . . . . . . . . . . . . . . 47

4.1 Sonograms obtained using (a) conventional, (b) sparse-array, and

(c) plane-wave imaging from simulated vessel phantoms with atten-
uation coefficients of 0 dB/cm, 1.5 dB/cm, and 3.5 dB/cm going
from left to right. . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

4.2 (a) Axial and (b) lateral profiles taken through the center of point-
spread-functions (PSF) for conventional, sparse-array, and plane-
wave imaging system. . . . . . . . . . . . . . . . . . . . . . . . . . 60

4.3 Montage of radial strain elastograms computed from RF echo

frames acquired with (a) conventional linear array, (b) sparse-array,
and (c) plane-wave imaging. Ideal strain elastograms are shown in
the first column of each montage, and the elastograms obtained
from vessel phantoms with attenuation coefficients of 0 dB/cm, 1.5
dB/cm, and 3.5 dB/cm are shown in the remain columns. . . . . . 61

4.4 Montage of circumferential strain elastograms computed from RF

echo frames acquired with (a) conventional linear array, (b) sparse-
array, and (c) plane-wave imaging. Ideal circumferential strain
elastograms are shown in the first column of each montage, and
the elastograms obtained from vessel phantoms with attenuation
coefficients of 0.6 dB/cm, 1.5 dB/cm, and 3.5 dB/cm are shown in
the remain columns. The two rectangular box in (a) denotes the
regions of interests corresponding to the plaque and vessel wall that
was used to compute CN Re values reported in Fig. 5.7. . . . . . . 62
 xvii

4.5 The normalized root-mean-error incurred when estimating (a) axial

and (b) lateral displacements with conventional linear array (first
column), sparse-array (second column), and plane-wave (third col-
umn) imaging. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

4.6 Montage of modulus elastograms corresponding to the strain elas-

tograms shown in Figs 3 and 4. The first column of each mon-
tage shows the actual modulus elastograms and the remain columns
shows the elastograms obtained from vessel phantoms with attenua-
tion coefficients of 0.6 dB/cm, 1.5 dB/cm, 3.5 dB/cm. Showing (a)
conventional, (b) sparse-array, and (c) plane-wave modulus elas-
tograms. The two rectangular box in (a) denotes the regions of
interests corresponding to the plaque and vessel wall that was used
to compute CN Re values reported in Fig. 4.7. . . . . . . . . . . . 64

4.7 (a) Recovered modulus contrast and (b) modulus error incurred
when elastography imaging was performed with conventional linear
array (left-column), sparse-array (middle-column) and plane-wave
(right-column) imaging. . . . . . . . . . . . . . . . . . . . . . . . 65

4.8 The contrast-to-noise ratio of strain (top and middle rows) and
modulus (bottom rows) elastograms obtained from vessel phan-
toms modulus contrast of -2.5 dB, -6.02 dB, and -12.04 dB; and
attenuation coefficients of 0.6 dB/cm to 3.5 dB/cm. . . . . . . . . 66

4.9 Sonograms and elastograms obtained from vessel phantoms #1 (a)

and #2 (b) when elastographic imaging was performed with CLA,
SA, and PW imaging. Showing (i& iv) Sonograms, (ii & v) cross-
correlation maps, and (iii & vi) modulus elastogram. In the sono-
gram, the crescent shape ROI denotes the plaque. . . . . . . . . . 67
 xviii

4.10 (left) Contrast recovery error, and (right) contrast-to-noise ratio

of modulus obtained with CLAI, SAI, PW. Showing experimental
results obtained from vessel phantoms with attenuation coefficient
0.6 dB/cm (phantom # 1), and 3.5 dB/cm (phantom #2). . . . . 69

5.1 (a) A schematic of the CE-IVUS system used in the experimental

study; (b) The spectral power of the IVUS transducer; (c) the peak
negative pressure of the IVUS transducer at depths of 2.25 and 3.5
mm; (d) the number weighted and volume weighted microbubble
distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

5.2 (a) The geometry of the simulated vessel phantom; (b) The ground
truth radial strain elastogram; (c) The measured radial strain elas-
togram from FIELD II simulations; (d) The ground truth circum-
ferential strain elastogram; (e) The measured circumferential strain
elastogram from FIELD II simulations. All strains are given as per-
centages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

5.3 (a,b) The contrast images of the simulated phantom using FIELD
II. Microvessel CTRs range from 6.7-16.5dB. The manual segmen-
tation is overlayed in white; (c-e) Youngs modulus elastograms
for the simulated phantom with 10 Pa microvessels (Pa) for the
ground truth, reconstructed elastogram with microvessel geometry
(mv) and reconstructed elastogram without microvessel geometry
(no mv); (f-h) The maximum principal stress elastograms for simu-
lated phantom with 10 Pa microvessels (Pa); (i-k) Close up images
of maximum principal stresses in the plaque cap (Pa) correspond-
ing to the box in Fig. 5.3 f. The paths between the two sets of
arrows in Fig. 5.3 g define the profiles in Fig. 5.4 . . . . . . . . . 84
 xix

5.4 (a,c,e) The radial strain, circumferential strain and principal stress
profiles through the left path depicted in Fig. 5.3 (b,d,f); The radial
strain, circumferential strain and principal stress profiles through
the right path passing through the simulated plaque. Plots a-d
show the ground truth, computed strain with microvessel geometry,
computed strain without microvessel geometry, and the measured
strain. Plots e-f show the ground truth, computed stress with mi-
crovessel geometry, and the computed stress without microvessel
geometry. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

5.5 The quantitative modulus and stress results for the elastographic
data in the simulation study vs. microvessel Youngs modulus.
(a-c) Mean Youngs modulus in the vessel wall, plaque, and cap
regions for both reconstruction types; (d) Mean maximum princi-
pal strain in the plaque cap for both reconstruction types; (e-h)
Reconstruction errors corresponding to plots (a-d). Units are in kPa. 88

5.6 (a-b) The fundamental and ultraharmonic responses for a phantom

with a microvessel 3.5 mm from the center of the lumen; (c-d) The
fundamental and ultraharmonic responses for a phantom with a
microvessel 2.2 mm from the catheter. . . . . . . . . . . . . . . . 89

5.7 The measured radial and circumferential strains for phantom 1

(a,b); The reconstructed Youngs modulus elastograms with and
without microvessel geometry for phantom 1 (kPa) (c,d); The re-
constructed maximum principal stress elastograms with and with-
out microvessel geometry for phantom 1 (kPa) (e,f). The arrows in
c define the profile shown in Fig. 5.9. . . . . . . . . . . . . . . . . 90
 xx

5.8 The measured radial and circumferential strains for phantom 2

(a,b); The reconstructed Youngs modulus elastograms with and
without microvessel geometry for phantom 2 (kPa) (c,d); The re-
constructed maximum principal stress elastograms with and with-
out microvessel geometry for phantom 2 (kPa) (e,f). The arrows in
c define the profile shown in Fig. 5.9. . . . . . . . . . . . . . . . . 92

5.9 (a,c,e) The radial strain, circumferential strain and principal stress
profiles through the path depicted in Fig. 7; (b,d,f) The radial
strain, circumferential strain and principal stress profiles through
the path depicted in Fig. 8. Plots a-d show the computed strain
with microvessel geometry, computed strain without microvessel
geometry, and the measured strain. Plots e-f show the computed
stress with microvessel geometry, and the computed stress without
microvessel geometry. . . . . . . . . . . . . . . . . . . . . . . . . . 93

5.10 The quantitative results for the elastographic data in phantoms 1

and 2, respectively. Error bars represent the standard deviation
between experiments. Units are in kPa. . . . . . . . . . . . . . . . 94

6.1 Graphical representation of a Newtons method inversion [De Borst

et al., 2012] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

6.2 Ground truth and reconstructed hyperelastic K-moduli elastograms

compared against four times the reconstructed linear Youngs mod-
ulus elastograms. Reconstructions shown for intraluminal pressures
between 8-56 mmHg . . . . . . . . . . . . . . . . . . . . . . . . . 110

6.3 Maximum principal Cauchy stress elastograms for the ground truth,
hyperelastic reconstructions, and linear reconstructions. These
elastograms are shown for intraluminal pressures of 8-56 mmHg . 111
 xxi

6.4 Enhanced maximum principal Cauchy stress elastograms for the

ground truth, hyperelastic reconstructions, and linear reconstruc-
tions. The displayed regions are defined by the black outline shown
in Fig. 6.3. These elastograms are shown for intraluminal pressures
of 8-56 mmHg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

6.5 EYM elastograms for the ground truth, hyperelastic reconstruc-

tions, and linear reconstructions. These elastograms are shown for
intraluminal pressures of 8-56 mmHg . . . . . . . . . . . . . . . . 113

6.6 Enhanced EYM elastograms for the ground truth, hyperelastic re-
constructions, and linear reconstructions showing the region defined
by the gray rectangle in Fig. 6.6. These elastograms are shown for
intraluminal pressures of 8-56 mmHg . . . . . . . . . . . . . . . . 114

6.7 Figure 6.7 a & b: Reconstruction results for the vessel wall and
inclusion regions, respectively. The ground truth and hyperelastic
reconstruction show the K-modulus, while the linear reconstruction
shows 14 E. Figure 6.7 c & d: These plots show the errors between
the reconstructions in Fig. 6.7 a & b and the ground truth values. 115

6.8 Figure 6.8 a & b: Reconstruction results for the EYM and Maxi-
mum principal Cauchy stress, respectively within the crescent be-
tween the lumen and the plaque as a function of intraluminal pres-
sure. Results are shown for the ground truth, hyperelastic recon-
struction, and linear reconstruction. Figure 6.7 c & d: These plots
show the errors between the reconstructions in Fig. 6.7 a & b and
the ground truth values. . . . . . . . . . . . . . . . . . . . . . . . 116

6.9 Stretch/stress curves for the vessel wall (a) and plaque (b) regions
measured by uniaxial compression testing. The gray lines show
individual trials, while the dotted red line shows the fitted curve. . 117
 xxii

6.10 Hyperelastic and linear reconstructed elastograms for the PVAc

phantom. The top row shows the K-modulus elastogram, the mid-
dle row shows the EYM, and the bottom row shows the maximum
principal stress for a pressure of 50 mmHg. . . . . . . . . . . . . . 119

6.11 a: Measured and reconstructed K-modulus values for the PVAc

phantom. b: Reconstructed EYM and stress values using the hy-
perelastic and linear algorithms. These values were computed by
averaging within the region defined by the black rectangle in Fig.
6.10. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

7.1 Error between the true EYM and computed value using the as-
sumption of a thick-walled pressure vessel. All units are percentages.131

7.2 B-scans for each of the four volunteers using SA and PW acquisition
displayed on a logarithmic scale . . . . . . . . . . . . . . . . . . . 133

7.3 Measured axial and lateral displacements for SA and PW for all
volunteers (m) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

7.4 Cross-correlation coefficients for each volunteer for SA and PW.

Results are averaged over all displacement pairs. . . . . . . . . . . 135

7.5 Box-and-whisker plots for the data in Fig. 7.4 . . . . . . . . . . . 135

7.6 RMSD between the measured and computed nodal displacements

as a function of solve iterations for all acquisition/reconstruction
pairs. Each line represents a different regularization value. The
white circles indicate the chosen regularization parameters and it-
eration numbers. . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

7.7 Relative EYM reconstructions for PWL, PWNL, and SAL acquisi-
tion/reconstruction pairs for all volunteers . . . . . . . . . . . . . 137
 xxiii

7.8 Relative maximum principal Cauchy stress reconstructions for

PWL, PWNL, and SAL acquisition/reconstruction pairs for all vol-
unteers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

7.9 Volunteer age verses the mean EYM for all acquisition/reconstruction
pairs. The error bars represent the standard deviation of the EYM
within each vessel. . . . . . . . . . . . . . . . . . . . . . . . . . . 139
 1

1 General Introduction

1.1 Pathophysiology of Plaques

Every 40 seconds, someone in the United States dies from cardiovascular dis-
ease, with many of these deaths occurring from atherosclerotic plaque rupture
[Go et al., 2014; Blacher et al., 1998]. Atherosclerosis may be aggravated by sev-
eral factors, including smoking and a high cholesterol diet [Hjermann et al., 1981;
De Backer et al., 2003]. The accumulation of lipids in the sub-endothelial in-
tima causes atherosclerotic plaques to form, which is accompanied by the pro-
duction of inflammatory biomarkers such as C-reactive proteins [Libby et al.,
2002]. This leads to an accumulation of collagen and elastin, which forms a
stiff fibrous cap encasing the lipid-rich plaque [Libby, 1995]. Inflammation pro-
duces macrophages within the plaque, which release proteolytic enzymes that
erode the fibrous cap [Virmani et al., 2000]. Further development of the plaque
occurs when additional phospholipids enter the plaque through microvessels, in-
cluding the vasa vasorum [Virmani et al., 2005]. All of these occurrences weaken
the fibrous cap, and accelerate the cycle of plaque buildup [Kwon et al., 1998;
Moulton et al., 1999]. If the forces exerted by the pulsatile blood pres-
sure and hemodynamic factors exceed the mechanical yield criterion of the fi-
brous cap, the cap will rupture and release the thrombogenic lipid core into
 2

the artery, triggering clots and restricting blood flow [Naghavi et al., 2003;
Blacher et al., 1998]. Clots in the carotid artery may cause stroke, while clots
in the coronary artery may cause heart attacks. This process of plaque buildup is
shown in Fig. 1.1.

Figure 1.1: Atherosclerotic plaque progression and rupture in the coronary artery
adapted from Libby et al. [Libby et al., 2002]

Atherosclerotic plaques are classified as stable or unstable based on their

likelihood to rupture. Stable plaques within the coronary artery have thick fi-
brous caps, which increase the pressure necessary for rupture. Unstable plaques
have large necrotic cores and thin fibrous caps (< 65m) [Naghavi et al., 2003;
Virmani et al., 2003]. It is estimated that plaques rupture when the external
stresses exceed 300 kPa [Cheng et al., 1993; Loree et al., 1992; Holzapfel et al.,
2005]. Identifying patients with rupture-prone plaques is difficult, because the
 3

progression of plaque buildup is largely asymptomatic [Furberg et al., 1994]. Vas-

cular angiography is commonly used to assess the likelihood of plaque rupture
by measuring the degree of stenosis, or vessel narrowing [Fayad et al., 2002;
Nieman et al., 2002]. However, evidence shows that stenosis is a poor indica-
tor for whether a plaque will rupture [Fayad et al., 2002]. Arterial remodeling
may lead to large stresses, without significant narrowing [Glagov et al., 1987].
To predict the propensity of rupture, the mechanical properties or composition
of the plaque must be known. The principal motivation of this research is
to develop methods for quantifying the mechanical properties and the
stress distribution within vascular tissues.

1.2 Imaging Techniques for Atherosclerosis and

Plaque Vulnerability

1.2.1 Vascular Angiography

Vascular angiography is a technique that uses X-ray imaging to view the constric-
tion of arteries. The arteries are differentiated from the surrounding tissue using
a contrast agent injected directly into the circulatory system. This contrast agent
appears opaque to the imaging machine. Stenosis within a vessel causes visible
narrowing of the blood flow, which can be measured to assess vulnerability [Little
et al., 1988].
 4

Figure 1.2: Vascular angiography image of an artery with stenosis [Akins et al., 1998]

1.2.2 Computed Tomography

Computed tomography (CT) uses X-ray response to assess atherosclerosis. CT

scans are non-invasive procedures which can produce high-resolution images of the
human body in three dimensions. Using a multi-slice approach, CT imaging can
achieve a resolution of 400 m in coronary imaging [Dettmer et al., 2013]. This
allows clinicians to view plaques and calcifications in addition to stenosis deep
within the body. The main strength of CT is its ability to identify calcification,
which may increase localized stresses within the artery, decreasing its stability
[Owen et al., 2011]. However, abdominal CT scans produce 600 times the ionizing
radiation of a chest x-ray, and may increase the risk of cancer after repeated scans
[Hausleiter et al., 2009]. Additionally, the specificity of CT may be hindered by
the varying levels of attenuation between plaque components [Owen et al., 2011].
 5

Figure 1.3: Computed tomography image of an artery with stenosis [Vale et al., 2012]

1.2.3 Magnetic Resonance Imaging

Unlike CT, Magnetic resonance imaging (MRI) is not based on ionizing radiation.
It also produces high contrast images between plaque features. High-resolution
MRI can recover the composition of plaques as well as stenosis [Owen et al.,
2011]. Accurate composition information could be used to model stresses within
the arteries. However, MRI has relatively poor resolution compared to CT (0.55
cm3 voxel) [Zhao et al., 2009; Saam et al., 2009; Jiang et al., 2011], and its high
operating costs make it an impractical screening tool.

Figure 1.4: Magnetic resonance imaging scan of artery with stenosis [Corti and Fuster,
2011]
 6

1.2.4 Non-invasive Ultrasound

Non-invasive ultrasound is an attractive method for routine screening due to its

low cost and portability. Standard clinical ultrasound methods can be used to non-
invasively view atherosclerotic plaques in superficial vessels, such as the carotid
artery. This method typically uses array transducers with center frequencies be-
tween 2 and 10 MHz, which allows for imaging resolutions between 0.5 and 0.8 mm
along the imaging axis. The lateral resolution is significantly poorer than the axial
resolution and is directly correlated with the number of transducer elements in the
array. Typical clinical transducers have between 64-128 elements. Beamforming
methods have been proposed in improve the lateral resolution [Korukonda et al.,
2013]. However, it is still insufficient to observe micron-scale features, such as the
plaque cap, or microcalcifications. However, cross-sectional imaging of arteries can
measure the area of plaques within a slice [Herder et al., 2012]. Imaging in the
longitudinal direction can show the thickness of the vessel as well as the degree of
stenosis along the artery. Measuring the intima-media thickness (IMT) of vessels
can monitor atherosclerotic progression [Herder et al., 2012; Inaba et al., 2012;
Hodis et al., 1998]. Prior research has shown that the IMT increases with plaque
vulnerability.

Additionally, Doppler ultrasound can be used to monitor pathological changes

in blood flow within superficial vessels. Large changes in velocity can be correlated
with visual stenosis to better quantify narrowing [Skjaerpe et al., 1985]. The
pulse wave velocity (PWV) can be used to measure the aortic compliance, which
correlates with increased risk of plaque rupture [Lehmann et al., 1998]. However,
none of these techniques can be used to assess stress accumulation within the
artery, and they are all subject to inter-observer variability.

Non-invasive vascular elastography (NIVE) is a recently proposed method to

measure relative stiffness within the carotid artery using clinical ultrasound sys-
 7

tems [Korukonda and Doyley, 2011b; Korukonda et al., 2013]. Vascular elastog-
raphy can use displacements caused by the natural expansion and contraction
within the artery to compute mechanical strain. Plaques appear as regions of
high strain. However, this technique will not measure the absolute stiffness, and
thus cannot be used to compute stresses.

Figure 1.5: Doppler ultrasound image of artery with stenosis [Gaitini and Soudack,
2005]

1.2.5 Intravascular Ultrasound

Intravascular ultrasound (IVUS) is used to image both superficial and coronary

arteries. It uses higher frequency transducers than non-invasive ultrasound be-
cause there is less tissue to travel through, and attenuation is less of a concern.
Typical frequencies for coronary imaging range between 20-45 MHz. This allows
for imaging resolutions between 34 and 77 m. IVUS images can be used to
measure stenosis. These measurements can be more accurate than those obtained
using vascular angiography because they can accurately show plaque area as well
as narrowing [Nissen and Yock, 2001; Moselewski et al., 2004].

IVUS virtual histology is a proposed method to determine the composition of

atherosclerotic plaques, including the lipid core, calcification, and the fibrous cap
[Moore et al., 1998; Diethrich et al., 2007; Nair et al., 2001]. Each of these features
 8

exhibits a unique spectral response, and these responses can be correlated with
those received by the transducer. In theory, this method can be used to obtain
real-time composition information. However, at this time, it is unclear whether
this method can achieve constant results using commercially available equipment
[Katouzian et al., 2008].

As with non-invasive ultrasound, IVUS may be used to measure the stiffness

within arteries using either elastography or palpography by computing strain.
Elastographic methods rely on natural vessel displacements to compute strain,
while palpography induces displacement from RF radiation force [Doyley et al.,
2001; De Korte et al., 1998].

Finally, IVUS can be used in molecular imaging to visualize microvessels such

as the vasa vasorum [Carlier et al., 2005; Staub et al., 2010]. Increased vasa va-
sorum density allows greater amounts of macrophages to enter into the vessel and
feed the plaque. In this modality, subjects may be injected with an ultrasound
contrast agent, consisting of lipid-based microbubbles. These microbubbles ex-
hibit nonlinear responses when subjected to RF radiation. The linear response of
tissue can be suppressed using signal processing techniques such as pulse inver-
sion [Shekhar and Doyley, 2013]. Coronary vasa vasorum have diameters ranging
between 67 and 160 microns [Kwon et al., 1998], which extend significantly be-
low the imaging resolution. Thus, this method relies on imaging clusters of vasa
vasorum which may show up as single large microvessels.
 9

Figure 1.6: Intravascular ultrasound image with virtual histology highlighting an

atherosclerotic plaque [Calvert et al., 2011]

1.2.6 Intracoronary Optical Coherence Tomography

Intracoronary optical coherence tomography (OCT) is an imaging method that

measures the optical scattering of near-infrared radiation emitted from a swept-
frequency laser with wavelengths between 1,250 and 1,350 nm. Similarly to IVUS,
this laser is placed on the end of a rotating catheter for use in the coronary artery.
It has a significantly higher resolution than IVUS due to the higher frequency
of light over sound. This resolution allows OCT to image the plaque cap and
microcalcifications, as well as stenosis [Kubo et al., 2007; Maehara et al., 2009;
Honda and Fitzgerald, 2008]. However, intracoronary OCT has drawbacks. The
high attenuation of light only allows for imaging depths of 3 m, not large enough
to view an entire plaque. Significant scattering is also incurred from the red blood
cells [Maehara et al., 2009; Honda and Fitzgerald, 2008].
 10

Figure 1.7: Intracoronary optical coherence tomography of a vessel with atherosclerotic

plaque [Prati et al., 2010]

1.3 Thesis Objectives

The goal of this thesis is to develop a technique to measure the stresses and Youngs
modulus elastograms within atherosclerotic plaques to assess the propensity of
atherosclerotic plaques to rupture. To achieve this, the following objectives had
to be satisfied.

Aim 1: Assess the feasibility of reconstructing modulus elas-

tograms of the carotid artery using sparse array ultrasound imag-
ing

In this aim, the effect of tissue attenuation on the performance of model-based

elastography was assessed using linear array ultrasound. Modulus elastograms
 11

were reconstructed with three acquisition approaches. Geometric constraints were

imposed to penalize inhomogeneities in the inverse problem. Simulated ultrasound
RF frames were collected for each imaging approach with varying levels of attenu-
ation and stiffness contrasts between the inclusion and background regions. These
data were verified through two experimental phantom reconstructions with differ-
ent levels of attenuation.

Aim 2: Assess the effects of neovascularization on stress recon-

struction

This aim determines the effect microvessels have on stress in the coronary
artery, and the ability to reconstruct these stresses both with and without knowl-
edge of the microvessel locations. A simulation study was conducted using a
phantom with wall, plaque, and fibrous cap regions. Six microvessels were located
within the plaque and the vessel wall. Simulated RF intravascular ultrasound
(IVUS) images were then computed. The results of the simulation study were
corroborated using two phantoms with microvessels in the plaque and wall regions
respectively. The microvessel locations were found using ultraharmonic contrast
imaging with the same IVUS system as used in the elastographic data collection.

Aim 3: Assess the effects of material nonlinearity on the perfor-

mance of modulus and stress imaging

In this aim, methods were developed to estimate the peak stress in vascular
tissues. More specifically, a finite element framework was developed to model and
reconstruct the nonlinear, hyperelastic behavior of vascular tissues based on an
exponential strain energy density function. A simulation study was completed
to reconstruct the mechanical properties from a vessel undergoing 35% strain.
This involved accumulating multiple sets of tracked displacements. These recon-
structions were performed using both linear and nonlinear mechanical models to
 12

determine the error incurred by assuming linearity under large strain conditions.
Phantom experiments were completed using IVUS, and a similar range of strains
as the simulation study. Intraluminal pressure was varied continuously to acquire
a set of frames encompassing the desired range of strains.

Aim 4: Assess the performance of in vivo modulus and stress

recovery as a screening technique

In this aim, the performance of the techniques developed in Aim #1 and

Aim #3 were tested within a clinical setting. More specifically, quantitative non-
invasive vascular imaging was performed on 4 healthy volunteers. We assessed
the quality of modulus and stress recovery using multiple acquisition methods
and reconstruction models.

1.4 Thesis Organization

The remainder of the thesis is divided into seven chapters:

Chapter 2 reviews the theory of linear and nonlinear tissue models within
the context of finite element modeling, and the proposed methods for recov-
ering the material properties of tissue using elastography. In this section,
both iterative and direct methods of quantitative elastography are examined.

Chapter 3 lays out the general methods used for data acquisition and
generation, computational algorithms used, and data analysis.

Chapter 4 introduces the soft-priors modulus reconstruction method for

non-invasive vascular elastography. This chapter reports the results of
Youngs modulus reconstruction using sparse array, plane wave, and con-
ventional acquisition methods. It examines the effects of modulus contrast
and attenuation on the accuracy of reconstruction.
 13

Chapter 5 examines the effects of microvessels on Youngs modulus and

principal Cauchy stress reconstructions using IVUS acquisition. Ultrahar-
monic contrast imaging was used to detect microvessels within simulated
and fabricated phantoms. The accuracy of the elastograms was compared
both with and without microvessel geometry as a priori information.

Chapter 6 extends the soft-priors reconstruction method to include hyper-

elastic mechanical models. This enhanced algorithm was used to reconstruct
simulated and fabricated phantoms with varying material parameters. Re-
sults are reported on the fit of hyperelastic models, the effect of increasing
strain on the reconstruction accuracy, and the accuracy and location of peak
Cauchy stresses within the vessels.

Chapter 7 reports the results of Youngs modulus and Cauchy stress recon-
structions of the carotid artery in vivo from four healthy volunteers. Data
for these results were acquired using non-invasive ultrasound.

Chapter 8 summarizes the results included in this thesis and discusses its
impact on the prediction of atherosclerotic plaque rupture and the potential
for clinical applications. Future challenges in this research are discussed and
suggestions are made to extend this work.
 14

2 Background

The elasticity of an object defines its stiffness and ease of deformation. It has
long been known that heterogeneities within human tissue may be indicators of
various diseases, such as cancer. Manual palpography is the most basic method to
detect these variations through toucha common example of this is locating breast
masses [Wellman et al., 1999; Wellman et al., 2001]. However, this technique can
only be used for superficial tissue, and cannot produce quantitative results. Tissue
elastography is a technique that was developed in the 1990s that quantitatively
characterizes tissue stiffness by measuring displacements between pairs of medical
images such as ultrasound or MRI [Ophir et al., 1991]. Such techniques can create
maps of mechanical strain within tissue, which are known as elastograms. The
observed strains correlate with the exact mechanical properties of the object. To
compute these properties, one must know the underlying mechanics of measured
material, often described through a constitutive model. This chapter examines
the various constitutive models that have been used in vascular elastography, as
well as techniques that have been used to measure and quantify these mechanical
properties.
 15

2.1 Review of Vascular Elasticity

2.1.1 Linear Elasticity

Linear elasticity describes a simple physical relationship between stress and strain
in an object. This relationship dictates that, at any location, stress must increase
linearly with strain. The stress-strain relationship is shown below:

= D, (2.1)

where  is the engineering strain tensor, and represents the Cauchy stress tensor
[Cook et al., 2007]. The matrix, D, is the linear elastic constitutive matrix. When
describing a planer slice of a three dimensional material, an assumption of plane
stress or plane strain must be made. Plane stress assumes that the out-of-plane
stress is equal to zero. This approximation is often used to model geometries that
are thin in the out-of-plane direction. The matrix D for this relation is

1 0
E
D= 0 , (2.2)

1
1 2
1
0 0 2

where E is the Youngs modulus and is the Poissons ratio [Cook et al., 2007].
Plane strain assumes that strains in the out-of-plane direction are equal to zero.
This assumption holds for geometries that are very long in the out-of-plane direc-
tion. In two dimensions, D is defined as [Cook et al., 2007];

1 0
E
D= 1 0 . (2.3)


(1 + ) (1 2)
12
0 0 2

A diagram comparing plane strain and plane stress is shown in Fig. 2.1
 16

Figure 2.1: a: An object modeled using plane stress that is thick in the z direction. b:
An object modeled using plane strain that is thin in the z direction. Image adapted
from [Carpinteri, 2002].

Like most soft tissues, the stress-strain relationship in arteries is nonlinear.

Specifically, the perceived stiffness increases with strain. Thus, linearity can only
be assumed when the tissue is in the small strain regime. le Floch et al. assumes
this range to be under 2% [Le Floch et al., 2010]. Zhou and colleagues used the
degree of nonlinearity as a method to assess the error due to incorrect assumptions
of tissue linearity. They found that stress errors ranged between 5-30% within
the carotid artery under physiological loading conditions. The lower limit is not
 17

zero as the artery is assumed to experience non-zero strain in its rest state due to
longitudinal tension [Zhou and Fung, 1997]. To overcome these limitations, Bergel
computed an incremental elastic modulus, which was only valid for a specific
pressure [Bergel, 1961]. This allowed him to make the assumption of linearity
over any small range of strains. Bergel used ex vivo tissue samples to measure the
change in radius while varying the pressure within the artery. These values were
then used to compute the Youngs modulus using the equation for stiffness in a
thin-walled pressure vessel,

2p(1 2 )Ri2 Ro
E= , (2.4)
Ro (Ro2 Ri2 )

where p is the pressure on the inner lumen of the vessel, Ri is the inner radius,
Ro is the outer radius, and Ro is the change in outer radius. Patel et al. ex-
panded on this technique by using a vessel model that was able to account for
the inherent anisotropy in vessels, computing stiffness in the r, , and z directions
[Patel and Janicki, 1970]. These models were used to show that arterial stiffness
increases with both age and intraluminal pressure. However, studies using these
implementations were tedious to conduct and posed challenges when comparing
multiple measurements [Hayashi et al., 1980]. Moritake and colleagues took a
different approach by computing an average Youngs modulus over all measured
strains [Moritake et al., 1974]. Averaging the stiffness can provide some insight
as to the modulus at an arbitrary strain state, but in nonlinear materials, it can-
not accurately measure the stiffness at a specific strain. Hayashi et al. combined
these two techniques to extrapolate the shear modulus at all pressures by combin-
ing the average stiffness and incremental stiffness using a predefined relationship
[Hayashi et al., 1980]. This allowed for the extrapolation of stiffnesses outside of
the measured range. However, this relationship may not hold true for all subjects
and has not been compared against measured stiffness curves from the mechanical
testing of arteries. It is therefore a less desirable model than the above. Like the
 18

model developed by Hayashi et al., nonlinear hyperelastic stiffness models can

extrapolate stiffness outside of the measured range of strain. However, hyperelas-
tic modeling can be computationally intensive. Additionally, not all tissues have
the same degree of nonlinearity, so a poorly chosen model may not be any more
accurate the simple linear case. In the past two decades, nonlinear hyperelastic
models have become more widely used than linear ones as computational cost has
decreased and advancements have been made in the mechanical characterization
of soft tissue.

2.1.2 Hyperelastic Elasticity

Hyperelasticity encompasses a broad range of material relationships that can

be described by a nonlinear strain energy density function, [De Borst et al.,
2012]. These energy density functions are often expressed with respect to the
right Cauchy-Green deformation tensor, C. This is a measure of strain that is
valid over both small and large strain regimes, and ignores displacements due to
rigid body motion. The Cauchy-Green deformation tensor is defined by C =F T F,
where F is the stretch tensor, and T is the matrix transpose. The stretch in a
material is defined as the ratio between a deformed length and a reference length.

To define the strain energy density functions as coordinate independent rela-

tionships, they are often written as functions of the tensor invariants of C ; where
the n th tensor invariant is equal to the n th component of the characteristic poly-
nomial of C. For a three-dimensional tensor, the tensor invariants can be written
as [De Borst et al., 2012]
I1C = Cxx + Cyy + Czz

I2C = Cxx Cyy + Cyy Czz + Cxx Czz + Cxy

2 2
+ Cxz 2
+ Cyz (2.5)

I3C = Cxx Cyy Czz + 2Cxy Cyz Czx Cxx Cyz

2 2
Cyy Czx 2
Czz Cxy .
 19

These strain invariants can be further decomposed into hydrostatic and deviatoric
components, where the hydrostatic component represents strain due to volumetric
change, and the deviatoric component represents strain due to shape change. The
deviatoric components are often used in nearly incompressible materials, and are
defined below:
J1C = I1C J 2/3

J2C = I2C J 4/3 (2.6)

J3C = J,
p
where J is the Jacobian constant defined by J = det(C) [De Borst et al., 2012].

The first hyperelastic model, which was used to describe the behavior of
rubber-like materials, was proposed by Melvin Mooney in 1940 where the strain
energy density function was defined by a single stiffness parameter [Mooney, 1940].
This was later improved upon by Rivlin, creating the Mooney-Rivlin model by
adding an additional stiffness parameter [Rivlin, 1948]. This allowed the model
to be applied to a wider range of stress-strain relationships. The energy density
function for this model is

= K1 (J1 3) + K2 (J2 3) , (2.7)

where K 1 and K 2 are material parameters. Under small strain conditions, the
shear modulus, , is related to the variables K 1 and K 2 by

= 2(K1 + K2 ). (2.8)

An example stress-stretch plot is shown below comparing a Mooney-Rivlin mate-

rial with a linear material that can be assumed to have the same shear modulus
under small strain. The plot shows that as stretch increases, the Mooney-Rivlin
 20

model diverges from the linear curve.

4 K1 = 10000, K2 = 5000
x 10
2
Linear
MooneyRivlin

Stress 1.5

0.5

0
1 1.1 1.2 1.3 1.4
Stretch

Figure 2.2: Linear and hyperelastic stress-stretch curves, with = 2(K1 + K2 )

A more accurate representation of human arteries was proposed by Holzapfel

et al. which accounted for the anisotropic nature of arteries [Holzapfel et al.,
2005]. This model is shown below:

M1  M2 (1)(I1 3)2 +(I4 1)2 

= (I1 3) + e 1 , (2.9)
M2

where M 1 is a stiffness constant, M 2 and are dimensionless constants, and I 4 is

an invariant related to the angular anisotropy.

Mechanical testing of arteries is often conducted to determine the stiff-

ness and validate the imposed constitutive relation using ex vivo tissue sam-
ples cut into rectangular sections. This is most commonly performed using
tensile testing [Gundiah et al., 2007; Lally et al., 2004; Holzapfel et al., 2000;
Holzapfel and Weizsacker, 1998]. These experiments are performed by attaching
a tissue sample to a tensile testing device, such that the sample can be stretched
uniaxially or biaxially. Biaxial testing may be used when there is suspected
 21

anisotropy, and is necessary when using the model described in Equation 2.9.
These tests apply strain, while measuring the tensile forces. Lally et al. per-
formed uniaxial and biaxial testing on porcine arteries over multiple straining
cycles [Lally et al., 2004]. Comparing uniaxial and biaxial results, they noticed
distinct anisotropy. Specifically, they found that the vessels appeared stiffer in
the longitudinal direction, with the fibers, than in the circumferential direction.
In situ, arteries are pre-strained longitudinally. The degree of this stretching may
alter the stress-stretch relationship of the vessel [Lally et al., 2004]. In addition
to tensile testing, some researchers have used indentation testing to measure the
stiffness in the radial direction. Barret et al. applied this technique and acquired
similar curves to Lally et al. and Holzapfel et al. [Barrett et al., 2009]. Using
these results, estimates of different components of vascular tissue can be mea-
sured, including cellular, hypocellular, and calcified regions [Barrett et al., 2009].
However, there is significant variation between these measurements due to the
challenge of isolating each component. For example, the Youngs modulus for
calcified regions has been reported between 350 and 1500 kPa [Lee et al., 1992;
Loree et al., 1994] over small strain. Fibrous regions have been reported between
33 and 1500 kPa [Barrett et al., 2009; Ebenstein et al., 2002]. Figure 2.3 shows
measured stress-stretch curves for different vascular tissues conducted by Holzapfel
et al.. This plot highlights the anisotropy of arteries.
 22

Figure 2.3: Measured stress-stretch curves for different vascular tissues in the circum-
ferential and longitudinal directions [Holzapfel et al., 2000]

2.2 Review of Direct Elastography

2.2.1 Quasi-static Strain Elastography

Quasi-static elastography visualizes the strain distribution within soft tissues

by performing cross-correlation analysis on a pair of ultrasound images ob-
tained before and after a small strain excitation is applied. This technique
was first proposed by Ophir et al. [Ophir et al., 1991]. A common method
for measuring displacement is through cross-correlation based speckle tracking
[Lubinski et al., 1996; Konofagou and Ophir, 1998; Chaturvedi et al., 1998;
Korukonda and Doyley, 2011b]. This technique performs cross-correlation anal-
ysis between equally-sized regions in both images. The region in one image is
sequentially shifted to find the best spatial match with the other. This technique
is simple, and easily automated, however it is only valid for small strains with
limited deformity between images. Varghese et al. showed that the accuracy of
 23

this method degrades steeply after 10% strain [Varghese and Ophir, 1997]. Fig-
ure 2.4 shows displacement and strain elastograms obtained by Korukonda et al.
from a simulated vessel phantom using cross-correlation displacement tracking
[Korukonda and Doyley, 2011b].

Figure 2.4: a & b: Measured axial and lateral displacement images, respectively, ob-
tained from cross-correlation speckle tracking from two simulated RF images using linear
array transducers. c & d: Radial and circumferential strain elastograms, respectively,
computed from the displacements in a & b. Adapted from [Korukonda and Doyley,
2011b].

Displacement tracking using image registration is more adept at recovering

large deformations because it can account for shape change between images. This
technique has also been used in elastography [Liang et al., 2008; Richards and
Doyley, 2013]. These methods distort one image to satisfy a cost function of
the mean square difference between the images. This is carried out through an
 24

iterative minimization algorithm. However, registration methods often require an

initial guess based on cross-correlation tracking to converge on the correct solution
and choosing a suitable regularization parameter is a non-trivial process [Richards
and Doyley, 2013].

Ophir and colleagues showed that strain within tissue is directly proportional
to the Youngs modulus, E, when one assumes linearity, orthotropy, and isotropy,
which allows strain to be used as a surrogate for stiffness [Ophir et al., 1991].
Under uniaxial compression or expansion, this relationship can be expressed as
follows:
11
E= (2.10)
11
22
= ,
11
where is the Poissons ratio; 11 and 11 are the the components of the Cauchy
stress and engineering strain tensors in the direction of compression, respectively;
22 is the strain orthogonal to the compression. In homogeneous tissue, the stress
tensor can be approximated as the external force applied to the object. However,
this may be a poor assumption in heterogeneous tissue.

To apply this technique to a vessel, the stiffness can be approximated by

assuming it can be modeled as a thick-walled homogeneous pressure vessel. This
relationship in cylindrical coordinates is as follows [de Korte et al., 1997; Schwan
and FitzHugh, 1969]:
1
r = (r ( + z ))
E
1
 = ( (r + z )) (2.11)
E
1
z = (z (r + )) .
E
These equations show that strain is still inversely proportional to the Youngs mod-
ulus in this geometry. These methods of quasi-static strain-based elastography
 25

have been tested clinically for the detection and characterization of atheroscle-
rotic plaques [Brusseau et al., 2001; De Korte et al., 2002]. In addition, this
method has been used for diagnosing breast and prostate cancer [Cespedes, 1993;
Garra et al., 1997] and as a guide for minimally invasive surgeries [Kallel et al.,
1999; Varghese et al., 2003].

The equilibrium equations can be generalized through the following set of

partial differential equations (PDEs) when plane strain is assumed:

1 p + 21 (11 ) + 22 (12 ) = 0 (2.12)

1 p + 21 (12 ) + 22 (11 ) = 0,

where the partial derivatives are with respect to the body forces, and  represents
the two-dimensional strain tensor. When Equations 2.11 and 2.12 are combined,
one can derive the relationship between stress and vessel radius as follows:

pi r 2 r2
 
rr = 2 i 2 1 o2 , (2.13)
ro ri r

where ro is the outer radius, ri is the inner radius, r is the radius at any point
along the cylinder, and pi is the internal pressure [Hamrock et al., 1999]. This
shows that the stress (and corresponding strain) will degrade as the inverse square
of the radius. Thus, noise reduction becomes critical when measuring strain at
the outer edges of vessels [Maurice et al., 2005a].

2.2.2 Dynamic and Transient Modulus Elastography

Sonoelastography is a technique developed in the mid-1980s to measure the vis-

coelastic properties of muscle tissue [Mol and Breddels, 1982; Lerner et al., 1988;
Parker et al., 1990]. In this method of imaging, the tissue is excited by ultrasonic
waves with frequencies between 10-1000 Hz. This excitation creates compressive
 26

and shear waves through the medium. The compressive waves are highly atten-
uated due to the incompressibility of the tissue, so it is the shear waves that
are examined. These shear waves can be tracked using standard ultrasound ei-
ther with Doppler imaging or speckle tracking. Inhomogeneities within the tissue
change the speed of the waves [Levinson et al., 1995]. Figure 2.5 shows sample
(a) b-mode and (b) sonoleastography images of a thermal lesion within a porcine
liver [Parker et al., 2011].

Figure 2.5: (a) b-mode and (b) sonoleastography images of a thermal lesion within a
porcine liver [Parker et al., 2011].

A simple model for relating the wave frequency to the Youngs modulus is
to assume a linear, isotropic medium, where the longitudinal wave is attenuated
significantly, such that it can be ignored. In this case, the Youngs modulus relates
to the shear wave speed, c s , as follows [Hoyt et al., 2007],
s
E
cs = , (2.14)
2 (1 + )

where is the material density. In addition to muscle imaging, this method has
been used to approximate the shear modulus in the prostate [Taylor et al., 2000],
cervical lymph node metastases [Lyshchik et al., 2007], and breast [Krouskop
et al., 1998]. However, this method has been hindered by its susceptibility to
 27

noise and its assumption of linear elasticity.

An improvement to this model is to use a viscoelastic constitutive model. Hu-

man tissue has been previously shown to exhibit viscoelasticity [Fung, 1993]. That
is, the relationship between stress and strain varies over time. The wavelength
through the tissue can then be related to the mechanical properties using a model,
such as Kelvin-Voigt [Fung, 1993; Levinson et al., 1995] shown below,
   

Tjk =2 + jk + + mm jk , (2.15)
t t

where T is the component of the Cauchy stress tensor in the axial direction,
and are the Lame constants, and and are the constants of viscoelasticity
in the shear and longitudinal directions. This equation can then be re-written
and converted to the Fourier domain to relate stiffness with the wave frequency
as follows [Levinson et al., 1995];

2 U = ( + i) 2 U + (( + ) + i ( + )) U, (2.16)

where is the wave frequency and U is the displacement. This can be simplified
as
2 rot (U ) = ( + i) 2 rot (U ) , (2.17)

where rot(U ) represents the rotational component of displacement in the fre-

quency domain. However, the second derivatives involved in the 2 operation
magnify the effects of measurement noise. In Levinson et al. this equation was
used to determine the Youngs modulus of leg muscles at different applied loads
and different levels of flexing [Levinson et al., 1995]. Harmonic waves were intro-
duced in the leg using a vibrating plate. Shear waves were then measured using
Doppler ultrasound. Results showed a linear trend between increasing Youngs
modulus and the downward force on the leg. This agreed with their hypothesisa-
 28

tions. However, no spatial variance could be measured within each muscle due to
noise [Levinson et al., 1995].

Sinkus et al. developed a noise reduction method to use this technique for
breast imaging [Sinkus et al., 2000]. Their algorithm added a regularization
term in the matrix inversion to account for the measurement noise [Anykeyev
et al., 1991]. This led to clearer elastograms than those in Levinson et al. Elas-
tic constants were derived in three dimensions to test for anisotropy. Eigen-
decomposition was performed on the shear modulus tensors. If the tensor is
isotropic, each eigenvalue except for one should be zero. This held true for the
simulation; however anisotropy was detected in the in vivo study. In the breast tis-
sue, two components of the principal modulus tensor had similar magnitudes. This
anisotropy was attributed to vascularization. The group later improved these re-
sults by removing the transverse component of the wave using a Helmholtz-Hodge
decomposition [Sinkus et al., 2005].

2.3 Review of Model-Based Elastography

Model-based elastography minimizes the difference between a set of measured

displacements within the tissue with a set of computed displacements found using
a finite element (FE) model. The computed displacements are updated by varying
the stiffness parameters in the model. This minimization is defined by a functional,
such as the simple Gauss-Newton example shown below [Yalavarthy et al., 2007b]:

(E) = kum uc (E)k2 . (2.18)

In this case, u c are the computed displacements, which are a function of the
Youngs modulus, E, and um are the measured displacements. Two-dimensional
cross-sectional FE models of the artery are often used, assuming either linear-
 29

or hyper-elasticity. Care must be taken when defining the model. Doyley et al.
showed that the use of Dirichlet conditions on the boundaries will only allow
for the reconstruction of relative material properties [Doyley et al., 2000]. Thus,
pressure or traction boundary conditions are often used to constrain the model.
Additionally, this problem is ill-posed, meaning there can be multiple sets of
material properties which may generate the same set of displacement maps.

The functional is minimized by using a Taylor expansion, and setting its deriva-
tive with respect to the Youngs modulus to zero. This process will set up a mini-
mization algorithm requiring one or two derivatives of the displacement function.
Alternatively, stochastic or linear programming methods can be used, requiring
only the value of the functional itself.

2.3.1 Without Functional Derivatives

Doyley et al. developed an iterative method for computing the Youngs modulus
through multiple solves of Hookes law for plane stress as shown in the equation
below [Doyley et al., 1996]:

i i
11 22 33
E i+1 = , (2.19)
m
11

where E was updated by multiple computations of the stress tensor, , using the
finite element method. However, this method requires the calculation of the strain
inverse, which will magnify the effects of noise [Samani et al., 2001]. To reduce
noise, it is necessary to assume homogeneous modulus distributions within prede-
fined regions. Inaccuracy in the segmentation will also cause increased reconstruc-
tion error [Samani et al., 2001]. If these regions are not known, the data array can
be tessellated, solving for each tile independently. The specific tessellations can
be randomized at each iteration to minimize boundary errors [Samani et al., 2001;
Doyley, 2012]. This method was successfully shown to detect fibroglandular and
 30

adipose tissue in vivo using MRE.

It can be improved through the use of a more robust minimization algorithm.

One such method is a simplex minimization. This algorithm computes the min-
imum value of the functional by traversing the feasible region, which is divided
into a series of nodes forming a polytope. The solution is reached by traveling
along the edges until the minimizing vertex is reached. It was implemented by
Wittek et al. to reconstruct the hyperelastic properties of human aorta [Wittek
et al., 2013]. However, simplex algorithms are prone to converge on local minima.
Additionally this method may take hundreds of iterations to converge [Wittek
et al., 2013]. This is because the worst case convergence is exponential. Wittek et
al. concluded that it could be used to measure the stiffness of the aorta using 4D
ultrasound. However, extensive investigation was limited due to the computation
time of two weeks per reconstruction.

The interior point method uses a similar procedure to the simplex algorithm,
but converges linearly in the worst case, making it a more efficient algorithm. This
method has been used to reconstruct the mechanical properties of coronary arteries
[Baldewsing et al., 2006; Le Floch et al., 2009]. In these cases, the segmentation
of plaque regions is unknown and must be determined imperially. Le Floch et al.
used gradient images to create segmentation contours through a process they refer
to as the imaging modulography technique (iMOD). The contours themselves are
created using a dynamic watershed method. Baldewsing et al. solved for the
plaque contours as well as the stiffness by modeling the plaque boundaries as
Bezier curves defined by N degrees of freedom. This type of minimization would
be nearly impossible to implement in a gradient-based reconstruction due to the
challenge of computing the functional derivatives. Both groups demonstrated the
ability to detect and model atherosclerotic plaques in vivo, although the nature of
their segmentation strategies introduces significant variation in the reconstructed
modulus values within each region.
 31

Stochastic minimization methods are appealing due to their simplicity. These

techniques begin with a seed pool of potential Youngs modulus solutions. This
field is narrowed by successive matings, or combinations, of the fittest elastograms
[Goldberg and Deb, 1991]. Using this method, the computation time can be de-
creased by using a predefined seed pool, although this relies on each image using
the same geometry. Larger seed pools will improve the accuracy of the results by
reducing the likelihood of converging on local minima, although this increases the
computational cost [Zhang et al., 2006]. Khalil and colleagues used this technique
to reconstruct the Youngs modulus of atherosclerotic plaques in simulated arter-
ies, where the exact boundaries of the plaque are known [Khalil et al., 2005]. Their
technique was able to reconstruct the material properties from noise-free displace-
ment maps in five iterations with between 0.68 and 12.9 percent error. However,
the reconstruction algorithm degraded sharply with Gaussian noise, where the
error increased to above 30% with 8% added noise [Khalil et al., 2006]. Their
method was improved upon by Karimi et al. by introducing mutations to help
extend the solution space, and by using a hyperelastic material model [Karimi
et al., 2008]. This method was able to achieve under 10% error with 10% white
Gaussian noise. Figure 2.6 shows a stress elastograms reconstructed using this
method [Karimi et al., 2008].
 32

Figure 2.6: Reconstructed stress elastogram in a simulated phantom with two inclusions
using a genetic algorithm approach [Karimi et al., 2008].

However, their results assume perfect region segmentation, and ignore the lipid
pool reconstruction [Karimi et al., 2008]. This was justified by the claim that,
due to its concealment behind the stiff fibrous cap, the lipid pool was too poorly
conditioned to reconstruct accurately.

2.3.2 With Functional Derivatives

Derivative-based optimizations minimize a penalty functional by setting its deriva-

tive with respect to the minimization parameters to zero. These methods can reach
convergence faster than when the derivatives are not used, and are less likely to
converge on local minima. However, for these methods to work, the derivatives
must be calculable, which may not be the case when minimizing both the mod-
ulus and spatial segmentations as in [Baldewsing et al., 2006]. This method was
 33

first applied to the field of elastography in 1996 [Kallel and Bertrand, 1996]. The
functional is often expressed as follows,

(E) = kum uc (E)k2 + kE E0 k2 , (2.20)

where is the value of the functional, E is the Youngs modulus, u m and u c are
the measured and computed displacements [Yalavarthy et al., 2007b]. The second
term penalizes against large values of the Youngs modulus, where E 0 is a guess of
the Youngs modulus [Yalavarthy et al., 2007b]. The constant is the weighting of
this penalty term. Ordinary Least-Squares (OLS) methods such as the Levenberg-
Marquardt technique are often used to solve problems of this form [Zhdanov, 2002;
Marquardt, 1963]. The functional is expanded using a Taylor series, and the
derivative of the entire expression is set to zero to arrive at the iterative update
equation shown below:

 T
J J + I Ei = J T i1 (Ei1 E0 ),

(2.21)

where J is known as the Jacobian matrix, containing the derivatives of displace-

ment with respect to the Youngs modulus, I is the identity matrix, is the
Kronecker delta, and T is the matrix transpose. The matrix, J T J, is known
as the Hessian. The Hessian matrix must be positive definite in this method
[Marquardt, 1963]. However, this is rarely the case. Furthermore, it often has
a near-singular inverse, implying a high sensitivity to noise. This is otherwise
known as an ill-conditioned problem. For this reason, I is often added to the
Hessian to improve its condition.

Several groups have included spatial information in the minimization func-

tional for near-infrared tomography and MRI imaging [Yalavarthy et al., 2007b;
Yalavarthy et al., 2007a; Brooksby et al., 2005b]. This information is intended to
make the problem more well-posed by constraining the solution space [Yalavarthy
 34

et al., 2007b]. In addition, these methods do not require the E0 term, which biases
the solution.

Yalavarthy et al. included spatial information in their minimization for dif-

fuse optical tomography using two techniques; soft and hard spatial priors. These
methods required the a priori segmentation of homogeneous regions within each
raw image [Yalavarthy et al., 2007b]. The hard-priors minimization assumes ho-
mogeneous properties within segmented regions. In this way, the Hessian matrix
becomes an n n matrix, where n is the number of segmented regions. This results
in an overdetermined problem, making the Hessian positive definite [Yalavarthy
et al., 2007b; Srinivasan et al., 2004]. This eliminates the need for additional regu-
larization. However, this method will not resolve inhomogeneities within specified
regions by definition. The soft-priors approach includes spatial information in the
regularization term. In this method, a penalty is imposed for variation within
each region.

Computing the Hessian matrix in the Marquardt based minimizations is com-

putationally intensive. Adjoint-weighted equation (AWE) methods such as the
Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm avoid this comoputation
by approximating the Hessian, and updating its value at every iteration. These
methods are attractive because they only require the gradient of the functional.
However, the lack of this information causes slower convergence requiring more it-
erations. The BFGS method has been used in vascular elastography both in simu-
lation and experiments [Richards and Doyley, 2011; Bertagna and Veneziani, 2014;
Perego et al., 2011]. Richards et al. used this method of minimization to compute
the Youngs modulus of vessel phantoms using IVUS elastography [Richards and
Doyley, 2011]. They found that useful modulus elastograms could be computed
when using spatial priors. Youngs modulus errors for these computations were
under 10%. Segmentation of the vessel wall and inclusion regions were segmented
a priori from IVUS B-scan images. However, this technique was not applied to in
 35

vivo or ex vivo artery samples. Additionally, nonlinear mechanical models were

not tested. Figure 2.7 shows their recovered modulus elastograms a) when ac-
curate spatial priors were used, and b) when no spatial priors were used. The
simulated inclusions were not visible without the structural information [Richards
and Doyley, 2011].

Figure 2.7: Reconstructed modulus elastograms from simulated IVUS images a) with
and b) without spatial information. Adapted from [Richards and Doyley, 2011].

Both OLS and AWE optimization methods compute local minima because
their underlying functionals are non-convex. Energy-based minimization tech-
niques use convex functionals to ensure a faster convergence to the global minima
than either OLS or AWE [Jadamba et al., 2014; Doyley et al., 2014]. One such
technique is the modified output least-squares method, which was used to com-
pute the shear modulus from simulated data sets [Jadamba et al., 2014]. Recently
developed, this algorithm that has yet to be tested on noisy data. Because of its
convex form, this method showed less sensitivity to the initial conditions than OLS
minimizations. The computational cost of the MOLS algorithm was also shown
to be less than the OLS algorithm, with the discrepancy increasing with the de-
grees of freedom. However, this method is more susceptible to noise than OLS or
AWE functionals, as it requires the computation of spatial derivatives. It also re-
 36

quires a full or overdetermined system, meaning the number of computed material

properties cannot exceed the number of displacement images. Ultrasound-based
elastography often uses just one component of displacement due to its poor lat-
eral resolution, which would pose a problem for this method. This same limitation
would preclude its use in multi-parameter material models, such as those used in
some hyperelastic reconstructions [Wittek et al., 2013]. An example of the MOLS
minimization is shown below in Fig. 2.8.

Figure 2.8: Reconstructed and ground truth shear modulus elastograms reconstructed
using the MOLS method on noise-free simulated displacements. Adapted from [Jadamba
et al., 2014].

2.4 Current Limitations of Quantitative Elas-

tography

It is common to use linear elastic modeling in Youngs modulus reconstructions

[Baldewsing et al., 2006; Le Floch et al., 2009; Richards and Doyley, 2011].
However, mechanical testing of human arteries shows a hyperelastic trend with
compression [Holzapfel et al., 2005; Barrett et al., 2009]. This hyperelasticity
will result in the tissue expressing an apparent stiffness greater than the lin-
ear value over large strains. Baldewsing et al. justifies their use of a linear
model by assuming the measured strain to be small (< 2%) [Baldewsing et al.,
 37

2004]. Other groups have used similar assumptions [Richards and Doyley, 2011;
Le Floch et al., 2009]. Although linear models allow us to characterize tissue com-
ponents, they are typically not useful when predicting the propensity of plaque
rupture. Thus, it is common to only report stiffness and not stresses within
the artery. Studies of ruptured and non-ruptured arteries have shown correla-
tion between stress and rupture but not stiffness and rupture [Cheng et al., 1993;
Loree et al., 1992]. These large stresses occur on the boundary between the plaque
and the plaque cap where the maximum strain is observed. It is therefore neces-
sary to measure large strains to compute peak stresses.

To mitigate the effects of noise on reconstruction, it is common to make

the assumption of homogeneity within specified regions. These regions are of-
ten selected a priori to represent different materials, i.e. plaque, wall, fi-
brous cap [Richards and Doyley, 2011; Karimi et al., 2008]. In an in vivo set-
ting, the region boundaries cannot be exactly determined, and researchers must
rely on b-mode or strain images to perform manual or automated segmenta-
tion. Error in these segmentations will directly impact the recovered parame-
ters. Additionally, some features such as microcalcifications or microvessels may
be too small to observe in ultrasound but may still impact the vessel stresses.
The presence of microcalcifications within the fibrous cap can increase the lo-
cal stresses by a factor of 50-300% [Bluestein et al., 2008; Wenk et al., 2010;
Vengrenyuk et al., 2006]. Microvessels in the lipid pool may increase local stresses
by over an order of magnitude [Teng et al., 2012]. These stresses cannot be de-
tected if homogeneity is assumed.
 38

3 General Methods

All of the simulation studies in the thesis required the same basic steps: 1) Artifi-
cial displacement fields were generated using finite element modeling, 2) Simulated
ultrasound RF images were created using the Field II environment, 3) Displace-
ments were tracked between pairs of RF frames, 4) Youngs modulus elastograms
were generated using the soft-priors reconstruction method, and 5) stresses and
strains were computed from the displacement fields and modulus elastograms.
Additionally, in phantom studies, polyvinyl alcohol cryogel vessel analogues were
created, and the stiffness of these vessels was measured using a uniaxial mechanical
testing system. These processes are described below.

3.1 Finite Element Modeling

The finite element (FE) method is often used to compute the solution for field
problems. This technique divides the field into a series of elements, each consisting
of three or more nodes, depending on the complexity of the model. Within each el-
ement, the spatial variation is approximated and discretized based on the number
of nodes per element. For example, a three-node triangular element can only rep-
resent linear variation between the nodes. These field quantities are described by
a set of differential equations. A simpler method to solve the differential equation
 39

is the finite difference method. Although this technique is less computationally in-
tensive, it is not as adept at handling boundary conditions. A custom (FE) solver
was created to model tissue as both a linear elastic and hyperelastic medium us-
ing MATLAB (Release 2014a, Mathworks, Natick, MA). Outlines of the imaged
vessels were used to define the boundaries of the finite element structure. These
outlines were either hand-segmented from ultrasound b-mode images or computer-
generated in the case of simulation studies. COMSOL (COMSOL inc. Burlington,
MA) was used to create finite element meshes within these structures using a De-
launay triangulation technique. Three-node isoparametric constant strain triangle
(CST) elements were used in all models. These elements use a single integration
point at their center, and are defined on reference coordinates, where the element
nodes occur at (0,0), (1,0), and (0,1). An example of this is shown in Fig. 3.1.

Figure 3.1: Sample finite element mesh created in Comsol using 1124 CST elements

The induced displacements at each node are governed by a set of partial differ-
ential equations, expressed in the weak form. Using this framework, displacements
can be related to nodal forces through the following relation (Hookes Law),

Ku = f, (3.1)
 40

where f is force, u is displacement, and K is the global stiffness matrix [Cook

et al., 2007]. In this case, K is a 2n 2n matrix, where n is the number of nodes
in the finite element model. K can be constructed by summing all elemental
stiffness matrices, after remapping each k e from elemental nodes (1-3) to global
nodes (1-n). These K e can be solved by using the following integral with respect
to the element domain, :
Z
Ke = hBeT De Be d, (3.2)

where B is the strain-displacement matrix, which relates nodal displacement with

strain at each elemental integration point; D is the constitutive relation for the
chosen mechanical model, which relates elemental strains with elemental stresses,
and h is the element thickness [Cook et al., 2007]. The constitutive matrices for
linear elasticity and hyperelasticity are given in Equations 2.2, 2.3, and 6.8. The
elemental B matrix has the following form:

y23 0 y31 0 y12 0
1
Be = 0 x32 0 x13 0 x21 , (3.3)

2A
x32 y23 x13 y13 x21 y12

where x ij is defined by x i -x j , A is the area of the element, and i and j are the
elemental node numbers [Cook et al., 2007]. Displacements can be computed from
Equation 3.1 through a matrix inversion.

These displacements can be induced by imposing Dirichlet or Traction bound-

ary conditions to the model. Dirichlet, or fixed, boundary conditions will assign
specific displacements to selected nodes. Traction boundary conditions can be
used to specify the loading force on selected nodes. These boundary conditions
 41

are implemented in the example below [Cook et al., 2007],

k11 k12 k13 u f
1 1
Global equation k21 k22 k23 u2 = f2


k31 k32 k33 u3 f3

k k k u f
11 12 13 1 1
Dirichlet boundary (u2 = a) 0 0 u2 = a (3.4)

1

k31 k32 k33 u3 f3

k k k u f
11 12 13 1 1
Traction boundary (f2 = b) k21 k22 k23 u2 = b


k31 k32 k33 u3 f3

3.2 RF Simulation

We used the Field II [Jensen, 1991] simulation environment to compute the acous-
tic response of the IVUS and linear array transducers used in these studies. We
simulated the acoustic response of the pre-deformed vessel by randomly distribut-
ing point scatterers (16 scatterers per wavelength) within the simulated vessels
[Wagner et al., 1983]. The speed of sound in the vessel models was set to 1540
m/s. Solving the forward elasticity problem produced displacements that were
used to redistribute the point scatterers of the pre-deformed vessels (i.e., compute
the acoustic response of the post-deformed vessel).
 42

3.3 Displacement Estimation

3.3.1 Cross-Correlation

Using this method, axial and lateral displacement elastograms were created by
applying a 2D cross-correlator [Korukonda and Doyley, 2011b; Korukonda et al.,
2013] to the pre-and post-deformed RF echo frames. A reference window of size Wz
Wx was chosen in the pre-compression image, RF1. A similarly sized window
was defined in the post-compression image, RF2, and was precessed within a search
window, Sx Sz . The cross-correlation was taken between the reference window
and each shifted window in RF2 to create a map of cross-correlation values over
the entire search region. A paraboloid was fit to this array, and the peak value
was used to determine the pixel motion between the two images, assuming the
true displacement will correspond to the highest cross-correlation. This process
was repeated across the entire image in steps of stepx and stepz for the x and z
axes. A diagram of this process is shown below:

Cross correlation

Fit paraboloid to determine

displacement

RF1 RF2

Figure 3.2: Diagram of cross-correlation displacement tracking algorithm: The blue

boxes are the pre- and post-compression images (RF1 and RF2). The orange box
represents the reference window, the red box represents the search window, and the
shaded lavender boxes represent the cross-correlation map within the search window.
 43

3.3.2 Image Registration

Using this method, displacements were computed by transforming RF2 to match

RF1. This was done for each pixel by minimizing the functional, , shown below
[Richards and Doyley, 2013],
Z Z
2
[u(x)] = 0.5 (I1 (x) I2 (x + u(x))) d + 0.5 G(x)(u(x)sym : u(x)sym d.

(3.5)
In the first term of this equation I 1 is the pre-compression image, I 2 is the post-
compression image, and u(x ) are the spatial displacements. refers to the entire
imaging domain. This portion of this functional minimizes the difference between
the images. The second term imposes a constraint on the smoothness of the
resulting displacement map by penalizing large strains. is a weighting constant,
u(x) is the displacement gradient, and G(x ) is a spatially varying weighting
matrix. G(x ) was chosen to penalize large strains near the lumen less than those
near the outer boundary, as these strains are expected to decrease with the square
inverse of the radius. An initial guess of u(x ) was chosen to be the result for the
cross-correlation displacement tracking.

3.4 Strain and Stress Estimation

Strains in a given coordinate system were computed by taking the gradient of the
displacements in that same system. The 2D stress tensor, ij , was computed as
follows:  

ij = E ij  + ij , (3.6)
1 2
where is the Poissons ratio,  is the strain tensor in an arbitrary coordinate
system, and is the Kronecker delta. The principal stresses were computed from
these arbitrary stresses to produce a coordinate independent tensor through the
 44

following transformation:

i = max(1 , 2 ) ii = min(1 , 2 ) , (3.7)

where 1 and 2 are the eigenvalues of the arbitrary stress tensor [Richards et al.,
2015].

3.5 Soft-Priors Reconstruction

An iterative inversion scheme was used to compute modulus elastograms [Doyley

et al., 2000; Kallel and Bertrand, 1996; Oberai et al., 2004], which estimates
the Youngs modulus by minimizing the difference between the measured and
computed displacements in a least squares sense. This method requires the a priori
segmentation of the vessel into R regions of assumed homogeneity. A diagram of
this process is shown below:
 45

Figure 3.3: Diagram of reconstruction flow

After displacement tracking, a homogeneous initial guess of the Youngs mod-

uli, E, was chosen and used to compute a set of displacements, u c , using the FE
method. The computed displacements were compared to the measured displace-
ments, um , and the Youngs moduli were updated using an optimization function
to minimize the two sets of displacements as functions of E.

The objective function, (E), for this optimization is based on the Levenberg-
Marquardt model in Equation 2.21 and has the following form:

(E) = kum uc (E)k2 + kL [E]k2 , (3.8)

where L is the 2n n penalty matrix that was defined as follows [Yalavarthy

 46

et al., 2007a; Yalavarthy et al., 2007b]:

1 i=j

Li,j = 0 /R ,
ij (3.9)


1

ij R
m

where m is the total number of nodes contained in a given region, and i and j
are indices of L. The L matrix (i.e., Laplacian) computes the average modulus in
each region explicitly as follows:

E2 E3 EN
L1 E = E1 E1 hEi . (3.10)
N N N

Minimizing Equation 3.8 with respect to the Youngs modulus produces the fol-
lowing matrix solution at the (k + 1)th iteration:

E k+1 = E k + [J T J + LT L]1 J T um uc (E k ) LT LE k .
 
(3.11)

Each column in the Jacobian matrix, J, was computed by taking the derivative of
du
Equation 3.1, and rearranging the resulting expression to find dEn
. This compu-
tation is carried out below:

du dK
K + u=0 (3.12)
dE dE

du dK
= K 1 u,
dEn dEn
du
where dEn
are a set of vectors defining the derivatives of displacement with respect
to each nodal Youngs modulus En .
 47

3.6 Phantom Fabrication

Polyvinyl alcohol cryogel (PVAc) phantoms were created using a suspension of 8%

cryogel in water using a controlled and repeatable process [Fromageau et al., 2007;
Maurice et al., 2005b; Richards and Doyley, 2011]. More specifically, we placed
two off-center rods (3.12 mm diameters) in a cylindrical mold (12 mm diameter
by 120 mm long). One rod was circular while the other was semicircular. Molten
polyvinyl alcohol (PVA) was poured in the vacant cavity between the rods and
the mold. The vessel wall was constructed by subjecting the sealed mold to
three freeze-thaw cycles. One thermal cycle was completed in 24 h where the
temperature was varied from 20 C to 20 C. We removed the semicircular rod
after thermal cycling and filled the vacant cavity with PVA, and subjected the
phantom to two additional freeze-thaw cycles. After thermal cycling, the phantom
was removed from the mold and stored at room temperature in water. A sample
phantom is shown in Fig. 3.4, where the soft inclusion appears as a gray region
due to the addition of silicon carbide (SiC).

Figure 3.4: Sample PVAc phantom, where the soft inclusion is gray due to the addition
of silicon carbide (SiC)
 48

3.7 Mechanical Testing

The linear Youngs moduli of the PVAc plaques and vessel walls were verified
through mechanical testing using a Landmark Servohydraulic Test System (MTS
Testing Systems, Eden Prairie, MN, USA). For this testing, five cylindrical phan-
toms (diameter: 2 cm, height: 2 cm) were fabricated from the same solutions as
the vessel phantoms. The linear Youngs modulus was computed by uniaxially
compressing each phantom by 5% in steps of 0.01%. The applied force was mea-
sured using a load cell. The Youngs modulus was computed using Equation 2.10.
The testing was completed on the same day as the elastography experiments.
 49

4 Quantitative Vascular
Elastography

4.1 Introduction

Non-invasive vascular elastography (NIVE) [Maurice et al., 2004] is an emerg-

ing ultrasound imaging technique that visualizes the strain distribution within
vascular tissuesinformation that is related to the external boundary condi-
tions and mechanical properties [de Korte et al., 2002]. Like its intravascular
counterpart [de Korte et al., 2000; Schaar et al., 2003], NIVE uses the pul-
sating intraluminal pressure as the source of mechanical stimulation; however,
unlike intravascular ultrasound elastography (IVUSE), it measures strain non-
invasively. Schaar et al. [Schaar et al., 2003] demonstrated that strain elas-
tograms (strain images) can reveal rupture-prone regions on the fibrous cap.
NIVE acquires strain elastograms from the cross-sectional plane of carotid ar-
teries, which are difficult to interpret. This difficulty occurs because NIVE
measures axial strains in the transducers frame of reference (Cartesian), which
doesnt represent the strains in the vessel coordinate system. To overcome
this limitation, several groups, including ours, have developed strategies to es-
timate both the axial and lateral components of displacements. More specif-
ically, the proposed strategies include (a) visualizing von Mises strain [Mau-
 50

rice et al., 2004]; (b) using ultrasound beam steering techniques to measure
radial and circumferential strains [Techavipoo et al., 2004; Rao et al., 2007;
Hansen et al., 2009; Hansen et al., 2010a; Hansen et al., 2010b]; (c) comput-
ing radial and circumferential strain elastograms from axial and lateral displace-
ments measured with sparse-array ultrasound imaging [Korukonda et al., 2013;
Korukonda and Doyley, 2011a].

Besides improving NIVE strain elastograms, better estimates of lateral dis-

placements will improve the performance of model-based elastography [Hansen
et al., 2013]. Ultrasound model-based elastography methods produce less accu-
rate modulus elastograms than their MRI counterpart [Doyley, 2012] because only
the axial component of displacement is included in the modulus recovery process
ultrasound provides imprecise estimates of the other components of displacements
[Lubinski et al., 1996].

The goal of this chapter is to develop a more quantitative approach to NIVE

based on sparse array imaginga technique we call sparse array quantitative vas-
cular elastography (QVE). Sparse array imaging provides high precision axial and
lateral displacements [Korukonda and Doyley, 2011a], but there are concerns that
the low transmit power of sparse-arrays could hamper clinical use. We hypothesize
that QVE can produce useful modulus elastograms in superficial organs such as the
carotid arteries. To corroborate this hypothesis, we assess the performance (ac-
curacy, contrast recovery, and contrast transfer efficiency) of modulus and strain
elastograms acquired with sparse-array imaging relative to those obtained using
conventional and plane-wave imaging.
 51

4.2 Methods

4.2.1 Ultrasound imaging

Conventional linear array imaging was performed with 64 active transmission

elements and 32 reception elementsthe default configuration for the ultrasound
scanner used in the experimental study. The ultrasound beam was focused by
applying delays during transmission, and using a Hanning apodization function on
the received echoes. Beamforming was performed with the delay-sum technique.
The lateral sampling frequency of all beamformed RF echo frames was increased
from 0.30 lines/mm to 52 lines/mm using the method described in Konofagou and
Ophir [Konofagou and Ophir, 1998].

Sparse-array imaging was performed by transmitting on 15 elements sequen-

tially, which spanned the full length of the array, and receiving on all 128 elements.
Like conventional linear array imaging, all RF echo frames were beamformed using
the delay-sum technique. Dynamic focusing was performed both on transmission
and reception. The received signals were apodized with a boxcar function [Ko-
rukonda and Doyley, 2011a].

Plane-wave imaging was performed by transmitting and receiving on all 128

elements. Beamforming was also performed with the delay-sum technique.

4.2.2 Beamforming

Radio-frequency echo frames were reconstructed by applying the delay-sum tech-

nique to conventional linear array, sparse-array, and plane-wave data. The
backscatter signal at point (x0 , z0 ) in the image was reconstructed as follows:

Ntx X
X Nrx
s(x0 , z0 ) = wij RFij (t (x0 , z0 )), (4.1)
i=1 j=1
 52

where Ntx and Nrx represent the number of transmit and receive elements in
the linear array; RFij (t) represents the RF echo when the ith element transmits
and j th element receives; t represents the time of flight of the echo; (x0 , z0 )
represents the round trip time from point (x0 , z0 ); wij represents the apodization
weight. During sparse-array imaging only a few elements in the array were active
during transmission, but all the elements were active during reception. We used
a transmit-receive (T/R) matrix to compute the apodization weights wij that
produced beam patterns similar to those generated with a fully populated array
[Chiao and Thomas, 1996; Korukonda and Doyley, 2011a; Korukonda and Doyley,
2012]. The apodization weights employed in plane-wave imaging were identical
to those employed in a fully populated array because all 128 elements were active
during transmission and reception. The time, tx , required to reach a given point
(x0 , z0 ) was computed as follows:
q 
2 2
tx = (xi x0 ) + z0 /c, (4.2)

where xi represents the location of the transmitting element and c represents the
speed of sound. During plane-wave imaging, the time required for the ultrasound
beam to travel to a given point was computed as follows:

z0
tx = . (4.3)
c

For all imaging techniques, the return trip time (rx ) was computed as follows:
q 
rx = (xj x0 )2 + z02 /c, (4.4)
 53

where xj represents the location of the receiving element. The total round trip
time to and from the point (x0 , z0 ) was computed as follows:

(x0 , z0 ) = tx + rx . (4.5)

Sparse-array and plane-wave RF echo frames were reconstructed on a 10 mm

10 mm grid that had a lateral sampling frequency of 52 lines/mm and an axial
sampling frequency of 40 MHz. The axial sampling frequency of beamformed
conventional RF echo frames was similar to those of sparse-array and plane-wave
echo frames; however, the lateral sampling frequency was lower 0.3 lines/mm.

4.2.3 Displacement Estimation

Axial and lateral displacement elastograms were generated by applying a 2D cross-

correlator [Korukonda and Doyley, 2011a; Korukonda et al., 2013] to the pre-and
post-deformed RF echo frames. All cross-correlation analyses were performed with
2 mm 2 mm kernels that overlap by 80% in both the axial and lateral direction.
Spurious displacement estimates were removed through median filtering with a 5
5 window.

4.2.4 Data Analysis

The elastographic contrast-to-noise ratio (CN Re ) and the normalized root-mean-

square performance metric were used to assess the performance of both strain
and modulus elastograms. The elastographic contrast-to-noise ratio was defined
as follows [Bilgen and Insana, 1997]:

2 (w p )2
CN Re [dB] = 20 log , (4.6)
(w2 + p2 )
 54

where w and p represent the mean strain or modulus values within selected
regions; and w and p represent the standard deviation in strain or modulus
values in the corresponding regions.

The normalized root-mean-squared error (NRMSE) was computed as follows:

s
N
(xm (i)xc (i))2
P
i=1
N
N RM SE = , (4.7)
max[xc ] min[xc ]

where x c and x m represent the actual and measured parameters (modulus or

strain), respectively, and N is the number of pixels in the region of interest (ROI).

4.3 Simulation Study

The goal of this simulation study was to corroborate the hypothesis that QVE
can recover useful modulus elastograms from attenuating materials. We compared
the quality of modulus elastograms reconstructed from axial and lateral displace-
ments measured from vessel phantoms with varying modulus contrasts (-12.04
dB, -6.02 dB, and -2.50 dB) and attenuation coefficients (0.6 dB/cm, 1.5 dB/cm,
and 3.5 dB/cm). Axial and lateral displacements were computed by applying
our two-dimensional echo tracking technique to RF echo frames acquired with
(a) conventional linear array imaging with lateral interpolation, (b) sparse-array
imaging, and (c) plane-wave imaging. For the reminder of this manuscript, the
first approach will be referred to as conventional linear array (CLA) imaging. A
two step process was used to synthesize RF echo frames. The proceeding sub-
sections describe (1) the two key stages of the simulation process: mechanical
(solving the forward elasticity problem) and acoustic modeling; (2) the modulus
reconstruction protocol.
 55

4.3.1 Mechanical Model

Simulated arteries were created by using the FE method, assuming linear elas-
ticity. The simulated arteries were incompressible ( 0.495) with inner and
outer radii of 1.5 and 6 mm respectively. We assigned a Youngs modulus of
50 kPa to the vessel wall, but varied the modulus of the simulated plaques to gen-
erate vessel phantoms with modulus contrasts of -12.04 dB, -6.02 dB, and -2.50
dB (12.5 kPa, 25.0 kPa, and 37.5 kPa). These mechanical parameters were repre-
sentative of those reported in Hansen et al. [Hansen et al., 2009]. A uniformly
distributed pressure (666.7 Pa) was applied to the inner lumen, which generated a
maximum strain of 1%. To minimize rigid body motion, we constrained the mo-
tion of two nodes on the inner and outer boundaries. More specifically, one node
on the outer boundary was not allowed to move in the circumferential direction,
while the node directly across from it on the inner boundary was not permitted
to move radially.

4.3.2 Acoustic Model

We used the Field II [Jensen, 1991] simulation environment to compute the acous-
tic response of the L14-5/38 linear transducer array (Prosonic Corporation, Ko-
rea) used in the experimental studies, when operating at 5 MHz. We simulated
the acoustic response of the pre-deformed vessel by randomly distributing point
scatterers (16 scatterers per wavelength) within the simulated vessels [Wagner
et al., 1983]. The speed of sound in the vessel models was set to 1540 m/s.
Solving the forward elasticity problem produced displacements that were used to
redistribute the point scatterers of the pre-deformed vessels (i.e., compute the
acoustic response of the post-deformed vessel). We simulated RF echo frames for
vessels with attenuation coefficients of 0.6 bD/cm, 1.5 dB/cm, and 3.5 dB/cm.
We used an additive Gaussian white noise model to simulate RF echo frames with
 56

signal-to-noise ratios (SNR) of 26 dB (CLA), 20 dB (SA), and 29 dB (PW). To

assess variability in the reconstruction approach, we performed 25 statistically
independent reconstructions at each contrast and attenuation coefficient.

4.3.3 Modulus Reconstructions

All modulus elastograms were reconstructed from a homogeneous Youngs mod-

ulus distribution of 50 kPa. Structural information was obtained by manually
segmenting either the modulus distribution used in the finite element model or
sonograms. All reconstructions were performed using a homogeneous finite el-
ement mesh consisting of 3910 elements. Reconstructions were performed with
three different values of the regularization parameter that we selected objectively
using the L-curve method [Vogel, 2002; Richards and Doyley, 2011]. More specifi-
cally, the regularization parameter was set to 7.2 109 , 5 1010 , and 1.2 109
when reconstruction was performed using displacements measured with CLA, SA,
and PW imaging, respectively.

4.4 Phantom Study

4.4.1 Phantom Fabrication

Two vessel phantoms were fabricated from a suspension of polyvinyl alcohol (El-
vanol 71-30, Dupont, Wilmington, DE, USA), ultra fine aluminum oxide (0.3 m,
Logitech Ltd., Glasgow, Scotland, UK), and silicon carbide (320 Grit, Fisher
Scientific, Fair Lawn, NJ, USA). We used aluminum oxide (Al2 O3 ) and silicon
carbide (SiC) to control the attenuation coefficient [Inglis et al., 2006] and the
echogenicity of the vessels, respectively. Table 1 shows the composition of each
phantom.
 57

Table 4.1: Composition of materials used to fabricate vessel phantoms

PVA % Al2 O3 % SiC % Water 18 M Thermal Cyles

Phantom #1: Plaque 8 0 3 89% 2
Phantom #1: Vess wall 8 0 1 91% 5
Phantom #2: Plaque 8 5 2 85% 2
Phantom #2: Vess wall 8 5 0 87% 5

4.4.2 Elastographic Data Acquisition

The equipment used for elastographic imaging consisted of a Sonix RP ultrasound

system (Ultrasonix, Peabody, MA, USA) that was equipped with a 128 element
linear transducer array (L145/38 probe), a multichannel data acquisition system
(Sonix DAQ , Ultrasonix, Peabody, MA, USA), a simple water column system,
and a pressure wire (Millar Instruments Mikro-Cath, Houston, TX, USA ).

We configured the Sonix RP ultrasound scanner to acquire (a) conventional

ultrasound echo images (an aperture consisting of 64 transmission elements and
32 reception elements), b) sparse-array echo data (15 transmission elements se-
quentially and 128 reception elements), and (c) plane-wave images (128 active
transmission and reception elements). All echo imaging was performed at 5 MHz,
and the received signal was sampled to 12 bits at 40 MHz. We used a simple
water column system to pressurize the vessels to 666.7 Pa. All three data sets
were collected with interleaved ultrasound scanning, which we implemented using
a software development kit (TEXO SDK, v5.7.1, Ultrasonix, Peabody, MA, USA,
). The delay-sum technique was used to beamform sparse-array and plane-wave
data.

4.4.3 Modulus Reconstructions

A typical mesh consisted of 7,500 elements. Pressure on the inner lumen was mea-
sured with a pressure wire during the elastographic imaging. We interpolated the
measured displacements to the nodal coordinates of all the finite element meshes
 58

using a cubic interpolation function. The boundary conditions used in the phan-
tom were similar to those used in the simulated study. All reconstructions were
performed using an homogeneous Youngs modulus of 50 kPa. The regularization
parameters were identical to those used in the simulation study.

4.4.4 Measuring the Attenuation Coefficient

We measured the attenuation coefficient of representative cylindrical-shaped sam-

ples of each tissue component at room temperature. All attenuation measurements
were performed at 5 MHz as follows [Surry et al., 2004]:
 
1 Pp
= 10 log 10 , (4.8)
d Pw

where P w and Pp represent the power received from water and the test sample,
respectively. d was the thickness of the test sample.

4.5 Results

4.5.1 Simulation Study

Figure 4.1 shows examples of sonograms acquired from simulated vessel phantoms
(attenuation coefficients of 0.6 dB/cm, 1.5 dB/cm, and 3.5 dB/cm) with CLA,
SA, and PW. Attenuation degraded the quality of all sonograms, but more quickly
when sparse-array imaging was performed. This occurred because less acoustic
power was transmitted during sparse-array imaging. Plane-wave sonograms con-
tained visible side-lobes because no focusing was applied during transmission.
 59

Figure 4.1: Sonograms obtained using (a) conventional, (b) sparse-array, and (c) plane-
wave imaging from simulated vessel phantoms with attenuation coefficients of 0 dB/cm,
1.5 dB/cm, and 3.5 dB/cm going from left to right.

Figure 4.2 shows axial and lateral profiles obtained from CLA, SA, and PW
point-spread functions (PSF) when imaging was performed in mediums with at-
tenuation coefficient of 0.6 dB/cm, 1.5 dB/cm, and 3.5dB/cm. The axial profiles
were similar because identical transmission frequencies were employed in all three
imaging approaches. The axial profiles were oscillatory and their amplitudes de-
creased with increasing attenuation. However, the lateral profiles were different.
Sparse-array and conventional linear array imaging had the narrowest and broad-
est lateral profiles (beamwidth), respectively. The lateral profiles of sparse-array
and plane-wave PSFs contained side-lobes; however, those obtained from the CLA
didnt. The box car apodization function and the larger transmission apertures
 60

employed during SA and PW imaging were responsible for the observed side-lobes.

Figure 4.2: (a) Axial and (b) lateral profiles taken through the center of point-spread-
functions (PSF) for conventional, sparse-array, and plane-wave imaging system.

Figures 4.3 and 4.4 show radial and circumferential strain elastograms acquired
from vessel phantoms with modulus contrast of -6.02 dB with CLA, SA, and PW
imaging. Increasing the attenuation coefficient from 0.6 dB/cm to 3.5 dB/cm
degraded the quality of all strain elastograms.
 61

Figure 4.3: Montage of radial strain elastograms computed from RF echo frames ac-
quired with (a) conventional linear array, (b) sparse-array, and (c) plane-wave imaging.
Ideal strain elastograms are shown in the first column of each montage, and the elas-
tograms obtained from vessel phantoms with attenuation coefficients of 0 dB/cm, 1.5
dB/cm, and 3.5 dB/cm are shown in the remain columns.
 62

Figure 4.4: Montage of circumferential strain elastograms computed from RF echo

frames acquired with (a) conventional linear array, (b) sparse-array, and (c) plane-
wave imaging. Ideal circumferential strain elastograms are shown in the first column
of each montage, and the elastograms obtained from vessel phantoms with attenuation
coefficients of 0.6 dB/cm, 1.5 dB/cm, and 3.5 dB/cm are shown in the remain columns.
The two rectangular box in (a) denotes the regions of interests corresponding to the
plaque and vessel wall that was used to compute CN Re values reported in Fig. 5.7.

Figure 4.5 shows the errors incurred when estimating axial and lateral dis-
placements with CLA, SA, and PW imaging plotted as a function of attenuation
coefficient and modulus contrast. The accuracy of axial displacement elastograms
measured with CLA and SA imaging depended on both attenuation coefficient
and modulus contrast imaging. More specifically, the accuracy of SA axial dis-
placement elastograms varied from 1.8% to 7% for the range of modulus contrast
and attenuation coefficient explored in this study; whereas, the accuracy of CLA
axial displacement elastograms varied from 4.9% to 8% for similar range of modu-
lus contrast and attenuation. The accuracy of PW axial displacement elastograms
 63

were marginally affected ( 2.5%) by attenuation and modulus contrast. Errors

incurred when measuring lateral displacement displayed a similar trend. More
specifically, the accuracy of lateral displacement elastograms varied from 1220%
(CLA), 1.83.9% (SA), and 2.23% (PW) for the range of attenuation coefficient
and modulus contrast explored in this study.

Figure 4.5: The normalized root-mean-error incurred when estimating (a) axial and (b)
lateral displacements with conventional linear array (first column), sparse-array (second
column), and plane-wave (third column) imaging.

Figure 4.6 shows the corresponding modulus elastograms reconstructed from

the axial and lateral displacement elastograms measured with CLA, SA, and PW
imaging (i.e., Fig. 4.4). All modulus elastograms revealed the plaque, but SA and
PW modulus elastograms contained less artifacts.
 64

Figure 4.6: Montage of modulus elastograms corresponding to the strain elastograms

shown in Figs 3 and 4. The first column of each montage shows the actual modu-
lus elastograms and the remain columns shows the elastograms obtained from vessel
phantoms with attenuation coefficients of 0.6 dB/cm, 1.5 dB/cm, 3.5 dB/cm. Showing
(a) conventional, (b) sparse-array, and (c) plane-wave modulus elastograms. The two
rectangular box in (a) denotes the regions of interests corresponding to the plaque and
vessel wall that was used to compute CN Re values reported in Fig. 4.7.

Figure 4.7 shows bar plots of the recovered contrast and the accuracy of mod-
ulus elastograms. The image reconstruction process recovered modulus contrast
more accurately when it was applied to displacements measured with SA and PW
than those measured with CLA. For CLA and SA imaging, the variance of the
recovered modulus contrast increased rapidly with attenuation; however, variance
increased slightly with increasing attenuation and modulus contrast during PW
imaging. Modulus elastograms computed with SA and PW were more accurate
than those computed with CLA. More specifically, the accuracy of modulus elas-
tograms varied from 1726% (CLA), 210% (SA), and 45% (PW), over the range
of modulus contrast and attenuation explored in this work.
 65

Figure 4.7: (a) Recovered modulus contrast and (b) modulus error incurred when elas-
tography imaging was performed with conventional linear array (left-column), sparse-
array (middle-column) and plane-wave (right-column) imaging.

Figure 4.8 shows the CN Re computed from strain and modulus elastograms
(Figs 4.4-4.6) plotted as a function of attenuation coefficient and modulus con-
trast. Both types of elastograms (i.e., strain and modulus) displayed a similar
trend. CN Re decreased with increasing attenuation, and increased with raising
modulus contrast; however, the CN Re of strain elastograms were lower than those
computed from modulus elastogram. This was due to differences in spatial reso-
lution of modulus and strain imaging [Doyley et al., 2005]. If spatial resolution of
the images are similar, then both images should have the same CNRe ; however,
equalizing the spatial resolution of modulus and strain elastograms is not trivial
given the complexity of the reconstruction process.
 66

Figure 4.8: The contrast-to-noise ratio of strain (top and middle rows) and modulus
(bottom rows) elastograms obtained from vessel phantoms modulus contrast of -2.5 dB,
-6.02 dB, and -12.04 dB; and attenuation coefficients of 0.6 dB/cm to 3.5 dB/cm.

4.5.2 Phantom Study

Figures 4.9i and 4.9iv show representative examples of sonograms obtained from
vessel phantoms with the three imaging methods (CLA, SA and PW). Side-lobes
were discernible in the plane wave sonograms, which was expected because no
focusing was performed during transmission.
 67

Figure 4.9: Sonograms and elastograms obtained from vessel phantoms #1 (a) and
#2 (b) when elastographic imaging was performed with CLA, SA, and PW imaging.
Showing (i& iv) Sonograms, (ii & v) cross-correlation maps, and (iii & vi) modulus
elastogram. In the sonogram, the crescent shape ROI denotes the plaque.
 68

Figures 4.9ii and 4.9v show example correlation images (i.e., the peak correla-
tion coefficient obtained at each pixel during echo tracking). When elastographic
imaging was performed on the phantom with the lower attention coefficient (0.6
dB/cm), most of pixels in CLA and SA correlation images exceeded the threshold
(i.e., 0.90) required to estimate displacement precisely. The higher side-lobes
levels in PW images is responsible for the lower cross correlation coefficient (de-
creased performance). This trend was reversed when elastography was performed
on the higher attenuating phantom, which was not surprising because PW sono-
grams had considerably higher sonographic SNR than either CLA or SA.

Figures 4.9ii and 4.9vi show examples of modulus elastograms recovered from
the attenuating phantoms with CLA, SA, and PW imaging. The plaque was
visible in all modulus elastograms as a localized region of low elasticity. Table
2 summarizes the mean modulus recovered from the plaque and vessel wall for
each phantom, which demonstrates that SA and PW modulus elastograms were
more accurate than those produced with CLA. Figure 10 shows bar plots of
accuracy and CN Re of the recovered modulus elastograms. CN Re of SA and
PW modulus elastograms were higher than those of CLA modulus elastograms,
which was consistent with the results of the simulation study. In general, the
contrast recovered from SA and PW modulus elastograms was better than those
obtained from CLA elastogram.
 69

Figure 4.10: (left) Contrast recovery error, and (right) contrast-to-noise ratio of modu-
lus obtained with CLAI, SAI, PW. Showing experimental results obtained from vessel
phantoms with attenuation coefficient 0.6 dB/cm (phantom # 1), and 3.5 dB/cm (phan-
tom #2).

Table 4.2: Mean and standard deviation of Youngs modulus of the plaques and vessel
walls. These values were estimated from the recovered elastograms recovered with the
three imaging approaches. We also report the values from independent mechanical
testing of representative samples.

PVA % Al2 O3 % SiC % Water 18 M Thermal Cyles

Phantom #1: Plaque 8 0 3 89% 2
Phantom #1: Vess wall 8 0 1 91% 5
Phantom #2: Plaque 8 5 2 85% 2
Phantom #2: Vess wall 8 5 0 87% 5

4.6 Discussion

In this study, we compared the performance of sparse-array elastograms (displace-

ments, strain, and modulus) to those measured with conventional linear array, and
plane-wave imaging. The primary findings of this study were as follows: (a) sparse
and conventional linear arrays produced the narrowest and broadest ultrasound
beams, respectively (Fig 4.2); (b) increasing modulus contrast and attenuation
coefficient reduced the accuracy of axial and lateral displacement elastograms
 70

measured with CLA and SA imaging, but marginally affected those produced
with PW imaging (Fig. 4.5); (c) all imaging methods produced stable modulus
elastograms (Figs. 4.6 and 4.9) whose variance increased linearly with raising at-
tenuation (Fig. 4.7); (d) the accuracy of modulus elastograms varied from 1824%
(CLA), 24% (SA), and 34% (PW).

Lateral sampling frequency of the ultrasound echo frames influenced the preci-
sion of lateral displacements (Fig. 4.3), which is congruent with results reported in
Konofagou and Ophir [Konofagou and Ophir, 1998]. Sampling frequencies higher
than 52 lines/mm had little effect on performance (not shown), because there is a
limit as to how much lateral sampling can improve precision [Korukonda and Doy-
ley, 2011a; Konofagou and Ophir, 1998]. Besides lateral sampling frequency, the
transmit and receive apertures also impact the precision of lateral displacements.
More specifically, larger apertures produce tighter beams than smaller ones that
improve precision, but at the expense of side-lobes [Karaman et al., 1995] (Fig.
4.2). Conventional linear array imaging produced the widest beam (beam-width
of 0.31 mm) because a smaller aperture (11 mm transmission and 5 mm reception)
was used to minimize side-lobes. The larger apertures employed in SA and PW
imaging (i.e., 38 mm for both transmission and reception) produced smaller lat-
eral beam-widths of 0.18 mm and 0.23 mm. Although similar size apertures were
used in SA and PW, SA produced the tighter beam. This occurred because dy-
namic focusing was performed during transmission and reception in SA imaging,
whereas in PW imaging this was done only during reception.

Strain elastograms acquired from the cross-sectional plane of the carotid ar-
teries are difficult to interpret because strain is coordinate-dependent. Quanti-
tative vascular elastography (QVE) can overcome this limitation by providing
coordinate-independent mechanical parameters, as demonstrated in Figs. 3 and
4. Von Mises strain is also coordinate-invariant [Maurice et al., 2005b], but strain
alone doesnt provide any information about stress. Besides improving the vi-
 71

sual interpretation of the elastograms, quantitative vascular elastography could

(a) provide reliable modulus estimates that are required to compute stress, and
(b) characterize the composition of different vessel componentsinformation that
could be used to identify life-threatening plaques and predict their propensity to
rupture. However, successful translation of QVE to the clinic will depend on the
availability of high precision axial and lateral displacementsperforming image
reconstruction with limited displacement fields (i.e., only the axial component)
produces less accurate modulus elastograms [Hansen et al., 2013].

Figure 4.5 demonstrates the influence of noise on the performance (accuracy

and precision) of modulus elastograms. In general, noise degrades the performance
of inverse schemes and may even cause them to produce erroneous elastograms
a major challenge when solving ill-posed problems like ours. To circumvent this
problem, smoothness constraints are used to stabilize the reconstruction process
in the presence of noise [Kallel and Bertrand, 1996; Doyley et al., 2000]. However,
in vascular imaging, additional constraints (i.e., geometric) are required to pro-
duce stable modulus elastograms [Baldewsing et al., 2005; Le Floch et al., 2009;
Samani et al., 2001; Richards and Doyley, 2011]. Geometrically constrained inver-
sion methods can tolerate higher noticeable amounts of measurement noise than
those with just smoothness constraints [Richards and Doyley, 2011; Le Floch
et al., 2009]. We performed a pair-wise Welchs t-test on the results reported in
Fig. 4.7, which revealed that modulus elastograms reconstructed with the three
imaging methods were not significantly different (p > 0.05). We anticipated that
this would occur because all reconstructions were performed with the optimum
choice of regularization parameter. However, in the phantom studies where it was
difficult to chose the optimum values of regularization parameter, the results were
different. More specifically, a pair-wise statistical analysis of the results (reported
in Fig. 4.10) revealed no statistical differences (p > 0.05) in PW and SA modulus
elastograms, but CLA modulus elastograms were significantly different (p < 0.05).
 72

The main limitation of this study is that the dynamic range of moduli used in
the vessel phantoms (simulated and physical) was low. More specifically, a typical
vessel could have a 50 dB modulus dynamic range between the fibrous cap (100
kPa), fatty tissue (16 kPa), and calcium deposits (5 MPa). There are concerns
that local minima could prevent gradient-based reconstruction methods such as
those used in this study from recovering such a wide range of modulus values
[Khalil et al., 2006; Khalil et al., 2005].

4.7 Summary

In this study, we demonstrated that quantitative sparse-array vascular elastogra-

phy can produce accurate (510%) modulus elastograms of superficial organs such
as the carotid artery with high contrast-to-noise ratio (60 dB).
 73

5 The Effects of Microvessels on

Quantitative Elastography

5.1 Introduction

The QVE iterative minimization algorithm requires geometric information (spa-

tial priors) to constrain the solution space [Doyley et al., 2006; Huntzicker et al.,
2014]. This technique includes a priori knowledge of homogeneous regions within
the tissue in the minimization algorithm. Advanced plaques may contain mi-
crovasculature, which is difficult to detect with conventional IVUS imaging. How-
ever, these features have been shown to locally impact displacements and stresses
[Teng et al., 2012]. The absence of microvessel morphology as a spatial prior may
cause inversion techniques to produce erroneous estimates of modulus regions with
significant microvasculature.

Plaque microvasculature originates from the vasa vasorum and is known to

promote intraplaque hemorrhage, which may destabilize atherosclerotic plaques
[Hellings et al., 2010]. Therefore, several researchers are actively develop-
ing strategies to image these features. Contrast-enhanced intravascular ultra-
sound (CE-IVUS) imaging is an emerging technique that can visualize plaque
microvasculature by exploiting the acoustic signature of ultrasound contrast
agents (UCAs) [Granada and Feinstein, 2008]. While linear [Vavuranakis
 74

et al., 2008] and nonlinear approaches have been reported [Goertz et al., 2006b;
Maresca et al., 2013], nonlinear imaging approaches can enhance imaging sen-
sitivity and specificity by up to 30 dB [Goertz et al., 2005]. Nonlinear CE-
IVUS techniques based on second harmonic [Goertz et al., 2006b], subhar-
monic [Goertz et al., 2007], higher harmonics [Ma et al., 2014] and ultrahar-
monic [Maresca et al., 2013] imaging modes have been reported. Nonlinear
imaging techniques typically require wide bandwidth transducers that are not
commercially available (> 70% fractional bandwidth) [Maresca et al., 2013].
Therefore, prototype dual frequency/wideband transducers were used to over-
come bandwidth in previous CE-IVUS imaging studies [Goertz et al., 2006b;
Ma et al., 2014]. In a study conducted with a prototype IVUS catheter, Maresca
and colleagues demonstrated that ultraharmonic imaging can be performed
with bandlimited (60% fractional bandwidth) transducers [Maresca et al., 2013;
Maresca et al., 2014]. Shekhar et al. have previously reported subhar-
monic imaging with a commercial IVUS transducer [Eisenbrey et al., 2012;
Sridharan et al., 2013]. However, the contrast-to-tissue ratio (CTR) produced
in these studies was limited by the transducer bandwidth limitations and the
lack of contrast-specific pulsing. Subsequently, chirp-coded excitation was re-
ported to enhance the nonlinear response of UCAs [Shekhar and Doyley, 2012;
Shekhar and Doyley, 2013]. A CE-IVUS system was developed by Dr. Himanshu
Shekhar, a previous member of the parametric imaging research laboratory us-
ing a commercially available catheter to image the peripheral arteries. However,
this system cannot be used to image the coronary artery, which is the most fre-
quent site of plaque rupture. Performing elastography based on CE-IVUS imaging
with a coronary imaging catheter would allow for the detection microvessels and
enhance the accuracy of modulus and stress images [Shekhar et al., 2014a].

To corroborate this hypothesis, I collaborated with Dr. Himanshu Shekhar,

who developed an ultraharmonic CE-IVUS system that was equipped with a 40
 75

MHz coronary imaging catheter (ILab, Boston Scientific/Scimed, Natick, MA,

USA), to determine the microvessel geometry. Phantom and simulation studies
were performed to assess the performance of modulus and principal stress elas-
tograms computed when geometric information is included and excluded in the
reconstruction process.

5.2 Methods

In this study a non-rigid image registration process was used to compute displace-
ments [Richards and Doyley, 2013], and the soft-prior reconstruction method to
compute the Youngs modulus elastograms. The principal stress elastograms were
computed by combining the computed modulus elastograms with the strain elas-
tograms.

5.2.1 Displacement Estimation

The non-rigid image registration method was used to compute displacements

according to Equation 3.5. The regularization factor, , was set to 10-10 for
both simulation and phantom studies. Axial displacements measured using cross-
correlation displacement tracking were used as a trial solution for the registration.
Cross-correlation analyses were performed with 0.57 mm 0.27 radian kernels
that overlap by 80% in both the axial and lateral directions.

5.2.2 Strain and Stress Estimation

Measured radial and circumferential strains were found by taking the gradient
of the measured radial and circumferential displacements. Principal strains were
computed using the following equation,
 76

ip = max(1 , 2 ) iip = min(1 , 2 ) , (5.1)

where ip and iip and the maximum and minimum principal strains, and are the
eigenvalues of the engineering strain tensor. For this computation the FE strains
were used because they were more accurate and less noisy than the measured
strains [Richards et al., 2015]. Principal stresses were computed using the plane
strain constitutive relation shown in Equation 2.3.

5.3 Simulation Study

The goal of this study was to corroborate the hypothesis that a priori knowledge of
microvessel geometry improves the accuracy of stress and modulus elastograms.
Three vessels with wall, plaque, microvessel, and fibrous cap components were
simulated to corroborate this hypothesis.

5.3.1 Mechanical Model

The finite element (FE) method was used to compute displacements and stresses
from the simulated vessel. The wall, plaque, and cap regions of the vessel were set
to have Youngs moduli of 1000, 25, and 1250 kPa, respectively. These values were
chosen to mimic the composition of the coronary artery [Baldewsing et al., 2004].
The Youngs moduli of microvasculature is not known. Thus, microvessels with
Youngs moduli of 1, 10, and 100 Pa, were simulated. In all cases, the Poissons
ratio was set to 0.495. The simulated vessels had an outer diameter of 12 mm and
an inner diameter of 3 mm [Huntzicker et al., 2014].

Each vessel contained six microvessels, both within the wall and the simulated
plaque. The diameters of the microvessels were modeled between 200 and 800 m.
These diameters were chosen to match microvessel diameters in vivo that can be
 77

detected using contrast imaging [Kwon et al., 1998; Goertz et al., 2006a]. To
prevent rigid body motion, the phantom was embedded in a soft, compressible
medium (E = 1 Pa, = 0.0001) with dimensions three times that of the phantom
itself [Hansen et al., 2013]. The boundaries of this outer region were fixed.

COMSOL (COMSOL Inc. Burlington, MA, USA) was used to create a uniform
mesh of 27,150 linear triangular elements for this analysis. Twenty mmHg of
pressure on the inner lumen was applied to simulate small strains due to blood
flow. This pressure was chosen to effect small strains in the vessel to minimize
decorrelation effects in displacement tracking [Varghese and Ophir, 1997]. The
maximum strain observed in all three phantoms was 4%.

5.3.2 Acoustic Model

The FIELD II simulation environment [Jensen, 1991] was used to compute the
acoustic response of a 40 MHz single element IVUS transducer with a 30% frac-
tional bandwidth. The response of the pre-deformed vessel model was simulated
by randomly distributing scatterers and generating 256 radial A-lines. The scat-
terer density was set to achieve fully developed speckle (16 scatterers per wave-
length) [Wagner et al., 1983], and the speed of sound was assumed to be 1540 m/s.
Post-deformed images were generated by redistributing the point scatterers of the
predeformed vessel according to the FE displacement field. Five sets of Gaussian
white noise were added to these simulated RF images to test the robustness of
the reconstruction. This noise was applied to create a signal-to-noise ratio (SNR)
of 12 dB to match that of our IVUS system [Richards and Doyley, 2011].

Field II cannot simulate the nonlinear acoustic response of microbubbles.

Thus, ultrasound contrast-enhanced images were approximated by modeling the
contrast-filled microvessels as highly echogenic regions. The amplitude of the scat-
terers within the microvessels was chosen to be 5 times that of the main vessel.
 78

This gives CTR ratios comparable to the experimental results. The CTR was
computed as follows,

Ic
CT R = 20log10 , (5.2)
It
where I c is the contrast intensity and I t is the residual tissue intensity in the
log-compressed ultra-harmonic images [Maresca et al., 2013; Goertz et al., 2005;
Goertz et al., 2006b]. Intensities for the contrast and suppressed tissue regions
were taken at the same radial distance [Maresca et al., 2013]. The regions of
interest (ROIs) were segmented manually in MATLAB.

5.3.3 Modulus Reconstruction

All modulus elastograms were reconstructed from an initial guess of 500 kPa for
the wall and cap regions, and 50 kPa for the plaque region. The wide range of
modulus values within the vessel necessitates the use of separate initial guesses,
as using a homogeneous initial guess can cause the reconstruction algorithm to
converge on local minima. We chose an initial guess of one order of magnitude less
than the vessel wall as a conservative estimate of contrast between the wall and
the plaque. Reconstructions were performed both with and without geometric
information. When used, the geometry was included as additional a priori re-
gions in the reconstruction procedure. All reconstructions were performed using
a uniform FE mesh. This mesh consisted of 8,999 linear triangular elements. We
applied 20 mmHg of pressure to the inner lumen, and eight consecutive nodes were
fixed along the outer boundary to prevent rigid body motion [Baldewsing et al.,
2004]. The regularization parameter, , was set to 10-10 for all reconstructions.
Image reconstruction was halted when the functional changed by less than 1%.
This typically took between 8-10 iterations.
 79

5.4 Phantom Study

5.4.1 Phantom Fabrication

We fabricated two vessel phantoms using a suspension of 10% polyvinyl alcohol

cryogel (PVAc) (Elvanol 71-30, Dupont, Wilmington, DE, USA) with 3% and 6%
cornstarch in the wall and plaque regions respectively for additional scattering.
Microvessels were represented by lengths of plastic monofilament (Zebco, Tulsa,
OK, USA) with a diameter of 750 m suspended in tension along the length of
the phantom.

5.4.2 Elasticity Imaging

We used a commercially available IVUS ultrasound scanner (ILab, Boston Sci-

entific/Scimed, Natick, MA, USA) and a 40 MHz Atlantis Pro imaging catheter
(Boston Scientific, Natick, MA, USA) with a 30% fractional bandwidth for all
acquisitions. All RF echo frames were digitized to 10 bits at a sampling frequency
of 200 MHz using a PCI bus and a data acquisition card (Compuscope 14200-
1GB; Gage Applied, Lockport, IL, USA) [Richards and Doyley, 2013]. Phantoms
were submerged in a bath of deionized water. The percentage gas saturation was
70%, as measured by a dissolved oxygen sensor (model MW-600, Milwaukee In-
struments, Rocky Mount, NC, USA). Static pressure on the inner lumen of the
phantom was controlled by a water column. This was monitored using a pressure
wire (Millar Instruments Mikro-Cath, Houston, TX, USA ). Column heights were
set to achieve a pressure differential of 2.5 mmHg. For each phantom, 5 sequential
sets of IVUS images at high and low pressure were collected at the same location
to measure variance.
 80

5.4.3 Ultraharmonic Imaging

We used the experimental setup shown in Fig. 5.1a to generate contrast-enhanced

ultrasound images (ultraharmonic). We used a breakout box (prototype, Boston
Scientific, Natick, MA, USA) to transmit custom, chirp-coded transmission pulses.
We created these pulses using Benchlink ProTM (Agilent, Santa Clara, CA, USA).
Each pulse had a center frequency of 30 MHz, with a 10% fractional bandwidth
[Shekhar et al., 2014b]. Pulse sequences were downloaded to arbitrary function

b
0

5
a

Spectral power [dB]

10

15

20

25
10 20 30 40 50
Frequency [MHz]
c
2.25 mm

Peak negative pressure (MPa)

1.5
3.5 mm

0.5

0
0 100 200 300 400 500 600
V [mV ]
input pp

d 1 Number weighted
Volume weighted
Count/ml (normalized)

0.8

0.6

0.4

0.2

0
1 1.5 2 2.5 3 3.5 4
Microbubble diameter [m]

Figure 5.1: (a) A schematic of the CE-IVUS system used in the experimental study;
(b) The spectral power of the IVUS transducer; (c) the peak negative pressure of the
IVUS transducer at depths of 2.25 and 3.5 mm; (d) the number weighted and volume
weighted microbubble distribution

generators (models 33210A and 81150A, Agilent) via a GPIB interface. A linear
pulse amplifier (MR 5000 DA, ENI, Rochester, NY, USA) increased the signal by
43 dB. The excitation pulse sequence was relayed to the IVUS transducer through
a diplexer (RDX-6, Ritec Inc., Warwick, RI, USA). We used a pulse repetition
frequency of 10.8 kHz to produce standard IVUS frame rates (30 frames/second).
 81

We used a calibrated broadband hydrophone (HL-0085, Onda Corporation, Sun-

nyvale, CA, USA) to characterize the IVUS transducer used in this study. The
impulse response, bandwidth, and the on-axis peak output pressures were mea-
sured (Fig. 1(b,c). The peak pressures on-axis at 2.5 mm and 3.5 mm from
the catheter were 1.4 and 0.52 MPa, and the center frequency is 37 Hz with
a 49% fractional bandwidth. Both the 30 MHz fundamental and the 45 MHz
ultraharmonic signals were within this band. We fired two inverted pulses sep-
arated by 20 s, and used a 70% Tukey window for gating to reduce distortion
and sidelobes in the excitation pulse [Maresca et al., 2012]. A 12th order But-
terworth filter with 3 dB cutoff points of 41 and 49 MHz isolated the 45 MHz
ultraharmonic response to obtain the final contrast-enhanced images using a 12th
order Butterworth filter with 3 dB cutoff points of 41 and 49 MHz. We used
Targestar-P-HF microbubbles (Targeson Inc. San Diego, CA, USA) as the ultra-
sound contrast agent (UCA). Targestar-P-HF contains decafluorobutane gas and
its shell is primarily composed of phospholipid and polyethylene glycol fatty acid
[Shekhar et al., 2014c]. Targestar-P-HF is especially designed for high frequency
nonlinear imaging its size distribution is smaller than conventional Targetsar-P.
We measured the size distribution and concentration of Targestar-PHF with a
Coulter counter (Multisizer IV, Beckman Coulter Inc., Fullerton, CA, USA) that
used a 50 m aperture, with 300 bins between 1-15 m (Fig. 1d). We used a
syringe to inject the contrast agent into the microvessels. Targestar-P-HF comes
in a solution diluted to 1.9 109 microbubbles/ml. This solution was further
diluted in deionized water to a ratio of 1:200.

After acquisition, the ultraharmonic RF images were registered with the elas-
tography RF images. This was achieved by rotating and shifting the ultraharmonic
images to align the centers of the main lumen and the microvessel features with
the elastography acquisitions.
 82

5.5 Results

5.5.1 Simulation Study

Figure 5.2a shows the geometry used for the simulation study. Figure 5.2(b,c)
show the ground truth and measured radial strain for the simulated phantom
with microvessel moduli of 10 Pa. Figure 5.2(d,e) shows the ground truth and
measured circumferential strains. The largest strains occurred around the largest
microvessel as a result of its compression.

b c
1
6 6

4 4 0

2 2
1
mm

0 0
2
a 2 2
6 4 4 3
4 6 6
4
2 5 0 5 5 0 5
mm mm
mm

0
d e
2 4
6 6
4
4 4 3
6
2 2
5 0 5 2
mm

mm 0 0
1
2 2

4 4 0

6 6
1
5 0 5 5 0 5
mm mm

Figure 5.2: (a) The geometry of the simulated vessel phantom; (b) The ground truth
radial strain elastogram; (c) The measured radial strain elastogram from FIELD II
simulations; (d) The ground truth circumferential strain elastogram; (e) The measured
circumferential strain elastogram from FIELD II simulations. All strains are given as
percentages

Figure 5.3(a,b) shows the simulated microvessels (dB). The CTR between the
microvessel and background regions ranged from 6.7 dB to 16.5 dB. In this simula-
 83

tion, all microvessels were visible, although brightness decreased with the distance
from the main lumen. Figure 3b shows the segmentation overlaid in white on the
B-mode frame. Figure 5.3(c-e) shows the ground truth and reconstructed Youngs
modulus elastograms (Pa). Figure 5.3(c,d) had visibly similar stiffnesses for the
wall and plaque. Figure 5.3e shows a visibly softer plaque when microvessel ge-
ometry is not used. Additionally, the largest microvessel was visible, although it
appeared larger and stiffer than the ground truth. Figure 5.3(f-h) shows the max-
imum principal stress elastograms (Pa). The stresses within the cap were over an
order of magnitude larger than those in the rest of the phantom. Figure 5.3(i-k)
shows close ups of the maximum principal stress in the plaque cap (Pa). Each
image depicts the vessel with a microvessel Youngs modulus of 10 Pa.
 84

Figure 5.3: (a,b) The contrast images of the simulated phantom using FIELD II. Mi-
crovessel CTRs range from 6.7-16.5dB. The manual segmentation is overlayed in white;
(c-e) Youngs modulus elastograms for the simulated phantom with 10 Pa microvessels
(Pa) for the ground truth, reconstructed elastogram with microvessel geometry (mv)
and reconstructed elastogram without microvessel geometry (no mv); (f-h) The maxi-
mum principal stress elastograms for simulated phantom with 10 Pa microvessels (Pa);
(i-k) Close up images of maximum principal stresses in the plaque cap (Pa) correspond-
ing to the box in Fig. 5.3 f. The paths between the two sets of arrows in Fig. 5.3 g
define the profiles in Fig. 5.4

Figure 5.4(a,c,e) shows radial strain, circumferential strain and principal stress
profiles through the left path depicted in Fig. 5.3, while Fig. 5.4(b,d,f) shows
 85

radial strain, circumferential strain and principal stress profiles through the right
path passing through the simulated plaque. The measured strain profiles in Fig.
5.4(a,c) followed the same radial decay as the computed and ground truth strains.
The microvessel had only a minor effect on the magnitude. The measured strain
profiles in Fig. 5.4(b,d) was not able to detect the large compressive strains
within the microvessel. However, these trends were detected in the computed
strains when the microvessel geometry was included. In Fig. 4(e,f), the computed
principal stresses showed the same trend as the ground truth. However, the peak
stress in Fig. 5.4f was larger than the ground truth when the microvessel geometry
was not included. All plots are for the simulated phantom with 10 Pa microvessels.
 86

a b
0 5

0.2
0
Strain

Strain
0.4
Ground truth
5
0.6 Comp. with MV Geometry
Comp. without MV Geometry
Measured
0.8 10
7 6 5 4 3 2 7 6 5 4 3 2 1 0
mm mm
c d
0.8 2

0.6 0
Strain

Strain
0.4 2
Ground truth
0.2 4 Comp. with MV Geometry
Comp. without MV Geometry
Measured
0 6
7 6 5 4 3 2 7 6 5 4 3 2 1 0
mm mm
e f
5 20
Ground truth
4 15 Comp. with MV Geometry
Comp. without MV Geometry
Stress Pa

Stress Pa

3 10

2 5

1 0

0 5
7 6 5 4 3 2 7 6 5 4 3 2 1 0
mm mm

Figure 5.4: (a,c,e) The radial strain, circumferential strain and principal stress profiles
through the left path depicted in Fig. 5.3 (b,d,f); The radial strain, circumferential
strain and principal stress profiles through the right path passing through the simulated
plaque. Plots a-d show the ground truth, computed strain with microvessel geometry,
computed strain without microvessel geometry, and the measured strain. Plots e-f show
the ground truth, computed stress with microvessel geometry, and the computed stress
without microvessel geometry.

Each plot in Fig. 5.5 shows reconstructed values within each material region
when microvessel stiffnesses of 1, 10, and 100 Pa were used. In all elastograms,
error between the five trials was under 1%, error bars were not discernible. Fig-
ure 5.5(a,c,e) plots the mean Youngs moduli in the vessel wall, plaque and cap
regions. Both elastograms underestimated the wall and overestimated the cap.
Figure 5.5g plots the mean maximum principal stresses within the cap. The true
 87

stress was similar to those reconstructed when the geometry was known. In these
stress images, there was less than 5% variation between all three microvessel stiff-
nesses. Figure 5.5(b,d,f,h) plots the errors in Fig. 5.5(a,c,e,g) with respect to
the ground truth. The peak stress error was less than 2% for all phantoms when
the microvessel geometry was known, while it was greater than 15% when it was
not. The error in both elastograms in Fig. 5.4b within the vessel wall were under
10%, whereas only the reconstructed elastogram with geometric information of
the microvessels was under 10% in the cap. Both elastograms had above 10%
error within the plaque region.
 88

a b

Modulus value kPa

Modulus error kPa

1100 10

1000 5

900 0
0 1 2 0 1 2
10 10 10 10 10 10
c d
Modulus value kPa

Modulus error kPa

80
40 60

40
20
20

0 0
0 1 2 0 1 2
10 10 10 10 10 10
e f
Modulus value kPa

Modulus error kPa

2000 40

20
1500

0
1000
0 1 2 0 1 2
10 10 10 10 10 10
g h
30 30
Stress value kPa

Stress error kPa

Ground Truth
Recon with MVS
Recon without MVS
20
20
10

0
10
0 1 2 0 1 2
10 10 10 10 10 10
Pa Pa

Figure 5.5: The quantitative modulus and stress results for the elastographic data in the
simulation study vs. microvessel Youngs modulus. (a-c) Mean Youngs modulus in the
vessel wall, plaque, and cap regions for both reconstruction types; (d) Mean maximum
principal strain in the plaque cap for both reconstruction types; (e-h) Reconstruction
errors corresponding to plots (a-d). Units are in kPa.
 89

5.5.2 Phantom Study

Figure 5.6(a,b) shows the fundamental and ultraharmonic responses to a phan-

tom with a microvessel 3.5 mm from the center of the lumen. The CTR of the
microvessel in the ultraharmonic image was 8.6 1.81 dB. Figure 5.6(c,d) shows
the fundamental and ultraharmonic responses to a phantom with a microvessel
2.2 mm from the center of the lumen. The CTR of the microvessel in the ultra-
harmonic image was 9.2 2.61 dB.

Figure 5.6: (a-b) The fundamental and ultraharmonic responses for a phantom with a
microvessel 3.5 mm from the center of the lumen; (c-d) The fundamental and ultrahar-
monic responses for a phantom with a microvessel 2.2 mm from the catheter.

Figure 5.7(a,b) shows the measured radial and circumferential strains for phan-
tom 1, which decayed radially, respectively. Figure 5.7(c,d) shows the modulus
elastograms, reconstructed with and without a priori microvessel information in
kPa. Figure 5.7(e,f) shows the maximum principal stress images both with and
without a priori microvessel geometry in kPa. The microvessel was not visible in
the elastogram without geometric information Fig. 5.7d, while it appeared as a
 90

soft region in Fig. 5.7c. The path between the arrows in Fig. 5.7a defines the
profiles shown in Fig. 5.9.

a b
0.5 2.5
0 2
5
0.5 1.5
mm

0 1 1

1.5 0.5

5 2 0

2.5 0.5
c d
60 60
5

40 40
mm

20 20

5 0 0

e f
1 1
5
0.8 0.8

0.6 0.6
mm

0 0.4 0.4

0.2 0.2

0 0
5

5 0 5 5 0 5
mm mm

Figure 5.7: The measured radial and circumferential strains for phantom 1 (a,b); The
reconstructed Youngs modulus elastograms with and without microvessel geometry for
phantom 1 (kPa) (c,d); The reconstructed maximum principal stress elastograms with
and without microvessel geometry for phantom 1 (kPa) (e,f). The arrows in c define
the profile shown in Fig. 5.9.

Figure 5.8(a,b) shows the measured radial and circumferential strains for phan-
 91

tom 2, which decayed radially. Figure 5.8(c,d) shows the reconstructed Youngs
modulus elastograms both with and without a priori microvessel geometry in
kPa. Figure 5.7(e,f) shows the maximum principal stress images both with and
without a priori microvessel geometry in kPa. The microvessel was not visible in
the elastogram without geometric information (Fig. 5.8d), while it appeared as a
soft region in Fig. 5.8c. The largest principal stress occurred between the plaque
and the main lumen for both elastograms. However, the reconstruction with mi-
crovessel geometric informations produced larger values. The path between the
arrows in Fig. 5.8e defines the profiles shown in Fig. 5.9.
 92

a b
0.5 2.5

5 0 2

0.5 1.5
mm

0 1 1

1.5 0.5
5 2 0

2.5 0.5
c d
60 60
5

40 40
mm

0
20 20

5 0 0

e f
1 1
5
0.8 0.8

0.6 0.6
mm

0 0.4 0.4

0.2 0.2

0 0
5

5 0 5 5 0 5
mm mm

Figure 5.8: The measured radial and circumferential strains for phantom 2 (a,b); The
reconstructed Youngs modulus elastograms with and without microvessel geometry for
phantom 2 (kPa) (c,d); The reconstructed maximum principal stress elastograms with
and without microvessel geometry for phantom 2 (kPa) (e,f). The arrows in c define
the profile shown in Fig. 5.9.

Figure 5.9(a,c,e) shows radial strain, circumferential strain and principal stress
profiles through the path depicted in Fig. 5.7, while Fig. 5.9(b,d,f) shows radial
strain, circumferential strain and principal stress profiles through the path de-
 93

picted in Fig. 5.8. The computed strains with microvessel geometry show peaks
at the microvessels that were not visible in the measured strains or those computed
without the microvessel geometry. There were stress peaks on either side of the
microvessel in Fig. 5.9e due to its contraction. These features are not observed in
the trace without microvessel geometry, which has a peak stress value that is 38%
smaller. Figure 5.9f shows a similar trend, although the peak stresses are larger
than in Fig. 5.9e. This due to the larger strains within the plaque. In this image,
the peak stress when microvessel geometry is included is 56% larger than when it
is omitted.

a b
1 0

0 1
Strain

Strain

1 2

Comp. with MV Geometry

2 3
Comp. without MV Geometry
Measured
3 4
7 6 5 4 3 2 1 7 6 5 4 3 2 1
mm mm
c d
3 4
Comp. with MV Geometry
Comp. without MV Geometry
2 3
Measured
Strain

Strain

1 2

0 1

1 0
7 6 5 4 3 2 1 7 6 5 4 3 2 1
mm mm
e f

Comp. with MV Geometry

1000 1000
Comp. without MV Geometry
Stress Pa

Stress Pa

500 500

0 0

7 6 5 4 3 2 1 7 6 5 4 3 2 1
mm mm

Figure 5.9: (a,c,e) The radial strain, circumferential strain and principal stress profiles
through the path depicted in Fig. 7; (b,d,f) The radial strain, circumferential strain
and principal stress profiles through the path depicted in Fig. 8. Plots a-d show
the computed strain with microvessel geometry, computed strain without microvessel
geometry, and the measured strain. Plots e-f show the computed stress with microvessel
geometry, and the computed stress without microvessel geometry.
 94

Figure 5.10 shows the quantitative Youngs modulus and peak maximum prin-
cipal stress values in the phantom study. The peak stresses were measured as the
average stress in a 10 10 pixel region around the maximum stress. Ground truth
stresses were computed for the mechanical testing results using a FE simulation.
We performed a pair-wise Welshs t-test (p > 0.05) on this data, which showed
that for all regions, the Youngs modulus results were statistically the same as
the measured values using mechanical testing. However, there was a statistically
significant difference between the peak stress elastogram without microvessel ge-
ometry and the measured values. The peak stress without microvessel geometry
showed 38% error, while the stress with microvessel geometry showed 3% error.

Vessel Wall Modulus Inclusion Modulus Maximum principal stress

50 12 1.4
Mechanical testing
Reconstruction with MVs
10 1.2
Reconstruction without MVs
40
1
8
30
0.8
KPa

KPa

6 KPa
0.6
20
4
0.4
10
2 0.2

0 0 0
Phantom 1 Phantom 2 Phantom 1 Phantom 2 Phantom 1 Phantom 2

Figure 5.10: The quantitative results for the elastographic data in phantoms 1 and 2,
respectively. Error bars represent the standard deviation between experiments. Units
are in kPa.

5.6 Discussion

In this study, we examined how microvessels affect our ability to reconstruct vas-
cular moduli and stresses. The primary findings of this study were as follows:
(a) Failing to account for microvessels within soft plaques produced peak stress
error of over 30% in reconstructed stresses; (b) Failing to account for microvessels
within soft plaques with stiff caps cause modulus errors of 20% within the cap,
 95

and 20% error in the peak stress; (c) The peak stress and cap modulus errors were
reduced to under 10% when geometric information was included; (d) 40 MHz ul-
traharmonic contrast-enhanced imaging correctly located the microvessels within
both phantoms.

Previous works have shown that peak stresses within vessels will occur between
soft plaques and the main lumen [Cheng et al., 1993; Richards et al., 2015]. Large
microvessels in this area may increase these stresses by causing larger local strains
[Teng et al., 2012]. We expected that this would have a negative impact on
the accuracy of our reconstruction for two reasons: First, vessel regions that
are not segmented a priori may be averaged into the background region. This
can cause them to appear larger than they are with reduced contrast, or not
appear at all [Yalavarthy et al., 2007b; Hansen et al., 2013]. This effect reduces
the reconstructed modulus contrast [McGarry et al., 2013]; the extent of which
depends on the relative size and quantity of the microvessels. Second, the failure
to reconstruct accurate modulus values within the microvessels will also reduce
the magnitude of the computed strains. Both the modulus and strain errors can
propagate through the entire phantom to compensate for the missing features.

There has been little investigation as to the effects of intimal microvessels on

arterial stresses. This is partially because the pressure within the microvessels
cannot be measured [Simon et al., 1973; Teng et al., 2012]. Current FE studies
have modeled these structures as hollow tubes with internal pressures of 10 mmHg,
while applying systolic pressure on the main lumen [Simon et al., 1973; Teng
et al., 2012; Lu et al., 2015]. Under these conditions, the microvessels have been
shown to contract. Teng et al. have also shown that there is no appreciable
difference in the mechanical behavior of microvessels pressurized at 10 mmHg vs.
5 mmHg [Teng et al., 2012]. Modeling the microvessels as soft regions with a
Poissons ration of 0.1, we also observe compression (Fig. 5.2). By modeling
the microvessels as solids rather than voids, we are also able to use the same
 96

FE mesh for reconstructed elastograms with and without microvessel geometry.

This eliminates any reconstruction differences due to mesh density. Local strains
adjacent to the microvessels increased by up to a factor of two compared with the
strains of equal radial distance further away from the microvessels in the ground
truth simulations. The magnitude of this increase was consistent with Teng et al.
[Teng et al., 2012]. We found that the local stresses on both sides of the plaque
microvessels were 40% larger than those in the rest of the cap (Fig. 5.3).

Figure 5.4 highlights why these large stresses could not be recovered without
prior knowledge of the microvessel locations. Figure 5.4(b,d) shows that the large
strains due the central microvessel in the simulated plaque could be measured in
the radial direction, but not in the circumferential direction. Because these strains
are highly localized, the accuracy of the radial strain benefited from the higher
imaging resolution in that direction.

The effects of varying the microvessel stiffness in Fig. 5.5 supported the claim
that similar stress and modulus results would be obtained over a range of mi-
crovessel properties. The ground truth stress in the plaque cap varied by less
than a percent over the range of microvessels tested, and the variance in stress
elastograms was under 5%. Figure 5.5 also shows that when microvessel geometry
was not included in the reconstruction, the plaque region appeared softer and the
cap region appeared harder than the ground truth, in line with the averaging ef-
fects previous studies have suggested. Comparing Fig. 5.5(g,h), the relative error
between the cap modulus (Fig. 5.5g) and cap stress (Fig. 5.5h) were within 5%
when the microvessel information was not included. However, there is a larger
difference when the microvessel geometry was included. This is because the area
of the largest microvessel was underestimated by 16% in the manual segmentation.
As Fig. 5.2 shows, microvessel size correlates with strain, and strain is directly
proportional to stress.

The experimental study was performed with Targestar-P-HF, which has a

 97

smaller mean size than clinical contrast agents. Nonlinear CE-IVUS imaging with
clinical UCAs becomes progressively challenging as the frequency is increased,
because they are designed to resonate at low frequencies (<10 MHz). However,
previous CE-IVUS imaging studies have demonstrated that ultraharmonic imag-
ing is feasible with Definity (Bristol-Myers Squibb Inc., New York, NY, USA) at
high frequencies, especially when its size distribution is modified [Maresca et al.,
2014]. Therefore we expect that the results of this study could be reproduced
with Definity and other clinically available phospholipid-encapsulated UCAs.

The stiff plaque cap could not be modeled experimentally, due to fabrication
constraints. Thus, the simulation study better mimicked the composition of the
coronary artery, while this phantom study highlighted the ability to combine 40
MHz contrast-enhanced imaging with QVE. The absence of the hard cap allowed
for more pronounced stress concentrations between the plaque and the lumen due
to the increased local strain (Figs. 5.7 and 5.8). The softer vessel wall, as com-
pared with simulation, also increased the mechanical impact of the microvessels.
This same trend was observed in Figs. 5.7 and 5.8. We previously reported that,
when using the same reconstruction regularization, phantom modulus elastograms
express more modulus variation than simulation, which could be observed in these
results as well. This was due to the increased noise in the system [Huntzicker et al.,
2014]. Scaling the regularization parameter with displacement could correct for
this issue, however this parameter variation would obfuscate comparisons between
the simulation and phantom studies. Unlike the simulation results, there was no
statistical difference in the plaque modulus of phantom 2 between elastograms
(Fig. 5.10). However, the cap stress is statistically different. Figure 5.9 sug-
gests that this occurred because the large strains around the microvessel could
not be recovered accurately when no microvessel information was available in the
reconstructionbut these local effects were not large enough to influence the mod-
ulus value of the entire region. This is also shown in Fig. 5.10, where the RMSD
 98

between the measured strains and the computed strains with microvessel geometry
is larger than the RMSD between the measured strains and the computed strains
without microvessel geometry. However, the tracked displacements still recovered
the effects of the microvessels as shown in Fig. 5.10. The RMSD between the
measured displacements and computed displacements with microvessel geometry
is lower than that without it. Thus, the simulation and phantom studies show
that the peak stresses may not be recoverable when the microvessel geometry is
not known. This is due to a combination of errors in the Youngs modulus because
of the inability to track local displacement variation, and errors in the peak recon-
structed strains because of incorrect geometry in the finite element model. The
peak stresses may be overestimated or underestimated depending on the specific
geometry and material properties of the vessel.

Researchers are currently developing custom transducers and instrumentation

to image plaque microvasulature [Ma et al., 2014]. Continued advances in this
direction will enable the detection of microvessels with enhanced spatial and con-
trast resolution, which should further improve our ability to accurately determine
spatial priors for stress reconstruction. In the current study, two ultrasound acqui-
sitions were needed to measure the ultraharmonic and standard vessel RF images.
This setup can be optimized by using a pulsatile pump to simulate blood flow, and
a syringe pump to inject contrast into the vessel. With such a setup, elastography
and contrast imaging could be measured simultaneously.

5.7 Summary

In this chapter, it was shown that knowledge of the microvessel locations within
the artery was necessary to reconstruct accurate stress elastograms (< 5% error).
However, failure to include this information led to inaccurate stress and modulus
elastograms (images with error on the order of 35%).
 99

6 Material Nonlinearity in
Vascular Elastography

6.1 Introduction

The FE algorithm used in QVE requires a predefined elastic model to com-

pute displacements. Many studies assume material linearity, which dictates that
stresses increase linearly with strain [Richards et al., 2015; Le Floch et al., 2009;
Karimi et al., 2008; Baldewsing et al., 2006]. This model is useful for two rea-
sons. First, it is computationally efficient, and second, the material properties
are described by just two parameters. However, vascular tissue only behaves lin-
early over small strain (< 5%). Mechanical testing has shown that human arteries
act as hyperelastic materials under large strain [Holzapfel and Weizsacker, 1998;
Walsh et al., 2014]. Thus, hyperelastic materials appear stiffer under high pres-
sure. Accurately modeling these nonlinear effects is critical when computing the
stresses within plaques. Plaques are most likely to rupture at systole [Davies,
1996], where the vessel experiences strains in excess of 300% in the plaque cap
[Cheng et al., 1993]. Thus, in this regime, the nonlinear effects cannot be ig-
nored. Traditional cross-correlation-based displacement tracking methods break
down when strains exceed 5% due to decorrelation between RF frames [Varghese
and Ophir, 1997]. Thus, displacements must be accumulated by incrementally
 100

measuring the small motion between successive frames and summing each set
of displacement fields in the material reference frame [McCormick et al., 2012;
Haupt, 2002] The vessel will experience gross distortion between the diastolic and
systolic phases, preventing the simple summation of the tracked displacements.

Little research has been done to determine the feasibility of reconstructing

hyperelastic properties within vessels, with only a single publication reconstruct-
ing the stresses within atherosclerotic plaques [Karimi et al., 2008]. This study
used an exponential strain energy density function to model elasticity, and com-
puted the associated mechanical properties using a stochastic genetic algorithm
approach. They demonstrated that the mechanical properties of the vessel wall
and plaque cap could be reconstructed with under 5% error when ideal strain im-
ages were used as inputs, having been corrupted with white Gaussian noise. To our
knowledge there have not been any reported studies that evaluate the performance
of nonlinear reconstructions assuming realistic noise models, or investigating the
errors incurred by assuming linearity when measuring stresses in hyperelastic ma-
terials. However, nonlinear tissue reconstruction has been performed on other
organs. Oberai et al. completed a study using an exponential strain energy den-
sity function to reconstruct the shear modulus in breast tissue [Oberai et al., 2009;
Goenezen et al., 2012]. Similar breast studies have been computed by other re-
searchers

In this chapter, the QVE imaging procedure was extended to measure displace-
ments with strains of up to 50%, and to account for material nonlinearity in both
forward and inverse modeling. We hypothesized that hyperelastic models are nec-
essary to reconstruct the peak cumulative stresses within the coronary artery. To
corroborate this hypothesis, the algorithm performance (modulus and stress re-
covery) was evaluated by imaging phantoms with nonlinear mechanical properties
and performing stress reconstructions with both linear and hyperelastic models.
These analyses were performed using both simulated and PVAc phantoms.
 101

6.2 Methods

6.2.1 Mechanical Model

An incompressible exponential strain energy density function, W, was used to

describe the mechanical nonlinearity observed in human arteries. The energy
density function relates stress accumulation with strain, typically as a function
of the right Cauchy-Green deformation tensor. To compute the Cauchy-Green
deformation tensor, the coordinate-independent principal strains were first defined
using Equation 5.1. Further, the principal stretch tensor, F, was defined as the
principal strain tensor plus the Kronecker delta. The three dimensional right
Cauchy-Green deformation tensor, C, was then computed as C =F T F, as shown
in Equation 6.1:

C1 0 0 21 0 0 (p1 + 1)2 0 0

p
C = 0 C2 0 = 0 22 0 = 2 . (6.1)

0 (2 + 1) 0

2 p 2
0 0 C3 0 0 3 0 0 (3 + 1)

In a two-dimensional model, 23 is assumed to be 1 to indicate zero strain in

the out-of-plain direction. Delfino proposed an exponential strain energy density
function for vascular tissue from measured stress-stretch curves [Delfino, 1996].
We modified this function, shown in Equation 6.3, to be a function of a single
parameter, K, which relates to the Youngs modulus, E, through Equation 6.2, for
small strains.

E = 4K. (6.2)

This parameter will be referred to as the K-modulus. This definition allows for the
direct comparisons between linear and nonlinear reconstructions. The modified
 102

strain energy density function is defined as

W = K e2(J1 3) 1 ,


(6.3)

where J 1 is the deviatoric component of the the first invariant of the right Cauchy-
Green deformation tensor. The Cauchy stress, most often used in vessel modeling,
can be computed from the energy density function by the following set of equa-
tions,

= J 1 F , (6.4)

where is the second Piola-Kirchhoff stress defined as

W J1
=2 = 4Ke2(J1 3) . (6.5)
C C
The first partial derivative of J1 with respect to C can be computed as shown in
Equation 6.6, with respect to the principal stretches.

J1C I C 1 I C
= (I3C )1/3 1 I1C (I3C )4/3 3 , (6.6)
C C 3 C
I1C I3C
where C
and C
are defined as

1 C
2
1 C1



I1C 1 I3C 1
C
=
C
=
.
(6.7)
0 0


0 0

0 0

The second Piola-Kirchhoff stress can be used to generate the stiffness matrix
using the following equation
 103

2 J1 2(J1 3) J1 J1
D=2 = 8e2(J1 3) + 16e , (6.8)
C C 2 C C
2 J1 2 J3
where C 2
and C 2
are defined as

T
2 J1 2

1/3 I1 4 7/3 I3 I3
2
= I3 2
+ I1 I3
C C 9 C C
" T T #
2 I3 I3
 
1 4/3 I1 I3 I1
I3 + I1 +
3 C C C 2 C C

T
2 J3 1 1/2 2 I3 1 3/2 I3

I3
= I I , (6.9)
C 2 2 3 C 2 4 3 C C
2 I1 2 IJ3
and C 2
and C 2
are defined as

2 I1
=0
C 2

0 1 C2 0 0 0

1 0 C1 0 0 0


2

I3 C2 C1 0
0 0 0
= . (6.10)
C 2 0 0 0 21
0 0


0 C21

0 0 0 0

C2
0 0 0 0 0 2

However, unlike the linear case, Equation 6.8 depends on strain (as a function of
displacement). Thus, the stiffness matrix K, shown in Equation 3.2 will also be
dependent on displacement. This means that Hookes law (Equation 3.1) cannot
be solved directly (i.e. inverting K ). We deal with this dependency by solving the
inversion using Newtons method [De Borst et al., 2012].
 104

Newtons Method Implementation

Newtons method can be used to compute the displacement field by minimizing

the residual between the computed forces in the FE model, f int =Ku, and the
imposed forces, f ext . The displacement field is incrementally built up by multiple
linearizations of the constitutive relation for each strain distribution. For this
process, fext is divided into n force steps (f t1 tn
ext through f ext ). The number of steps

must be chosen to limit the nonlinear effects.

An initial zero-strain state of the Cauchy-Green deformation tensor, C 0 , was

assumed, which was used to computed K 0 . From this assumption, we can compute
a displacement field, u 1 , as follows:

u1 = (K 0 )1 fext
t1
. (6.11)

C 1 and K 1 are then computed from u 1 according to Equations 6.1 and 3.2. These
values are used to update the internal force vector,

1
fint = K 1 u1 . (6.12)

This process iterates as generalized below,

um = um1 + (K n1 )1 (fext
t1 n1
fint ) (6.13)

n
fint = f n1 + K n un .

The iterations continue until the residual between f int and f t1

ext falls below a spec-
n n+1
ified tolerance. At this point, fext is incremented to fext . Figure 6.1 depicts this
process graphically.
 105

Figure 6.1: Graphical representation of a Newtons method inversion [De Borst et al.,
2012]

6.2.2 Principal Cauchy Stress and Effective Youngs Mod-

ulus Computation

The maximum principal stresses were computed using the plane stress and hyper-
elastic constitutive relations shown in Equations 2.2 and 6.8, using the definition
for principal strains in Equation 5.1 and the computed Youngs moduli and K-
moduli respectively.

It is difficult to compare the K-modulus with the linear Youngs modulus in

the large strain regime, because the K-modulus does not represent an absolute
stiffness, but a varying stiffness depending on the strain distribution. Thus, we
define the effective Youngs modulus (EYM) as the local, instantaneous Youngs
modulus distribution for a given strain state. This value was found at every FE
node by determining the equivalent Youngs modulus necessary to achieve the
same stress/strain state. This was computed through the following equation:

4pi (1 2 )
EY M = i , (6.14)
p + 0.5iip
 106

where pi is the maximum principal stress. The EYM for the linear reconstruction
is simply the Youngs modulus, because stiffness is independent from strain or
pressure in this model.

6.2.3 Displacement Estimation

Displacement maps were generated by aggregating individual sets of displacements

between successive pairs of RF images. Frames were chosen every 4 mmHg in-
crease in intraluminal pressure. This limited the strain between each pair to 5%.
After tracking, each set of displacements was mapped onto the pre-deformed ge-
ometry and summed together. The non-rigid image registration method was used
to compute each set of displacements according to Equation 3.5. The regulariza-
tion factor, , was set to 109 for both simulation and phantom studies. Axial
displacements measured using cross-correlation displacement tracking were used
as a trial solution for the registration. Cross-correlation analyses were performed
with 0.57 mm 0.27 radian kernels that overlap by 80% in both the axial and
lateral directions.

6.3 Simulation Study

6.3.1 Phantom Design

A single phantom was generated using the hyperelastic FE solver (inner radius =
1.5 mm, outer radius = 6mm). The K-moduli were set to 1541 and 5537 for plaque
and wall regions of the vessel, respectively. The plaque was embedded within the
vessel wall 0.2 mm from the inner lumen. These stiffness parameters were set to
mimic the material properties of PVAc. The vessel was pressurized between 0-56
mmHg in increments of 4 mmHg. This model used 6656 constant strain triangle
(CST) elements created using COMSOL (COMSOL Inc. Burlington, MA, USA).
 107

To prevent rigid body motion, an arc of eight nodes on the outer boundary was
fixed at the measured values [Baldewsing et al., 2004]. The FE algorithm used 10
Newton steps, with a convergence tolerance of 0.01. These values were chosen such
that convergence was reached between one and two iterations per step. Increasing
the number of steps had a negligible effect on the displacements.

6.3.2 Acoustic Model

We used the FIELD II simulation environment [Jensen, 1991] to generate the

acoustic response of a 40 MHz single element IVUS transducer with a 30% frac-
tional bandwidth. We simulated the response of the pre-deformed vessel model by
randomly distributing scatterers and generating 256 radial A-lines. The scatterer
density was set to achieve fully developed speckle (16 scatterers per wavelength)
[Wagner et al., 1983], and the speed of sound was assumed to be 1540 m/s. Post-
deformed images were created by interpolating the scatterer locations onto the
model displacements and re-simulating the acoustic response. Gaussian white
noise was added to these images to create a signal-to-noise Ratio (SNR) of 12 dB
to match that of our IVUS system [Richards and Doyley, 2011].

6.3.3 Modulus and Stress Reconstruction

The K-modulus was reconstructed for the nonlinear model, and 14 E was recon-
structed for the linear model. This dictates that both algorithms would com-
pute the same value for infinitesimal strain, allowing for better comparison. All
modulus elastograms were reconstructed from an initial guess of 5 kPa. All re-
constructions were performed using a uniform FE mesh. This mesh consisted of
6626 CST elements. The boundary conditions were the same as those used in the
ground truth computation. The regularization parameter, , was set to 10-14 for
all reconstructions. The reconstructions were computed using both a plane stress
 108

linear model and the exponential hyperelastic model. Optimization was halted
when the functional changed by less than 1%. This typically took between 5-10
iterations for the linear solve, and 20-25 iterations for the nonlinear solve. The
Newton iteration parameters were the same as the ground truth computation.

6.4 Phantom Study

6.4.1 Phantom Fabrication

A vessel phantom was produced using a suspension of 10% PVAc (Elvanol 71-
30, Dupont, Wilmington, DE, USA) with 3% and 6% cornstarch in the wall and
plaque regions, respectively, for additional scattering.

6.4.2 Elasticity Imaging

We used a commercially available IVUS ultrasound scanner (ILab, Boston Sci-

entific/Scimed, Natick, MA, USA) and a 40 MHz Atlantis Pro imaging catheter
(Boston Scientific, Natick, MA, USA) with a 30% fractional bandwidth for all
acquisitions. All RF echo frames were digitized to 10 bits at a sampling frequency
of 200 MHz using a PCI bus and a data acquisition card (Compuscope 14200-
1GB; Gage Applied, Lockport, IL, USA) [Richards and Doyley, 2013]. Phantoms
were submerged in a bath of deionized water. The percentage gas saturation was
70%, as measured by a dissolved oxygen sensor (model MW-600, Milwaukee In-
struments, Rocky Mount, NC, USA). Pressure on the inner lumen of the phantom
was varied by manually raising a water column. For each experiment, 100 IVUS
rotations were collected while the pressure was dynamically varied between 0-50
mmHg. The instantaneous pressure at each rotation was collected using a pres-
sure wire (Millar Instruments Mikro-Cath, Houston, TX, USA ). Data from four
 109

successive 0-50 mmHg cycles was collected at a single location along the vessel to
measure reconstruction variance.

6.4.3 Material Testing

The hyperelastic K-moduli for the plaque and wall portions of the PVAc vessel
were verified through mechanical testing using a Landmark Servohydraulic Test
System (MTS Testing Systems, Eden Prairie, MN, USA). For this testing, four
additional cylindrical phantoms (diameter: 2 cm, height: 2 cm) were fabricated
from the same solutions as the vessel phantoms. These samples were subjected to
a uniaxial compressive strain of 40%. The resulting stress-stretch curve was fit to
Equation 6.5 using a Levenberg-Marquardt minimization.

6.5 Results

6.5.1 Simulation Study

Figure 6.2 shows the K-modulus reconstructions for the ground truth and hy-
1
perelastic reconstructions, as well as 4
E for the linear reconstruction. These
elastograms are displayed for intraluminal pressures between 8-56 mmHg. The
ground truth K-modulus does not vary with pressure, and the hyperelastic recon-
struction does not show any visible increase in stiffness. However, the modulus
of both the plaque and wall regions of the linear elastograms increase at every
pressure point.
 110

Ground Truth Hyperelastic Recon. Linear Recon. 10

4 Ground Truth Hyperelastic Recon. Linear Recon. 10
4
-6 2 -6 2

-4 -4
1.5 1.5
-2 -2

40 mmHg
8 mmHg

mm

mm
0 1 0 1

2 2
0.5 0.5
4 4
6 0 6 0
4 4
10 10
-6 2 -6 2

-4 -4
1.5 1.5
-2 -2
16 mmHg

48 mmHg
mm

mm
0 1 0 1

2 2
0.5 0.5
4 4
6 0 6 0
4 4
10 10
-6 2 -6 2

-4 -4
1.5 1.5
-2 -2
24 mmHg

56 mmHg
mm

mm
0 1 0 1

2 2
0.5 0.5
4 4
6 0 6 0
4
-4 -2 0 2 4 6 -4 -2 0 2 4 6 -4 -2 0 2 4 6
10
-6 2

-4
1.5
-2
32 mmHg

mm

0 1

2
0.5
4
6 0
-4 -2 0 2 4 6 -4 -2 0 2 4 6 -4 -2 0 2 4 6
mm mm mm

Figure 6.2: Ground truth and reconstructed hyperelastic K-moduli elastograms com-
pared against four times the reconstructed linear Youngs modulus elastograms. Recon-
structions shown for intraluminal pressures between 8-56 mmHg

Figure 6.3 shows the maximum principal Cauchy stress elastograms for the
ground truth, hyperelastic reconstruction, and linear reconstruction. These elas-
tograms are shown for intraluminal pressures between 8-56 mmHg. These elas-
tograms show stress decaying radially, with low stress in the plaque. The black
box in the top left elastogram shows the region enhanced in Fig. 6.4.
 111

Ground Truth Hyperelastic Recon. Linear Recon. Ground Truth Hyperelastic Recon. Linear Recon. 10
4
-6 4000 -6 2.5

-4 -4 2
3000
-2 -2

40 mmHg
8 mmHg

1.5
mm

mm
0 2000 0
1
2 2
1000
4 4 0.5

6 0 6 0
4
10
-6 7000 -6 3.5

-4 6000 -4 3
5000 2.5
-2 -2
16 mmHg

48 mmHg
4000 2
mm

mm
0 0
3000 1.5
2 2
2000 1
4 1000 4 0.5
6 0 6 0
4
10
-6 -6
10000 4
-4 -4
8000
-2 -2 3
24 mmHg

56 mmHg
mm

6000

mm
0 0
2
2 4000 2
1
4 2000 4
6 0 6 0
-4 -2 0 2 4 6 -4 -2 0 2 4 6 -4 -2 0 2 4 6
-6
15000
-4
-2
32 mmHg

10000
mm

0
2 5000
4
6 0
-4 -2 0 2 4 6 -4 -2 0 2 4 6 -4 -2 0 2 4 6
mm mm mm

Figure 6.3: Maximum principal Cauchy stress elastograms for the ground truth, hy-
perelastic reconstructions, and linear reconstructions. These elastograms are shown for
intraluminal pressures of 8-56 mmHg

Figure 6.4 shows enhanced images of the maximum principal Cauchy stress
elastograms shown in Fig. 6.3. These images are centered on the plaque shoulder,
following the black rectangle in Fig. 6.3. These elastograms show that the peak
stresses are located in the wall region between the lumen and the soft plaque,
and that the stresses increase with rising intraluminal pressure. The hyperelastic
reconstruction shows little deviation from the ground truth. However, the mag-
nitude of the peak stress in the linear reconstruction is visibly smaller than that
of the ground truth and hyperelastic reconstruction.
 112

Ground Truth Hyperelastic Recon. Linear Recon. Ground Truth Hyperelastic Recon. Linear Recon. 10
4
-4 4000 -4 2.5

-3.5 -3.5 2
3000
-3 -3

40 mmHg
8 mmHg

1.5
mm

mm
-2.5 2000 -2.5
1
-2 -2
1000
-1.5 -1.5 0.5

-1 0 -1 0
4
10
-4 7000 -4 3.5

-3.5 6000 -3.5 3

5000 2.5
-3 -3
16 mmHg

48 mmHg
4000 2
mm

mm
-2.5 -2.5
3000 1.5
-2 -2
2000 1
-1.5 1000 -1.5 0.5
-1 0 -1 0
4
10
-4 -4
10000 4
-3.5 -3.5
8000
-3 -3 3
24 mmHg

56 mmHg
mm

6000

mm
-2.5 -2.5
2
-2 4000 -2
1
-1.5 2000 -1.5
-1 0 -1 0
-2 0 2 -2 0 2 -2 0 2
-4
15000
-3.5
-3
32 mmHg

10000
mm

-2.5
-2 5000
-1.5
-1 0
-2 0 2 -2 0 2 -2 0 2
mm mm mm

Figure 6.4: Enhanced maximum principal Cauchy stress elastograms for the ground
truth, hyperelastic reconstructions, and linear reconstructions. The displayed regions
are defined by the black outline shown in Fig. 6.3. These elastograms are shown for
intraluminal pressures of 8-56 mmHg

Figure 6.5 shows EYM elastograms for the ground truth, hyperelastic recon-
struction, and linear reconstruction for intraluminal pressures ranging between 8-
56 mmHg. These images show that the Youngs modulus of the linear reconstruc-
tion is much closer to the hyperelastic EYM than the reconstructed K-modulus,
due to the stiffness adjustment. The black box in the top left elastogram shows
the region enhanced in Fig. 6.6.
 113

Ground Truth Hyperelastic Recon. Linear Recon. 10 4 Ground Truth Hyperelastic Recon. Linear Recon. 10
5
-6 10 -6 2.5

-4 8 -4 2
-2 -2

40 mmHg
8 mmHg

mm 6 1.5

mm
0 0
4 1
2 2
4 2 0.5
4
6 0 6 0
4 5
10 10
-6 10 -6 3

-4 8 -4 2.5

-2 -2 2
16 mmHg

48 mmHg
6
mm

mm
0 0 1.5
4
2 2 1
4 2
4 0.5
6 0 6 0
4 5
10 10
-6 15 -6 3.5

-4 -4 3

-2 10 2.5
-2
24 mmHg

56 mmHg
2
mm

mm
0 0
1.5
2 5 2
1
4 4 0.5
6 0 6 0
5 -4 -2 0 2 4 6 -4 -2 0 2 4 6 -4 -2 0 2 4 6
10
-6 2

-4
1.5
-2
32 mmHg

mm

0 1

2
0.5
4
6 0
-4 -2 0 2 4 6 -4 -2 0 2 4 6 -4 -2 0 2 4 6
mm mm mm

Figure 6.5: EYM elastograms for the ground truth, hyperelastic reconstructions, and
linear reconstructions. These elastograms are shown for intraluminal pressures of 8-56
mmHg

Figure 6.6 shows enhanced images of the EYM elastograms defined by the
region outlined in black in Fig. 6.5 over all pressures. These images show that
the region between the lumen and the plaque experiences the largest change in
stiffness with respect to the zero strain case. The linear reconstruction cannot
reproduce these variations, and thus the wall modulus is computed to be greater
than the true EYM far from the lumen, but less than the EYM of the ground
truth close to the lumen.
 114

Ground Truth Hyperelastic Recon. Linear Recon. 10

4 Ground Truth Hyperelastic Recon. Linear Recon. 10
5
-4 10 -4 2.5

-3.5 8 -3.5 2
-3 -3

40 mmHg
8 mmHg

6 1.5
mm

mm
-2.5 -2.5
4 1
-2 -2
-1.5 2 -1.5 0.5

-1 0 -1 0
4 5
10 10
-4 10 -4 3

-3.5 8 -3.5 2.5

-3 -3 2
16 mmHg

48 mmHg
6
mm

mm
-2.5 -2.5 1.5
4
-2 -2 1

-1.5 2 -1.5 0.5

-1 0 -1 0
4 5
10 10
-4 15 -4 3.5

-3.5 -3.5 3
2.5
-3 10 -3
24 mmHg

56 mmHg
2
mm

mm
-2.5 -2.5
1.5
-2 5 -2
1
-1.5 -1.5 0.5
-1 0 -1 0
5
-2 0 2 -2 0 2 -2 0 2
10
-4 2

-3.5
1.5
-3
32 mmHg

mm

-2.5 1

-2
0.5
-1.5
-1 0
-2 0 2 -2 0 2 -2 0 2
mm mm mm

Figure 6.6: Enhanced EYM elastograms for the ground truth, hyperelastic reconstruc-
tions, and linear reconstructions showing the region defined by the gray rectangle in
Fig. 6.6. These elastograms are shown for intraluminal pressures of 8-56 mmHg

Figure 6.7 quantifies the results of Fig. 6.2. Figure 6.7 a shows the K-modulus
for the ground truth and both reconstructions in the vessel wall region vs. the in-
traluminal pressure. The ground truth K-modulus does not change with pressure;
the hyperelastic reconstruction increases slightly with pressure, and the linear
reconstruction shows a distinct increase stiffness with pressure, which is approx-
imately linear. Figure 6.7 c show the error associated with Fig. 6.7 a. The
hyperelastic error is under 20% for all reconstructions, whereas the linear error
exceeds 160% at 56 mmHg. Figure 6.7 b shows the K-modulus for the ground
truth and hyperelastic reconstruction, and 4E for the linear reconstruction in the
plaque region vs. the intraluminal pressure. Both the ground truth and hyper-
elastic reconstructions show no overall upwards or downwards trends, however,
the linear reconstruction increases steadily with pressure. The errors associated
with this plot are shown in Fig. 6.7 d, where the hyperelastic reconstruction has
below 25% error for all reconstructions, while the linear error exceeds 300% for a
 115

pressure of 56 mmHg.

a b
16 6.5
Ground truth
6
Hyperelastic reconstruction
14 Linear reconstruction
5.5

5
12
4.5
kPa

kPa
10 4

3.5
8
3

2.5
6
2

4 1.5
0 10 20 30 40 50 60 0 10 20 30 40 50 60
mmHg mmHg

c d
180 350

160
300
140
250
120
Percent error

Percent error
100 200

80 150

60
100
40
50
20

0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
mmHg mmHg

Figure 6.7: Figure 6.7 a & b: Reconstruction results for the vessel wall and inclusion
regions, respectively. The ground truth and hyperelastic reconstruction show the K-
modulus, while the linear reconstruction shows 41 E. Figure 6.7 c & d: These plots show
the errors between the reconstructions in Fig. 6.7 a & b and the ground truth values.

Figure 6.8 quantifies the results of Figs. 6.3 and 6.5. Figure 6.8 a shows the
EYM for the ground truth and both reconstructions with respect to intraluminal
pressure in the vessel wall crescent between the lumen and the plaque. The linear
reconstruction shows an approximately linear curve, while the slope of the ground
truth and hyperelastic reconstruction increases slightly with pressure. For all
reconstructions, the linear values are smaller than the equivalent hyperelastic
values. At 8 mmHg, the linear reconstruction, hyperelastic reconstruction, and
ground truth are superimposed on each other, while they begin to diverge at larger
pressures. Figure 6.8 c shows the error associated with Fig. 6.8 a. There is no
positive or negative trend in the hyperelastic error, with all values below 10%.
However, the linear error increases with pressure, reaching 58% at 56 mmHg.
 116

Figure 6.8 b shows the maximum principal Cauchy stress for the ground truth
and both reconstructions with respect to the intraluminal pressure in the vessel
wall crescent between the lumen and the plaque. The trends in this plot mirror
those in Fig. 6.8 a. The errors associated with this plot are shown in Fig. 6.8
d. The error in the hyperelastic reconstruction is below 15% for all intraluminal
pressures, while the error in the linear reconstruction increases to 66% at a pressure
of 56 mmHg.

a b
20 160
Ground truth
18
Hyperelastic reconstruction
140
Linear reconstruction
16

14 120

12
100
kPa

kPa
10
80
8

6 60
4
40
2

0 20
0 10 20 30 40 50 60 0 10 20 30 40 50 60
mmHg mmHg

c d
60 70

60
50

50
40
Percent error

Percent error

40
30
30

20
20

10
10

0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
mmHg mmHg

Figure 6.8: Figure 6.8 a & b: Reconstruction results for the EYM and Maximum
principal Cauchy stress, respectively within the crescent between the lumen and the
plaque as a function of intraluminal pressure. Results are shown for the ground truth,
hyperelastic reconstruction, and linear reconstruction. Figure 6.7 c & d: These plots
show the errors between the reconstructions in Fig. 6.7 a & b and the ground truth
values.

6.5.2 Phantom Study

Figure 6.9 shows the actual and modeled stretch-stress curves for the PVAc vessel
phantoms measured by uniaxial compression testing. Each gray line represents
 117

test data from the four individual samples. The dotted red line shows the fitted
curve for the hyperelastic model. Each sample was subjected to stretches between
1 and 1.4. The proposed hyperelastic model fits the plaque data with an r2 value
of 0.9993, while the vessel wall samples showed more variation and had an r2 value
of 0.9908.

Figure 6.9: Stretch/stress curves for the vessel wall (a) and plaque (b) regions measured
by uniaxial compression testing. The gray lines show individual trials, while the dotted
red line shows the fitted curve.

Figure 6.10 shows the computed elastograms for the PVAc phantom using
both a hyperelastic model and a linear model in the reconstruction when the
lumen was pressurized to 50 mmHg. The top row shows the results of the K
modulus for the hyperelastic reconstruction, and the equivalent 14 E value for the
linear reconstruction. The reconstructed values using the linear model are larger
than those using the hyperelastic model. Additionally, there is more inter-region
variation in the nonlinear reconstruction than there is in the linear reconstruction.
The middle row shows the EYM for both reconstructions. The wall region in the
linear elastogram was computed to be stiffer than the hyperelastic elastogram
far from the lumen, while it was smaller than the hyperelastic elastogram close
to the lumen. The plaque was stiffer in the linear elastogram compared to the
 118

hyperelastic elastogram. The third row shows the maximum principal Cauchy
stress elastograms for both reconstructions. The peak stress around the lumen is
greater for the hyperelastic reconstruction, while the vessel wall stress is larger for
the linear reconstruction. The black rectangle in the top right image shows the
ROI used to compute the EYM and principal stress in Fig. 6.11.
 119

Nonlinear reconstruction Linear reconstruction

-6 8000

7000
-4
6000
-2
5000
K-modulus

mm

0 4000

3000
2
2000
4
1000

4
10
-6

-4 2

-2
1.5
EYM

mm

0
1
2

0.5
4

-6 6000

-4 5000
Maximum principal stress

-2 4000
mm

0 3000

2 2000

4 1000

0
-6 -4 -2 0 2 4 -6 -4 -2 0 2 4
mm mm

Figure 6.10: Hyperelastic and linear reconstructed elastograms for the PVAc phantom.
The top row shows the K-modulus elastogram, the middle row shows the EYM, and the
bottom row shows the maximum principal stress for a pressure of 50 mmHg.
 120

Figure 6.11 a shows the measured and computed K-modulus values for the
vessel wall and plaque regions, using the hyperelastic reconstruction method. The
error bars represent the standard deviation between four measurements of both
the uniaxial compression testing and the modulus reconstruction. The modulus
reconstruction overestimates both values. Figure 6.11 b shows the EYM and
principal stress values in the area marked in the black square in Fig. 6.10. The
error bars represent the standard deviation between the four data sets. The linear
1
reconstruction underestimates both parameters. 4
EYM is plotted so that it can
be displayed next to the principal stress for better comparison.
a
7000
Measured Kvalue
Reconstructed Kvalue
6000

5000

4000
Pa

3000

2000

1000

0
Vessel wall Plaque

b
5000
Nonlinear reconstruction
4500 Linear reconstruction

4000

3500

3000
Pa

2500

2000

1500

1000

500

0
EYM/4 Principal stress

Figure 6.11: a: Measured and reconstructed K-modulus values for the PVAc phantom.
b: Reconstructed EYM and stress values using the hyperelastic and linear algorithms.
These values were computed by averaging within the region defined by the black rect-
angle in Fig. 6.10.
 121

6.6 Discussion

In this chapter, the effects of material nonlinearity and large deformation on mod-
ulus and stress recovery were examined for both linear and hyperelastic reconstruc-
tion models. The primary findings of this study were as follows: (a) Incrementally
built up displacement fields produce results that can be used to accurately recon-
struct stress and modulus under large strains (b) The peak maximum principal
Cauchy stress can be reconstructed with under 15% error in simulation studies,
(c) nonlinear material effects cause stiffness increases of up to 250% in high strain
regions of the phantom, (d) The reconstructed linear Youngs modulus varies by
over 300% depending on the intraluminal pressure, and (e) linear reconstructions
of peak stress underestimated the ground truth values by 65%.

Previous work has shown that human arteries exhibit hyperelastic behavior
[Holzapfel et al., 2005; Walsh et al., 2014]. Specifically, the tissue becomes stiffer
as strain increases. This stiffening will not be reflected in linear models, which will
lead to noticeable error when strains exceed 5%. Stress is directly proportional to
stiffness, and as such, the underestimation of stiffness will also lead to the under-
estimation of stress. This becomes a critical problem when determining plaque
stability using stress elastograms, because the strains in the plaque shoulders can
exceed 100%. Examining these effects can be difficult, because large strain intro-
duces decorrelation artifacts into tracked displacements and strains. Researchers
have previously avoided this problem by computing displacements between multi-
ple sets of frames with low strain, and summing the results. Kim et al. proposed
a method to recover vascular strain in the nonlinear regime by accumulating dis-
placements measured between each frame in the Lagrangian presentation (i.e. ves-
sel coordinates) [Kim et al., 2004]. Using this methods, they were able to observe
the nonlinear displacement-pressure relationship in the brachial artery. A second
method to characterize nonlinearity would be to accumulate strain rather than
 122

displacement. Pavan et al. used strain accumulation to determine the stress-strain

curve in hyperelastic materials [Pavan et al., 2012]. However, these methods can-
not be used alone to compute the stress or modulus distribution in heterogeneous
tissue. Model-based hyperelastic reconstruction can be used to determine vascu-
lar properties, however, multi-parameter models have been found to increase the
ill-posedness of the problem, and the relative sensitivity difference between each
parameter makes it difficult to converge to the correct solution. Wittek et al. used
a model-based reconstruction method to determine the hyperelastic properties of
homogeneous vessels using a five-parameter aortic model [Wittek et al., 2013;
Gasser et al., 2006]. The resulting error in each reconstructed parameter reached
86%. A two-parameter reconstruction of simulated arteries performed by Khalil
et al. achieved under 10% error when using ground truth displacements corrupted
by white noise. However, this method required constraining each parameter to
vary by under one order of magnitude, centered around the ground truth. This
limitation may not be possible in vivo where the ground truth is unknown. For
these reasons, we used a one parameter hyperelastic model with an exponential
strain energy density function.

The nonlinear effects of this model are visible in Fig. 6.1, where the linear re-
construction of 14 E diverged from the K-modulus as the intraluminal pressure in-
creases. The reason for this trend becomes apparent when looking at the adjusted
EYM elastograms in Figs. 6.4 and 6.5, which become noticeably stiffer. These
plots show less variation between the linear and hyperelastic reconstructions after
the true stiffness was computed. Figure 6.5 also highlights one of the causes of the
underestimation in the linear reconstruction. The ground truth and hyperelastic
reconstruction show that the EYM is spatially variant even within homogeneous
material regions. These fluctuations cannot be recovered using a linear soft-priors
reconstruction, which attempts to limit the variance within each material region.
Thus, this method converged to a solution that removes these inhomogeneities.
 123

This caused the high-strain region between the lumen and the plaque to be un-
derestimated, and the low-strain regions at the outer edge of the phantom to be
overestimated. These same trends appeared in the Cauchy stress elastograms in
Figs. 6.3 and 6.4. Figure 6.8 b shows that for all models there was an approx-
imately linear trend between pressure and stress for all elastograms. This trend
would be exactly linear in a symmetric, homogeneous material of linear or hyper-
elastic elasticity, to conserve forces. The hyperelastic stresses show some slight
nonlinearity in their curves, which is highlighted in the error plot in Fig. 6.8 d.
This is due to the inhomogeneities in the materialnamely that the stress distribu-
tion changes depending on the EYM distribution. Several studies have shown that
as the plaque cap becomes stiffer, the stress distribution changes [Cilla et al., 2012;
Akyildiz et al., 2011]. In the modeled hyperelastic phantom, the region between
the plaque and the lumen gets stiffer as pressure increases, leading to this change
in stress distribution. For all pressures, the hyperelastic reconstruction recovered
the EYM and maximum stress with under 15% error, showing that the incremen-
tal displacement tracking could accurately recover the true displacements, and
the inverse solver could converge to the correct solution. The curves in Fig. 6.7
show quantitatively the extent of the variation in the measured Youngs modulus
in the linear reconstruction. The Youngs modulus in the inclusion increased by
250% as the pressure is raised from 8-56 mmHg. This highlights the pitfalls when
using large strains for modulus reconstruction. If hyperelasticity is not known or
accounted for, the recovered values will not agree with what has been previously
reported in the small strain regime. We have performed previous studies exam-
ining the effects of noise on reproducibility using this reconstruction algorithm.
These studies showed that measurement noise does not significantly affect the
error, and as such, this analysis has been omitted in the simulation results.

Several previous studies have examined material nonlinearity in PVAc vessel

phantoms [Pazos et al., 2009; Mehrabian and Samani, 2009; Jiang et al., 2013].
 124

Pazos et al. determined that the degree of nonlinearity depended on the number of
freeze-thaw cycles used in fabrication, while Jiang et al. also noted that the degree
of nonlinearity can be increased by adding NaCl to the PVAc solution. In all cases,
the nonlinearity was clearly visible after the sample reached a stretch ratio of 1.3.
Figure 6.9 shows this same trend. The exponential fit for the inclusion (Fig. 6.9
b) is slightly better than that of the vessel wall (Fig. 6.9 a), which may be due to
differing degrees of nonlinearity between freeze thaw cycles [Pazos et al., 2009], or
the variance between trials. The elastograms in Fig. 6.10 show the same trend as
the simulation study. Namely, that the linear reconstruction underestimated the
peak stress and EYM with respect to the hyperelastic reconstruction, while 14 E was
overestimated with respect to the K-modulus. The EYM and stress results also
show more radial symmetry than was observed in simulation, with less observable
stress concentration between the plaque and the lumen. This is because the
plaque is further from the lumen, leading to smaller strain in that region. The
quantitative results in Fig. 6.11 reflect the trends seen in the elastograms in Fig.
6.10. The computed K-modulus using the hyperelastic reconstruction agrees with
the results found using uniaxial testing. Additionally, the linear reconstruction
underestimated the peak EYM by 19%, while it underestimated the peak stress
by 48%.

One limitation of this work is that the exponential strain energy density func-
tion that was used is only a function of a single parameter. This may cause inac-
curacies in the fit for different types of hyperelastic behavior. Multiple parameters
could also provide insight on whether the degree of nonlinearity can provide useful
pathological information about the plaque composition. This study increased the
intraluminal pressure linearly rather than following a pulsatile pattern as would
be observed in vivo. This can be justified by observing that IVUS can collect data
at a high enough rate that frames can be selected for analysis that show a lin-
ear increase in pressure. However, pulsatile excitation could introduce additional
 125

motion artifacts not observed here.

6.7 Summary

In this chapter, the effects of material nonlinearity on stress and modulus recon-
struction were observed. By accounting for the hyperelastic effects, peak stresses
in IVUS simulation results could be recovered with under 15% error under strains
of up to 30%. In both simulation and PVA phantom studies, displacements could
be accumulated over multiple frames to achieve these results. The linear recon-
struction failed to reconstruct the correct stresses in part, because it could not
recover the spatially varying EYM stiffness field.
 126

7 In Vivo Quantitative
Elastography for Routine
Screening

7.1 Introduction

In this thesis, quantitative vascular elastography has been shown to successfully

reconstruct modulus and stress elastograms in both simulated and PVAc vessel
phantoms. This method has proven to be resilient against noise, and compatible
with both linear array and IVUS acquisition methods. However, to be a clinically
viable screening method, QVE must be able to compute elastograms in vivo. This
poses two new challenges. First, there will be increased noise in the tracked dis-
placements, due to vessel motion and increased attenuation. Even if the patient
is holding still, motion due to breathing may displace the vessel. Tissue inho-
mogeneities and greater depths will cause the acoustic attenuation to increase.
Second, there may be model-data mismatch between the tissue mechanics and
the chosen mechanical model. Arteries exhibit viscoelastic behavior as well as hy-
perelasticity, which is not modeled [Learoyd and Taylor, 1966; Cheng et al., 2002;
Garca et al., 2011]. Additionally, the single-parameter tissue model we employ
may not be able to model all tissues. We believe that the quantitative vascular
 127

elastography method can overcome these challenges due to the soft-priors recon-
struction method. This has already been proven to be robust against measurement
noise in Chapter 4, using conventional linear array imaging, which has a poorer
resolution and power than the plane wave imaging used here. Additional, soft-
priors will allow for inter-region variation, which may allow the EYM to match
the true relationship by adjusting the local stiffness to compensate for model
inaccuracies.

Previously, several research groups completed studies that measure stiffness

variations within human arteries using elastography. One method to combat
the high measurement noise is to use an iterative modulus reconstruction while
imposing a hard-priors constraint to force homogeneity within specified regions
[Yalavarthy et al., 2007a]. As in vivo data is often too noisy to discern different
plaque components from B-scans, these studies used strain estimates to segment
potential inhomogeneities [Baldewsing et al., 2008; Le Floch et al., 2012]. While
these studies yielded promising results, they used IVUS in the data collection,
which can produce higher resolution images than NIVE, and is not suited for
screening. Schmitt et al. proposed a modulus reconstruction technique using
NIVE for in vivo scans of healthy arteries [Schmitt et al., 2007]. This method
computed the elastic modulus distribution directly from measured displacements
by assuming the vessel could be modeled as a homogeneous pressure vessel. How-
ever, vessels with plaques can have modulus variations of over two orders of mag-
nitude, precluding this method from being used to assess cardiovascular disease
[Barrett et al., 2009]. Wittek et al. also proposed a method for determining the
elastic properties of in vivo arteries using NIVE [Wittek et al., 2013]. This method
used an iterative inversion scheme to compute the hyperelastic stiffness and stress
within vessels. However, in this study, the vessels were modeled as hollow shells,
which would not be able to reconstruct radial inhomogeneities caused by plaques.
Using the soft-priors regularization of QVE would allow for the measurement of
 128

spatial variation within pre-selected regions. If no plaque features are visible,

the vessel can be modeled as a single region and still recover inhomogeneities.
However, this technique poses a new challenge in determining an optimal regular-
ization parameter that will maximize inter-region variations, while still achieving
convergence in the inversion.

In the chapter, the QVE framework was used to compute the effective Youngs
modulus (EYM) and principal Cauchy stress elastograms from in vivo carotid
artery scans using NIVE, with both hyperelastic and linear elastic models. We ex-
amined the viability of both sparse synthetic aperture (SA) and plane wave (PW)
acquisition methods, using a Sonix RP ultrasound system. We hypothesized that
QVE would be able to compute spatially varying Youngs modulus elastograms
without prior knowledge of vessel inhomogeneities. To test this hypothesis, PW
and SA scans were collected from four volunteers, and the computed elastograms
were analyzed by observing trends in the modulus variation between acquisition
and reconstruction methods, comparing the optimal regularization parameters,
and correlating the modulus elastograms trends with the age of the volunteers.

7.2 Methods

7.2.1 Data Acquisition

Four healthy volunteers were scanned using a Sonix RP ultrasound system (Ul-
trasonix, Peabody, MA, USA) that was equipped with a 128 element linear trans-
ducer array (L145/38 probe) and a multichannel data acquisition system (Sonix
DAQ , Ultrasonix, Peabody, MA, USA). While the volunteers were lying in a
supine position, 64 interleaved synthetic aperture and plane wave acquisitions
were obtained from the right carotid artery using the imaging parameters speci-
fied in Chapter 4.2. These frames were all taken during the diastolic period. The
 129

age, systolic blood pressure, and diastolic blood pressure of each volunteer were
recorded.

7.2.2 Displacement and Strain Estimation

Displacements were computed between every two acquired frames using cross-
correlation speckle tracking. These displacements were aggregated using the co-
ordinate transformation system described in Chapter 6.2.

Each pair of displacements was measured using cross-correlation-based speckle

tracking, with 1.8 mm 1.8 mm kernels that overlap by 80% in both the axial and
lateral direction. We used a 5 5 median filter to remove spurious displacement
estimates. The mean vessel expansion, d, was computed by taking the difference
between the mean radial displacements on opposite sides of the vessel in the axial
direction. These values were computed for each acquisition by taking the average
displacements in two 10 10 windows. The mean radial strain was defined by
taking the average of the radial strains within the same 10 10 windows. The
windows were oriented axially to take advantage of the superior axial resolution
of PW and SA.

7.2.3 Effective Youngs Modulus and Principal Stress Es-

timation

The vessel geometry was determined by manually measuring the inner diameter
of each vessel, and assuming a 2 mm thickness, and that each vessel was circu-
lar. FE meshes were created from these measurements such that the maximum
element radius was 400 m, to be less than the 600 m axial resolution of the
transducer array. This used between 3900 and 4580 constant strain triangle (CST)
elements. Displacement boundary conditions were used on the inner lumen of the
vessel to improve the robustness of the reconstruction. These boundary conditions
 130

can recreate the variable radial displacements along the inner lumen better than
forces can. These irregularities may be due to non-uniform blood flow or may be
artifacts of the tracking algorithm. The outer boundary was left free. Both linear
and hyperelastic reconstructions were performed assuming plane stress. The hy-
perelastic model used the strain energy density function shown in Equation 6.3,
and the EYM was calculated using Equation 6.14. Ten steps were used for the
Newtons method displacement computation per iteration, and the convergence
was set to 0.01, to match the parameters used in Chapter 6. For both reconstruc-
tion techniques, the algorithm was halted after 50 iterations. By this point, either
the RMSE between the measured and computed displacements varied by under
0.1%, or the solution had begun to diverge.

Without explicit pressure information, the reconstruction will only produce rel-
ative values of the EYM, centered around the initial guess [Doyley et al., 2000]. To
add this information back into the computation, we computed the mean Youngs
modulus for each vessel, assuming it could be treated as a thick-walled pressure
vessel, using the following equation [Raj and Ramasamy, 1983],

ri2 r ro2 ri2

 
Pi
E= (1 ) 2 + (1 + ) , (7.1)
ur ro ri2 r (ro2 ri2 )

where p is the pressure difference, d is the vessel diameter, t is the vessel

thickness, d is the radial expansion, and is the Poissons ratio set to 0.5.
Equation 7.1 assumes material linearity, which is not true in arteries. To as-
sess the incurred error due to nonlinearity, 12 homogeneous vessels were sim-
ulated using the vessel geometry from one of the volunteers. The K-modulus
was set to 25,000 to correspond to a small-strain Youngs modulus of 100,000
Pa, which is similar to experimentally derived values [Barrett et al., 2009;
Loree et al., 1992]. The intraluminal pressure was varied between 5-60 mmHg
in steps of 5 mmHg. For each simulation, the EYM was computed using Equa-
 131

tions 7.1 and 6.14. Figure 7.1 plots the error between the true EYM and the
computed EYM using Equation 7.1. Each of the values in Fig. 7.1 are for a point
halfway between the inner and outer lumens. The range of strains tested was set
to match the maximum observed strain from the volunteers.

Figure 7.1: Error between the true EYM and computed value using the assumption of
a thick-walled pressure vessel. All units are percentages.

This plot shows that the relative error can reach 31% for a strain of 6.2%.
When computing the error between the simulated and computed EYMs for a
linear material, the error was under 1% for all strains.

Due to limitations of the ultrasound scanner, we were not able to measure

the maximum and minimum blood pressure in the imaged frames, and had to
rely on systolic and diastolic readings from a sphygmomanometer. The range of
imaged pressures was assumed to be much smaller than the full pressure range,
which would have a linear scaling effect on the values computed in Equation 7.1.
Because of this, we treated the computed Youngs modulus values as proportion-
ality constants, making the assumption that the measured pressure range was the
same for all volunteers. After scaling the reconstructed modulus elastograms by
the proportionality constants, the new elastograms were normalized to the small-
 132

est value. These were used to compute the maximum principal Cauchy stresses
as shown in Equations 2.2 and 6.8 and normalized to the stress elastogram with
the smallest mean.

The regularization parameter, , was optimized for each modulus reconstruc-

tion independently. All reconstructions were performed sweeping between 1016
and 1014 . For all reconstructions, this range contained the lowest value of that
reached a smooth convergence when measured over 50 iterations. This value was
chosen to be the optimal regularization. Within the chosen reconstruction, the
final elastogram was selected at the iteration that minimized the RMS difference
between the measured and computed displacements. This does not necessarily
occur at the final iteration, as some reconstructions ultimately diverge. Stress
and EYM results for the hyperelastic reconstruction using SA were not included
in this study because this combination immediately diverged and never arrived at
a minimized elastogram due to the increased noise in the SA displacements with
respect to PW. The maximum principal Cauchy stresses were computed for both
the linear and hyperelastic reconstructions.

7.3 Results

Figure 7.2 shows the B-scans for each of the four volunteers using SA and PW
acquisitions. Each image is displayed on a logarithmic scale with reference to
the PW scans. In all cases, the SA images were of lower intensity than the
corresponding PW scans. There are portions of the vessels in both the PW and
SA scans that are indistinguishable from the background, although these instances
are more prevalent in SA. The axes are displayed in units of mm.
 133

1 2 3 4
15 10 0

10 10 15 -5
20
Plane wave

20
15 15 -10
25
25
-15
20 20
30
30
-20
25 25
35 35
-5 0 5 10 -5 0 5 10 0 5 10 15 -10 -5 0 5 10
-25
15 10
-30
10 10 15
20
Sparse array

-35
20
15 15
25
-40
25
20 20
30 -45
30

25 25
35 35 -50
-5 0 5 10 -5 0 5 10 0 5 10 15 -10 -5 0 5 10

Figure 7.2: B-scans for each of the four volunteers using SA and PW acquisition dis-
played on a logarithmic scale

Figure 7.3 shows the measured axial and lateral displacements for SA and PW
in units of meters, aggregated over all frames. The displacements from Volunteer 4
showed the most obvious radial expansion with horizontal displacements recorded
in the lateral direction and vertical displacements recorded in the axial direction.
The displacements are not centered at zero, implying significant free body motion
during imaging. For all subjects, the axial SA scans showed similar trends to the
PW scans, but are more noisy. In the lateral direction, the SA displacements
differed more prevalently from the PW images.
 134

1 10 -4 2 10 -3 3 10 -4 4 10 -3
1.5 1

4 6

3 1
PW 0.5
4
2
0.5
1
2 0
0
Lateral 0
-1
0 -0.5
-2
-0.5
-3
SA -2
-1
-4 -1

-5 -4

-6 -1.5 -1.5

-4 -4 -4 -4
10 10 10 10
-1

-2
-2
-2
-2
PW -4 -3
-3
-3 -4
-6

-4 -5
Axial -4 -8

-5 -6

-10
-5 -7
-6
SA -8
-12
-6
-7
-9
-14

-7 -8 -10

Figure 7.3: Measured axial and lateral displacements for SA and PW for all volunteers
(m)

Figure 7.4 shows the mean cross-correlation coefficients for each volunteer over
all displacement measurements. In all cases, the PW maps had higher correlation
coefficients between frames than SA. The regions of poor correlation correspond
to the highly attenuated regions in the B-scans displayed in Fig. 7.2. The figure
axes are in units of mm.
 135

1 2 3 4
16 1
18 18 18
18
0.95
20 20 20
20
Plane wave

22 22 22
22 0.9
24 24 24
24
26 26 26 0.85
26
28 28 28
28 0.8
30 30 30
-5 0 5 -5 0 5 -5 0 5 -5 0 5
0.75

16
18 18 18 0.7
18
20 20 20
20
Sparse array

0.65
22 22 22
22
24 24 24 0.6
24
26 26 26
26 0.55
28 28 28
28
30 30 30 0.5
-5 0 5 -5 0 5 -5 0 5 -5 0 5

Figure 7.4: Cross-correlation coefficients for each volunteer for SA and PW. Results are
averaged over all displacement pairs.

Figure 7.5 shows box-and-whisker plots for the cross-correlation images in Fig.
7.4. These plots show that the mean values for the PW data are always larger
than those for the SA data.

Figure 7.5: Box-and-whisker plots for the data in Fig. 7.4

 136

Figure 7.6 shows the root mean square difference (RMSD) between the mea-
sured and computed nodal displacements over all reconstruction types and reg-
ularization factors for Volunteer 1. The white circles in each figure denote the
iteration and regularization values used to create the elastograms in Figs. 7.7
and 7.8. The linear and hyperelastic reconstructions converged absolutely for PW
displacements, while the SA reconstructions began to diverge after the twelfth
iteration. For the PW reconstructions, the linear RMSD were slightly smaller
than the hyperelastic RMSD, implying that it was a better model of the tissue
mechanics over the range of strain tested.

10 -6 Linear Reconstruction - PW 10 -6 Nonlinear Reconstruction - PW

1.8 1.9
6E-16 2E-15
7E-16 1.89 3E-15
8E-16 4E-15
1.78
9E-16 1.88 5E-15
10E-16 6E-15
11E-16 7E-15
1.87
12E-16 8E-15
Displacement difference

Displacement difference

1.76
1.86

1.74 1.85

1.84
1.72
1.83

1.82
1.7
1.81

1.68 1.8
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Iterations Iterations

10 -6 Linear Reconstruction - SA
1.74
9E-16
10E-16
1.73
11E-16
12E-16
1.72 13E-16
14E-16
15E-16
Displacement difference

1.71

1.7

1.69

1.68

1.67

1.66
0 10 20 30 40 50 60 70
Iterations

Figure 7.6: RMSD between the measured and computed nodal displacements as a func-
tion of solve iterations for all acquisition/reconstruction pairs. Each line represents
a different regularization value. The white circles indicate the chosen regularization
parameters and iteration numbers.
 137

Figure 7.7 shows the relative EYM elastograms for each volunteer using plane
wave with linear reconstruction (PWL), plane wave with hyperelastic reconstruc-
tion (PWNL), and sparse array with linear reconstruction (SAL). The PWL and
PWNL elastograms had the same mean values because they used the same scaling
factor, although the hyperelastic reconstructions were noisier, leading to higher
standard deviations. The SAL elastograms were all stiffer than the PWL and
PWNL reconstructions.

1 2 3 4
2 2 2 2

1.8 1.8 1.8 1.8

PWL 1.6 1.6 1.6 1.6

1.4 1.4 1.4 1.4

1.2 1.2 1.2 1.2

1 1 1 1

0.8 0.8 0.8 0.8

0.6 0.6 0.6 0.6

PWNL
0.4 0.4 0.4 0.4

0.2 0.2 0.2 0.2

0 0 0 0

4 4 7 4
6
3 3 5 3

4
SAL 2 2 2
3

1 1 2 1
1
0 0 0 0

Figure 7.7: Relative EYM reconstructions for PWL, PWNL, and SAL acquisi-
tion/reconstruction pairs for all volunteers

Figure 7.8 shows the relative maximum principal Cauchy stress elastograms
for all volunteers and acquisition/reconstruction pairs. All stress elastograms had
more inter-vessel variation than the corresponding EYM elastograms, which was
expected because stress is a function of both the strain and the modulus. There
was little variation between the PWL and PWNL reconstructions. However, the
 138

sparse array elastograms do not show the same stress distribution as the plane
wave elastograms.

1 2 3 4
10 10 10 10

9 9 9 9

8 8 8 8
PWL
7 7 7 7

6 6 6 6

5 5 5 5

4 4 4 4

3 3 3 3

PWNL 2 2 2 2

1 1 1 1

0 0 0 0

-1 -1 -1 -1

10 10 10 10

8 8 8 8

6 6 6 6
SAL
4 4 4 4

2 2 2 2

0 0 0 0

Figure 7.8: Relative maximum principal Cauchy stress reconstructions for PWL,
PWNL, and SAL acquisition/reconstruction pairs for all volunteers

Figure 7.9 plots the volunteer age verses the mean EYM for all acquisi-
tion/reconstruction pairs. The error bars represent the standard deviation of
the EYM within each vessel. In all cases, the EYM of the 49 year old subject
was larger than that of the 37-39 year old subjects. The PWNL and SAL EYMs
have more variance than the PWL reconstruction (note that the SAL plot is on a
different scale due to the wider variation in reconstructed values).
 139

PWL PWNL SAL

1.6 1.6 5.5

1.5 5
1.5
4.5
1.4
1.4
4
Normalized EYM

Normalized EYM

Normalized EYM
1.3
1.3 3.5
1.2
1.2 3
1.1
2.5
1.1
1
2
1 0.9 1.5

0.9 0.8 1
36 38 40 42 44 46 48 50 36 38 40 42 44 46 48 50 36 38 40 42 44 46 48 50
Age Age Age

Figure 7.9: Volunteer age verses the mean EYM for all acquisition/reconstruction pairs.
The error bars represent the standard deviation of the EYM within each vessel.

Table 7.1 quantifies the normalized mean measured strain, regularization value,
RMSD, EYM and maximum principal Cauchy stress values for each volunteer.
The strain in Volunteer 4 was larger than that of the other subjects by over
a factor of two. The mean regularization factors for the PWL reconstructions
was larger than that of the SAL reconstruction, suggesting that the sparse array
displacements were noisier. The RMSDs between the PWL and PWNL recon-
structions varied by under 5%, suggesting that both elastic models could be used
to represent the vessel under the given strain. For both EYM and stress, the SAL
reconstructions had more variance within the vessels.
 140

Volunteer Strain Method Reg. RMSD EYM (Pa) Stress (Pa)

PWL 9.00E-16 1.75E-06 1.04 0.03 1.35 1.01
1 2.72 PWNL 4.00E-15 1.82E-06 1.04 0.07 1.29 1.03
SAL 1.00E-15 1.69E-06 1.89 0.14 2.12 1.75

PWL 4.00E-15 2.43E-06 1.08 0.09 3.21 2.65

2 2.72 PWNL 9.00E-15 2.30E-06 1.08 0.27 3.13 255
SAL 1.10E-14 5.52E-06 1.08 0.06 2.12 208

PWL 3.00E-15 2.09E-06 1.50 0.05 1.01 0.96

3 2.88 PWNL 8.00E-15 2.06E-06 1.50 0.09 1.00 1.00
SAL 5.00E-16 1.13E-06 5.11 0.23 2.51 2.22

PWL 4.00E-15 1.45E-06 1.00 0.06 2.28 1.67

4 6.18 PWNL 3.00E-14 1.52E-06 1.00 0.07 2.16 1.81
SAL 6.00E-15 2.55E-06 1.15 0.06 2.40 2.23

Table 7.1: Normalized Mean strain, regularization parameter, RMSD, EYM and max-
imum principal Cauchy stress values for each volunteer and acquisition/reconstruction
combination

7.4 Discussion

In this study we performed preliminary in vivo elastography experiments using

QVE. These tests were conducted using a linear array transducer to compute the
EYM in the right carotid arteries of four volunteers. Both SA and PW beam-
forming techniques were used, and modulus reconstructions were performed using
both linear and a hyperelastic mechanical models. The preliminary findings of
this study were as follows: (a) Both LA and PW could be used to generate ac-
cumulated displacement maps, (b) The QVE algorithm reached convergence for
PWL, PWNL, and SAL acquisition/reconstruction pairs and produced EYM and
maximum principal Cauchy stress elastograms, and (c) The EYM correlated with
age for all acquisition/reconstruction pairs.

Previous work has shown that while SA imaging performs better than PW in
phantom studies and simulation, this is not the case in vivo due to the increased
 141

attenuation [Korukonda et al., 2013]. The results in Fig. 7.2 showed that the SA
B-mode images had intensities that were reduced by over 20 dB in places, com-
pared with the PW scans. There wes more signal deterioration than was reported
in Chapter 4, because there was no additional attenuating medium separating
the vessels from the transducer. Previous works that compute in vivo displace-
ments have reduced the noise due to attenuation by summing multiple sets of dis-
placements, acquired incrementally [McCormick et al., 2012; Larsson et al., 2015;
Widman et al., 2014]. Using these methods, the authors were able to reconstruct
displacement and strain fields over a full cardiac cycle. The effects of random
noise decrease over multiple acquisitions, allowing for images with higher SNRs.
This effect can be observed in Fig. 7.3. Although there was still observable error
in the fields, the PW results are smooth, due to the averaging. The SA results
show more variance, although the trends mimic those observed in PW. These high
variance regions in the SA displacements correspond to regions with low cross-
correlation coefficients, shown in Fig. 7.4. For all volunteers, the cross-correlation
coefficients were higher in PW, suggesting that it was better able to recover the
true tissue motion. However, the PW displacements did not show perfect radial
expansion. In Fig. 7.3, the PW axial field for Volunteer 2 shows a region of high
displacement on the left hand side, where a radially expanding homogeneous ves-
sel would have little axial motion. This anomaly corresponds to a region in Fig.
7.2 where the vessel wall cannot be observed in PW or SA. Thus, increasing the
power of the imaging could minimize these effects. In addition to power, some
of the non-radial expansion could be physical, due to vessel inhomogeneities or
nonuniform fluid flow [Buchanan et al., 1999].

In this study, we were unable to segment any structural inhomogeneities within

the vessel due to poor resolution and contrast. Thus, it was necessary to choose a
reconstruction regularization parameter that could maximize the allowable inter-
region variation, while achieving convergence. Several studies have implemented
 142

a regularization scheme for soft-priors reconstructions that chooses the initial reg-
ularization parameter to be 10 times greater than the maximum component of
the Hessian matrix, and reduces it at each iteration [Brooksby et al., 2005b;
Brooksby et al., 2005a; Brooksby et al., 2006]. However, this method does not
maximize the inter-region variation. Figure 7.6 shows our proposed regularization
method, which sweeps a range of potential regularization parameters. For low
regularizations, the displacement error oscillates and eventually diverges. This
is due insufficient damping in the inversion [Golub et al., 1999]. Regularization
parameters where chosen as the lowest value that does not exhibit this behavior.
In the SA reconstruction, the solution ultimately diverges even when the problem
is sufficiently regularized. This effect is due to semiconvergence [Bjorck, 1996;
Hansen, 1998; Mojabi and LoVetri, 2009]. This is because te effects of noise be-
come more prevalent as the inversion converges on the true solution. To account
for this, the final iteration was considered to be the one that best minimized the
RMS difference between the measured and computed displacements. Table 7.1
shows that for all linear reconstructions, PW images required lower regularization
parameters due to their lower noise. Using this regularization method, each of the
EYM reconstructions in Fig. 7.7 show similar spatial variation. The hyperelastic
PW reconstruction had a higher variance that the linear reconstruction (Table 7.1,
Fig. 7.7), which was likely due to the fact that the stiffness is strain-dependent,
and noisy strain fields will effect noisy EYM elastograms. The maximum principal
Cauchy stress elastograms in Fig. 7.8 showed that the linear and hyperelastic PW
reconstructions were visibly similar, while there is noticeable difference in the SA
elastogram. It was expected that both PW reconstructions would look similar
because the strains reported in Table 7.1 were all under 10%.

Many studies have reported that artery stiffness increases with age due to
vessel hardening [Laurent et al., 2006; Vaitkevicius et al., 1993; Kawasaki et al.,
1987]. Figure 7.9 expresses this trend as well, although more data is necessary to
 143

validate this plot.

One limitation of this work is that we were unable to obtain absolute modulus
and stress values due to the lack of pressure information. This could be rectified
in the future by using a system that can acquire more RF frames to ensure that
the entire cardiac cycle is covered. Doing this, the systolic and diastolic pressures
could be assumed to be the imaged range. Using this information, arteries contain-
ing atherosclerotic plaques could be analyzed to compute the stress elastograms
and determine the likelihood of plaque rupture.

7.5 Summary

In this chapter, preliminary tests of in vivo vascular elastography were completed

using QVE. These experiments showed that QVE could reconstruct EYM and
stress elastograms of healthy arteries, and that the reconstructed stiffnesses in-
creased with age. These results warrant further investigation into the use of QVE
as a screening method for atherosclerosis.
 144

8 Conclusion and Future Work

8.1 Summary of Results

Cardiovascular disease causes 2160 deaths in the United States every day [Go
et al., 2014], with a majority of these deaths occurring because of atherosclerotic
plaque rupture within the carotid or coronary arteries. These plaques rupture
when with internal stresses exceed 300 kPa [Cheng et al., 1993]. However, there
are no current screening or diagnostic techniques that can be used to measure these
stresses. The overarching goal of this thesis was to advance the field of ultrasound-
based vascular elastography by developing a technique to quantitatively assess the
material properties and stresses within atherosclerotic plaques. Specifically, the
research performed in this thesis achieved four main goals:

1. A framework was developed to compute the Youngs modulus of vessels con-

taining soft plaques using a model-based, iterative inversion scheme. This
method used a priori structural information to achieve high accuracy mod-
ulus elastograms in attenuating media. Specifically, three non-invasive ac-
quisition methods were compared: Sparse synthetic aperture, plane wave,
and conventional linear array. All three methods were able to reconstruct
the soft inclusion, while sparse array and plane wave both provided high
 145

accuracy results despite the low people of sparse synthetic aperture.

2. It was discovered that the presence of intimal microvessels in advanced

plaques affect the local strain and displacement fields, causing inaccura-
cies in modulus and stress reconstruction. These effects were largest for
microvessels within the plaque itself. It was demonstrated that these effects
could be accounted for using contrast-enhanced intravascular ultrasound
imaging. The microvessel locations, found using an ultraharmonic pulse
inversion technique, were successfully used in the reconstruction to create
accurate modulus and stress elastograms.

3. It was demonstrated that the use of a hyperelastic mechanical model pro-

duces more accurate modulus and stress reconstructions of vessel phantoms
at large strains than a linear model. In a phantom study, the differences
between linear and hyperelastic reconstructions of hyperelastic vessel phan-
toms was quantified with respect to increasing pressure, showing that the
peaks stresses would be underestimated for high strain situations. These
results were confirmed in an experimental study using PVAc, which behaves
nonlinearly at large strains.

4. Preliminary in vivo tests were conducted to determine the feasibility of using

NIVE to reconstruct the stresses and modulus distributions in a clinical
setting. This study demonstrated that plane wave imaging provides RF
images with sufficient SNR and resolution to reconstruct the mechanical
properties when paired with the robust QVE inversion method.

8.2 Future Work

The work reported in this thesis advances the field of quantitative vascular elas-
tography, and provides a useful framework for identifying life-threatening plaques
 146

in a clinical setting. However, there are still many ways to extend this framework,
which are outside of the scope of this thesis. The methods described below could
be used to expand the applicability and diagnostic capabilities of quantitative
vascular elastography.

8.2.1 Automated Plaque Segmentation

This thesis has shown that quantitative vascular elastography using NIVE can
accurately reconstruct plaques in phantoms and simulation as well as showing
modulus and stress variation in the carotid artery. However, the resolution of
NIVE does not allow for the a priori segmentation of the plaque cap, which we
show to have a significant impact on stress concentration. This will introduce error
in the region-based soft-priors reconstruction method. For this reason, IVUS was
used to image the coronary artery with sufficient resolution to identify and segment
the plaque cap. However, this is not an ideal solution, because IVUS is an invasive
procedure. Previous work in IVUS elastography has been done to dynamically
segment plaque features as the inverse solver iterates [Baldewsing et al., 2006;
Le Floch et al., 2009]. Implementing such techniques in NIVE may allow it to
achieve the same level of accuracy as IVUS. This dynamic segmentation could
be implemented in several ways, including the previously explored technique of
strain-based segmentation [Le Floch et al., 2009]. One novel approach would be to
use the inter-region modulus variation allowed by the soft-priors reconstruction to
segment regions based on stiffness concentrations. These automatically generated
regions could be regenerated at each solver iteration. This method would not
only be useful for identifying the plaque cap, but also for identifying microvessels
without the use of contrast imaging, and the detection of microcalcifications
 147

8.2.2 Expanding the Mechanical Model to Include Crack

Propagation

In this thesis, a mechanical model was used that accounts for material nonlinearity,
but not geometric nonlinearity. In particular, this method does not account for
the effects of crack-propagation in the fibrous cap. In this work, as well as many
other studies, idealized vessel geometries are used in FE modeling. Even in vivo
and ex vivo studies assumed smooth lumen boundaries, while this may not be
true on a micron scale. Previous work in FE modeling has shown that micron-
scale features can cause large stress concentrations in the their immediate vicinity
[Vengrenyuk et al., 2006; Bluestein et al., 2008; Teng et al., 2012]. It is unknown
whether these minuscule stress concentrations can lead to rupture. However,
insight on this could be gained in crack propagation models were used in the FE
reconstruction. In solid mechanics, stresses above the yield strength of a material
can lead to either breaking or fracturing [Rossmann and Dym, 2008]. A model
that accounts for this could determine a) whether a stress concentration will cause
a rupture or a fracture, b) how fracturing will effect the material nonlinearity
of the cap, and c) How long it will take the fracture to propagate through the
plaque due to fatigue. Fatigue models have been used previously in vascular FE
models, but have never been applied to elastography [Gasser and Holzapfel, 2007;
Versluis et al., 2006].

8.2.3 Monitoring Plaque Progression

The in vivo results in this thesis are promising, showing a correlation between
the IMT and the mean vessel stiffness. However, conclusive results can only be
found using large-scale testing, when vascular pressure information is available.
In such a study, definitive correlations between the IMT, age, and stiffness could
be obtained. Additionally, regular scans of patients suffering from atherosclerosis
 148

could be conducted to determine 1) how the stiffness of the plaque and cap change
as the plaque progresses, and 2) how the sensitivity of QVE compares to other
indicators of atherosclerosis such as stenosis and IMT. If plaque rupture occurs in
any patients, the most recent stress elastograms could be used to approximate the
peak stresses in the vessels to better estimate the yield criterion. All of these data
could be used to develop a screening protocol to assess the likelihood of plaque
rupture in systematic and non-systematic patients.
 149

Bibliography

[Akins et al., 1998] Paul T Akins, Thomas K Pilgram, DeWitte T Cross, and
Christopher J Moran, Natural history of stenosis from intracranial atheroscle-
rosis by serial angiography, Stroke, 29(2):433438, 1998.

[Akyildiz et al., 2011] Ali C Akyildiz, Lambert Speelman, Harald van Brumme-
len, Miguel A Gutierrez, Renu Virmani, Aad van der Lugt, AF Van Der Steen,
Jolanda J Wentzel, and FJ Gijsen, Effects of intima stiffness and plaque mor-
phology on peak cap stress, Biomed Eng Online, 10(1):113, 2011.

[Anykeyev et al., 1991] VB Anykeyev, AA Spiridonov, and VP Zhigunov, Com-

parative investigation of unfolding methods, Nuclear Instruments and Methods
in Physics Research Section A: Accelerators, Spectrometers, Detectors and As-
sociated Equipment, 303(2):350369, 1991.

[Baldewsing et al., 2008] Radj A Baldewsing, Mikhail G Danilouchkine, Frits

Mastik, Johannes A Schaar, Patrick W Serruys, and Anton FW van der Steen,
An inverse method for imaging the local elasticity of atherosclerotic coro-
nary plaques, Information Technology in Biomedicine, IEEE Transactions
on, 12(3):277289, 2008.

[Baldewsing et al., 2004] Radj A Baldewsing, Chris L de Korte, Johannes A

Schaar, Frits Mastik, and Antonius FW van Der Steen, A finite element model
for performing intravascular ultrasound elastography of human atherosclerotic
coronary arteries, Ultrasound in medicine & biology, 30(6):803813, 2004.
 150

[Baldewsing et al., 2005] Radj A Baldewsing, Frits Mastik, Johannes A Schaar,

Patrick W Serruys, and Antonius FW van Der Steen, Robustness of re-
constructing the Youngs modulus distribution of vulnerable atherosclerotic
plaques using a parametric plaque model, Ultrasound in medicine & biology,
31(12):16311645, 2005.

[Baldewsing et al., 2006] Radj A Baldewsing, Frits Mastik, Johannes A Schaar,

Patrick W Serruys, and Antonius FW van der Steen, Youngs modulus re-
construction of vulnerable atherosclerotic plaque components using deformable
curves, Ultrasound in medicine & biology, 32(2):201210, 2006.

[Barrett et al., 2009] SRH Barrett, MPF Sutcliffe, S Howarth, Z-Y Li, and
JH Gillard, Experimental measurement of the mechanical properties of carotid
atherothrombotic plaque fibrous cap, Journal of biomechanics, 42(11):1650
1655, 2009.

[Bergel, 1961] DH Bergel, The static elastic properties of the arterial wall, The
Journal of physiology, 156(3):445, 1961.

[Bertagna and Veneziani, 2014] Luca Bertagna and Alessandro Veneziani, A

model reduction approach for the variational estimation of vascular compliance
by solving an inverse fluidstructure interaction problem, Inverse Problems,
30(5):055006, 2014.

[Bilgen and Insana, 1997] M Bilgen and MF Insana, Predicting target de-
tectability in acoustic elastography, In Ultrasonics Symposium, 1997. Pro-
ceedings., 1997 IEEE, volume 2, pages 14271430. IEEE, 1997.

[Bjorck, 1996] Ake Bjorck, Numerical methods for least squares problems, Siam,
1996.

[Blacher et al., 1998] Jacques Blacher, Bruno Pannier, Alain P Guerin, Sylvain J
Marchais, Michel E Safar, and Gerard M London, Carotid arterial stiffness as
 151

a predictor of cardiovascular and all-cause mortality in end-stage renal disease,

Hypertension, 32(3):570574, 1998.

[Bluestein et al., 2008] Danny Bluestein, Yared Alemu, Idit Avrahami, Morteza
Gharib, Kris Dumont, John J Ricotta, and Shmuel Einav, Influence of micro-
calcifications on vulnerable plaque mechanics using FSI modeling, Journal of
biomechanics, 41(5):11111118, 2008.

[Brooksby et al., 2006] Ben Brooksby, Brian W Pogue, Shudong Jiang, Hamid
Dehghani, Subhadra Srinivasan, Christine Kogel, Tor D Tosteson, John Weaver,
Steven P Poplack, and Keith D Paulsen, Imaging breast adipose and fi-
broglandular tissue molecular signatures by using hybrid MRI-guided near-
infrared spectral tomography, Proceedings of the National Academy of Sci-
ences, 103(23):88288833, 2006.

[Brooksby et al., 2005a] Ben Brooksby, Subhadra Srinivasan, Shudong Jiang,

Hamid Dehghani, Brian W Pogue, Keith D Paulsen, John Weaver, Christine
Kogel, and Steven P Poplack, Spectral priors improve near-infrared diffuse
tomography more than spatial priors, Optics letters, 30(15):19681970, 2005.

[Brooksby et al., 2005b] Ben Brooksby, John Weaver, Christine Kogel, Shudong
Jiang, Hamid Dehghani, Steven P Poplack, Brian W Pogue, and Keith D
Paulsen, Combining near-infrared tomography and magnetic resonance imag-
ing to study in vivo breast tissue: implementation of a Laplacian-type regu-
larization to incorporate magnetic resonance structure, Journal of biomedical
optics, 10(5):051504051504, 2005.

[Brusseau et al., 2001] Elisabeth Brusseau, Jeremie Fromageau, Gerard Finet,

Philippe Delachartre, and Didier Vray, Axial strain imaging of intravascu-
lar data: results on polyvinyl alcohol cryogel phantoms and carotid artery,
Ultrasound in medicine & biology, 27(12):16311642, 2001.
 152

[Buchanan et al., 1999] John R Buchanan, Clement Kleinstreuer, George A

Truskey, and Ming Lei, Relation between non-uniform hemodynamics and
sites of altered permeability and lesion growth at the rabbit aorto-celiac junc-
tion, Atherosclerosis, 143(1):2740, 1999.

[Calvert et al., 2011] PA Calvert, DR Obaid, NEJ West, LM Shapiro, D Mc-

Nab, CG Densem, PM Schofield, D Braganza, SC Clarke, M OSullivan, et al.,
B VH-IVUS findings predict major adverse cardiovascular events. The Viva
Study (virtual histology intravascular ultrasound in vulnerable atherosclero-
sis), Heart, 97(Suppl 1):A2A2, 2011.

[Carlier et al., 2005] Stephane Carlier, Ioannis A Kakadiaris, Nabil Dib, Manolis
Vavuranakis, Sean M OMalley, Khawar Gul, Craig J Hartley, Ralph Metcalfe,
Roxana Mehran, Christodoulos Stefanadis, et al., Vasa vasorum imaging: a
new window to the clinical detection of vulnerable atherosclerotic plaques,
Current atherosclerosis reports, 7(2):164169, 2005.

[Carpinteri, 2002] Alberto Carpinteri, Structural mechanics: a unified approach,

CRC Press, 2002.

[Cespedes, 1993] EI Cespedes, Elastography: Imaging of biological tissue elas-

ticity, PhD dissertationUniversity of Houston, Texas, 1993.

[Chaturvedi et al., 1998] Pawan Chaturvedi, Michael F Insana, and Timothy J

Hall, 2-D companding for noise reduction in strain imaging, Ultrasonics,
Ferroelectrics and Frequency Control, IEEE Transactions on, 45(1):179191,
1998.

[Cheng et al., 1993] George C Cheng, Howard M Loree, Roger D Kamm,

Michael C Fishbein, and Richard T Lee, Distribution of circumferential
stress in ruptured and stable atherosclerotic lesions. A structural analysis with
histopathological correlation., Circulation, 87(4):11791187, 1993.
 153

[Cheng et al., 2002] Koon-Sung Cheng, Alok Tiwari, Cara R Baker, Richard Mor-
ris, George Hamilton, and Alexander M Seifalian, Impaired carotid and
femoral viscoelastic properties and elevated intimamedia thickness in periph-
eral vascular disease, Atherosclerosis, 164(1):113120, 2002.

[Chiao and Thomas, 1996] Richard Y Chiao and LJ Thomas, Aperture forma-
tion on reduced-channel arrays using the transmit-receive apodization matrix,
In Ultrasonics Symposium, 1996. Proceedings., 1996 IEEE, volume 2, pages
15671571. IEEE, 1996.

[Cilla et al., 2012] Myriam Cilla, Javier Martinez, Estefania Pena, and
Miguel Angel Martnez, Machine learning techniques as a helpful tool to-
ward determination of plaque vulnerability, Biomedical Engineering, IEEE
Transactions on, 59(4):11551161, 2012.

[Cook et al., 2007] Robert D Cook et al., Concepts and applications of finite
element analysis, John Wiley & Sons, 2007.

[Corti and Fuster, 2011] Roberto Corti and Valentin Fuster, Imaging of
atherosclerosis: magnetic resonance imaging, European heart journal, page
ehr068, 2011.

[Davies, 1996] Michael J Davies, Stability and instability: two faces of coronary
atherosclerosis The Paul Dudley White Lecture 1995, Circulation, 94(8):2013
2020, 1996.

[De Backer et al., 2003] Guy De Backer, Ettore Ambrosionie, Knut Borch-
Johnsen, Carlos Brotons, Renata Cifkova, Jean Dallongeville, Shah Ebrahim,
Ole Faergeman, Ian Graham, Giuseppe Mancia, et al., European guidelines
on cardiovascular disease prevention in clinical practice: third joint task force
of European and other societies on cardiovascular disease prevention in clinical
 154

practice (constituted by representatives of eight societies and by invited ex-

perts), European Journal of Cardiovascular Prevention & Rehabilitation, 10(1
suppl):S1S78, 2003.

[De Borst et al., 2012] Rene De Borst, Mike A Crisfield, Joris JC Remmers, and
Clemens V Verhoosel, Nonlinear finite element analysis of solids and structures,
John Wiley & Sons, 2012.

[de Korte et al., 1997] Chris L de Korte, E Ignacio Cespedes, Antonius FW

van der Steen, and Charles T Lancee, Intravascular elasticity imaging us-
ing ultrasound: feasibility studies in phantoms, Ultrasound in medicine &
biology, 23(5):735746, 1997.

[de Korte et al., 2002] Chris L de Korte, Marion J Sierevogel, Frits Mastik,
Chaylendra Strijder, Johannes A Schaar, Evelyn Velema, Gerard Pasterkamp,
PW Serruys, and Anton FW van der Steen, Identification of atherosclerotic
plaque components with intravascular ultrasound elastography in vivo A Yu-
catan pig study, Circulation, 105(14):16271630, 2002.

[de Korte et al., 2000] Chris L de Korte, Anton FW van der Steen, E Ignacio
Cespedes, Gerard Pasterkamp, Stephane G Carlier, F Mastik, Arjen H Schon-
eveld, Patrick W Serruys, and Nicolaas Bom, Characterization of plaque com-
ponents and vulnerability with intravascular ultrasound elastography, Physics
in medicine and biology, 45(6):1465, 2000.

[De Korte et al., 2002] CL De Korte, SG Carlier, Frits Mastik, MM Doyley, AFW
Van Der Steen, PW Serruys, and N Bom, Morphological and mechanical infor-
mation of coronary arteries obtained with intravascular elastography. Feasibility
study in vivo, European heart journal, 23(5):405413, 2002.

[De Korte et al., 1998] CL De Korte, HA Woutman, AFW Van der Steen, EI Ce-
spedes, and G Pasterkamp, IVUS elastography: a potential identifier of vul-
 155

nerable atherosclerotic plaque, In Ultrasonics Symposium, 1998. Proceedings.,

1998 IEEE, volume 2, pages 17291732. IEEE, 1998.

[Delfino, 1996] Antonio Delfino, Analysis of stress field in a model of the human
carotid bifurcation, PhD thesis, EPFL, 1996.

[Dettmer et al., 2013] Matthias Dettmer, Nicola Glaser-Gallion, Paul Stolzmann,

Florian Glaser-Gallion, Juergen Fornaro, Gudrun Feuchtner, Wolfram Jochum,
Hatem Alkadhi, Simon Wildermuth, and Sebastian Leschka, Quantifica-
tion of coronary artery stenosis with high-resolution CT in comparison with
histopathology in an ex vivo study, European journal of radiology, 82(2):264
269, 2013.

[Diethrich et al., 2007] EB Diethrich, Margolis M Pauliina, DB Reid, A Burke,

V Ramaiah, JA Rodriguez-Lopez, G Wheatley, D Olsen, and R Virmani, Vir-
tual histology intravascular ultrasound assessment of carotid artery disease: the
Carotid Artery Plaque Virtual Histology Evaluation (CAPITAL) study., Jour-
nal of endovascular therapy: an official journal of the International Society of
Endovascular Specialists, 14(5):676686, 2007.

[Doyley et al., 2006] Marvin M Doyley, Seshadri Srinivasan, Eugene Dimidenko,

Nirmal Soni, and Jonathan Ophir, Enhancing the performance of model-based
elastography by incorporating additional a priori information in the modulus
image reconstruction process, Physics in medicine and biology, 51(1):95, 2006.

[Doyley et al., 2005] Marvin M Doyley, Seshadri Srinivasan, Sarah A Pender-

grass, Ziji Wu, and Jonathan Ophir, Comparative evaluation of strain-based
and model-based modulus elastography, Ultrasound in medicine & biology,
31(6):787802, 2005.

[Doyley, 2012] MM Doyley, Model-based elastography: a survey of approaches

 156

to the inverse elasticity problem, Physics in medicine and biology, 57(3):R35,

2012.

[Doyley et al., 1996] MM Doyley, JC Bamber, T Shiina, and MO Leach, Recon-

struction of elastic modulus distribution from envelope detected B-mode data,
In Ultrasonics Symposium, 1996. Proceedings., 1996 IEEE, volume 2, pages
16111614. IEEE, 1996.

[Doyley et al., 2014] MM Doyley, B Jadamba, AA Khan, M Sama, and B Winkler,

A new energy inversion for parameter identification in saddle point problems
with an application to the elasticity imaging inverse problem of predicting tumor
location, Numerical Functional Analysis and Optimization, 35(7-9):9841017,
2014.

[Doyley et al., 2001] MM Doyley, Frits Mastik, CL De Korte, SG Carlier, EI Ce-

spedes, PW Serruys, Nicolaas Bom, and AFW Van der Steen, Advancing
intravascular ultrasonic palpation toward clinical applications, Ultrasound in
medicine & biology, 27(11):14711480, 2001.

[Doyley et al., 2000] MM Doyley, PM Meaney, and JC Bamber, Evaluation of

an iterative reconstruction method for quantitative elastography, Physics in
Medicine and Biology, 45(6):1521, 2000.

[Ebenstein et al., 2002] DM Ebenstein, JM Chapman, C Li, D Saloner, J Rapp,

and LA Pruitt, Assessing structure-property relations of diseased tissues using
nanoindentation and FTIR, In MATERIALS RESEARCH SOCIETY SYM-
POSIUM PROCEEDINGS, volume 711, pages 4752, 2002.

[Eisenbrey et al., 2012] John R Eisenbrey, Anush Sridharan, Marvin M Doyley,

and Flemming Forsberg, Parametric Subharmonic Imaging Using a Commer-
cial Intravascular Ultrasound Scanner An In Vivo Feasibility Study, Journal
of Ultrasound in Medicine, 31(3):361371, 2012.
 157

[Fayad et al., 2002] Zahi A Fayad, Valentin Fuster, Konstantin Nikolaou, and
Christoph Becker, Computed tomography and magnetic resonance imaging
for noninvasive coronary angiography and plaque imaging current and poten-
tial future concepts, Circulation, 106(15):20262034, 2002.

[Fromageau et al., 2007] Jeremie Fromageau, J-L Gennisson, CLMR Schmitt,

Roch L Maurice, Rosaire Mongrain, and Guy Cloutier, Estimation of polyvinyl
alcohol cryogel mechanical properties with four ultrasound elastography meth-
ods and comparison with gold standard testings, Ultrasonics, Ferroelectrics
and Frequency Control, IEEE Transactions on, 54(3):498509, 2007.

[Fung, 1993] YC Fung, Biomechanics: material properties of living tissues,

Springer, 1993.

[Furberg et al., 1994] Curt D Furberg, HP Adams, William B Applegate,

Robert P Byington, Mark A Espeland, Tyler Hartwell, Donald B Hunninghake,
David S Lefkowitz, Jeffrey Probstfield, and Ward A Riley, Effect of lovas-
tatin on early carotid atherosclerosis and cardiovascular events. Asymptomatic
Carotid Artery Progression Study (ACAPS) Research Group., Circulation,
90(4):16791687, 1994.

[Gaitini and Soudack, 2005] Diana Gaitini and Michalle Soudack, Diagnosing
Carotid Stenosis by Doppler Sonography State of the Art, Journal of ultra-
sound in medicine, 24(8):11271136, 2005.

[Garca et al., 2011] A Garca, MA Martnez, and E Pena, Viscoelastic proper-

ties of the passive mechanical behavior of the porcine carotid artery: influence
of proximal and distal positions., Biorheology, 49(4):271288, 2011.

[Garra et al., 1997] Brian S Garra, E Ignacio Cespedes, J Ophir, Stephen R

Spratt, Rebecca A Zuurbier, Colette M Magnant, and Marie F Pennanen, Elas-
 158

tography of breast lesions: initial clinical results., Radiology, 202(1):7986,

1997.

[Gasser and Holzapfel, 2007] T Christian Gasser and Gerhard A Holzapfel, Mod-
eling plaque fissuring and dissection during balloon angioplasty intervention,
Annals of biomedical engineering, 35(5):711723, 2007.

[Gasser et al., 2006] T Christian Gasser, Ray W Ogden, and Gerhard A Holzapfel,
Hyperelastic modelling of arterial layers with distributed collagen fibre orien-
tations, Journal of the royal society interface, 3(6):1535, 2006.

[Glagov et al., 1987] Seymour Glagov, Elliot Weisenberg, Christopher K Zarins,

Regina Stankunavicius, and George J Kolettis, Compensatory enlargement of
human atherosclerotic coronary arteries, New England Journal of Medicine,
316(22):13711375, 1987.

[Go et al., 2014] Alan S Go, Dariush Mozaffarian, Veronique L Roger, Emelia J
Benjamin, Jarett D Berry, Michael J Blaha, Shifan Dai, Earl S Ford, Caroline S
Fox, Sheila Franco, et al., Heart disease and stroke statistics2014 update: a
report from the American Heart Association., Circulation, 129(3):e28, 2014.

[Goenezen et al., 2012] Sevan Goenezen, J-F Dord, Zac Sink, Paul E Barbone,
Jingfeng Jiang, Timothy J Hall, and Assad A Oberai, Linear and nonlinear
elastic modulus imaging: an application to breast cancer diagnosis, Medical
Imaging, IEEE Transactions on, 31(8):16281637, 2012.

[Goertz et al., 2005] David E Goertz, Emmanuel Cherin, Andrew Needles, Raffi
Karshafian, Allison S Brown, Peter N Burns, and F Stuart Foster, High
frequency nonlinear B-scan imaging of microbubble contrast agents, Ultrason-
ics, Ferroelectrics, and Frequency Control, IEEE Transactions on, 52(1):6579,
2005.
 159

[Goertz et al., 2006a] David E Goertz, Martijn E Frijlink, Nico de Jong, and An-
tonius FW van der Steen, Nonlinear intravascular ultrasound contrast imag-
ing, Ultrasound in medicine & biology, 32(4):491502, 2006.

[Goertz et al., 2007] David E Goertz, Martijn E Frijlink, Dennie Tempel, Vijay
Bhagwandas, Andries Gisolf, Robert Krams, Nico de Jong, and Antonius FW
van der Steen, Subharmonic contrast intravascular ultrasound for vasa vaso-
rum imaging, Ultrasound in medicine & biology, 33(12):18591872, 2007.

[Goertz et al., 2006b] David E Goertz, Martijn E Frijlink, Dennie Tempel,

Luc CA van Damme, Robert Krams, Johannes A Schaar, J Folkert, Patrick W
Serruys, Nico de Jong, and Antonius FW van der Steen, Contrast harmonic
intravascular ultrasound: a feasibility study for vasa vasorum imaging, Inves-
tigative radiology, 41(8):631638, 2006.

[Goldberg and Deb, 1991] David E Goldberg and Kalyanmoy Deb, A compara-
tive analysis of selection schemes used in genetic algorithms, Urbana, 51:61801
2996, 1991.

[Golub et al., 1999] Gene H Golub, Per Christian Hansen, and Dianne P OLeary,
Tikhonov regularization and total least squares, SIAM Journal on Matrix
Analysis and Applications, 21(1):185194, 1999.

[Granada and Feinstein, 2008] Juan F Granada and Steven B Feinstein, Imaging
of the vasa vasorum, Nature Clinical Practice Cardiovascular Medicine, 5:S18
S25, 2008.

[Gundiah et al., 2007] Namrata Gundiah, Mark B Ratcliffe, and Lisa A Pruitt,
Determination of strain energy function for arterial elastin: experiments using
histology and mechanical tests, Journal of biomechanics, 40(3):586594, 2007.
 160

[Hamrock et al., 1999] Bernard J Hamrock, Bo O Jacobson, Steven R Schmid,

Bo Jacobson, and Bo Jacobson, Fundamentals of machine elements,
WCB/McGraw-Hill Singapore, 1999.

[Hansen et al., 2009] Hendrik HG Hansen, Richard GP Lopata, and Chris L

de Korte, Noninvasive carotid strain imaging using angular compounding
at large beam steered angles: validation in vessel phantoms, Medical Imaging,
IEEE Transactions on, 28(6):872880, 2009.

[Hansen et al., 2010a] Hendrik HG Hansen, Richard GP Lopata, Tim Idzenga,

and Chris L de Korte, An angular compounding technique using displacement
projection for noninvasive ultrasound strain imaging of vessel cross-sections,
Ultrasound in medicine & biology, 36(11):19471956, 2010.

[Hansen et al., 2013] Hendrik HG Hansen, Michael S Richards, Marvin M Doy-

ley, and Chris L de Korte, Noninvasive vascular displacement estimation for
relative elastic modulus reconstruction in transversal imaging planes, Sensors,
13(3):33413357, 2013.

[Hansen et al., 2010b] HHG Hansen, RGP Lopata, T Idzenga, and CL De Korte,
Full 2D displacement vector and strain tensor estimation for superficial tis-
sue using beam-steered ultrasound imaging, Physics in Medicine and biology,
55(11):3201, 2010.

[Hansen, 1998] Per Christian Hansen, Rank-deficient and discrete ill-posed prob-
lems: numerical aspects of linear inversion, volume 4, Siam, 1998.

[Haupt, 2002] Peter Haupt, Continuum mechanics and theory of materials,

Springer Science & Business Media, 2002.

[Hausleiter et al., 2009] Jorg Hausleiter, Tanja Meyer, Franziska Hermann, Mar-
tin Hadamitzky, Markus Krebs, Thomas C Gerber, Cynthia McCollough, Stefan
 161

Martinoff, Adnan Kastrati, Albert Schomig, et al., Estimated radiation dose

associated with cardiac CT angiography, Jama, 301(5):500507, 2009.

[Hayashi et al., 1980] K Hayashi, H Handa, S Nagasawa, A Okumura, and

K Moritake, Stiffness and elastic behavior of human intracranial and ex-
tracranial arteries, Journal of biomechanics, 13(2):175184, 1980.

[Hellings et al., 2010] Willem E. Hellings, Wouter Peeters, Frans L. Moll, Sebas-
tiaan R.D. Piers, Jessica van Setten, Peter J. Van der Spek, Jean-Paul P.M.
de Vries, Kees A. Seldenrijk, Peter C. De Bruin, Aryan Vink, Evelyn Velema,
Dominique P.V. de Kleijn, and Gerard Pasterkamp, Composition of carotid
atherosclerotic plaque is associated with cardiovascular outcome a prognostic
study, Circulation, 121(17):19411950, 2010.

[Herder et al., 2012] Marit Herder, Stein Harald Johnsen, Kjell Arne Arntzen, and
Ellisiv B Mathiesen, Risk Factors for Progression of Carotid Intima-Media
Thickness and Total Plaque Area A 13-Year Follow-Up Study: The Troms
Study, Stroke, 43(7):18181823, 2012.

[Hjermann et al., 1981] I Hjermann, I Holme, K Velve Byre, and P Leren, Effect
of diet and smoking intervention on the incidence of coronary heart disease:
report from the Oslo Study Group of a randomised trial in healthy men, The
Lancet, 318(8259):13031310, 1981.

[Hodis et al., 1998] Howard N Hodis, Wendy J Mack, Laurie LaBree, Robert H
Selzer, Chao-ran Liu, Ci-hua Liu, and Stanley P Azen, The role of carotid
arterial intima-media thickness in predicting clinical coronary events, Annals
of internal medicine, 128(4):262269, 1998.

[Holzapfel et al., 2000] Gerhard A Holzapfel, Thomas C Gasser, and Ray W Og-
den, A new constitutive framework for arterial wall mechanics and a compar-
 162

ative study of material models, Journal of elasticity and the physical science
of solids, 61(1-3):148, 2000.

[Holzapfel et al., 2005] Gerhard A Holzapfel, Gerhard Sommer, Christian T

Gasser, and Peter Regitnig, Determination of layer-specific mechanical prop-
erties of human coronary arteries with nonatherosclrotic intimal thickening and
related constitutive modeling, American Journal of Physiology-Heart and Cir-
culatory Physiology, 289(5):H2048H2058, 2005.

[Holzapfel and Weizsacker, 1998] Gerhard A Holzapfel and Hans W Weizsacker,

Biomechanical behavior of the arterial wall and its numerical characteriza-
tion, Computers in biology and medicine, 28(4):377392, 1998.

[Honda and Fitzgerald, 2008] Yasuhiro Honda and Peter J Fitzgerald, Frontiers
in intravascular imaging technologies, Circulation, 117(15):20242037, 2008.

[Hoyt et al., 2007] Kenneth Hoyt, Kevin J Parker, and Deborah J Rubens, Real-
time shear velocity imaging using sonoelastographic techniques, Ultrasound in
medicine & biology, 33(7):10861097, 2007.

[Huntzicker et al., 2014] Steven Huntzicker, Rohit Nayak, and Marvin M Doyley,
Quantitative sparse array vascular elastography: the impact of tissue atten-
uation and modulus contrast on performance, Journal of Medical Imaging,
1(2):027001027001, 2014.

[Inaba et al., 2012] Yoichi Inaba, Jennifer A Chen, and Steven R Bergmann,
Carotid plaque, compared with carotid intima-media thickness, more accu-
rately predicts coronary artery disease events: a meta-analysis, Atherosclero-
sis, 220(1):128133, 2012.

[Inglis et al., 2006] Scott Inglis, Kumar V Ramnarine, John N Plevris, and
W Norman McDicken, An anthropomorphic tissue-mimicking phantom of
 163

the oesophagus for endoscopic ultrasound, Ultrasound in medicine & biology,

32(2):249259, 2006.

[Jadamba et al., 2014] B Jadamba, AA Khan, G Rus, M Sama, and B Winkler,

A new convex inversion framework for parameter identification in saddle point
problems with an application to the elasticity imaging inverse problem of pre-
dicting tumor location, SIAM Journal on Applied Mathematics, 74(5):1486
1510, 2014.

[Jensen, 1991] Jrgen Arendt Jensen, A model for the propagation and scatter-
ing of ultrasound in tissue, Acoustical Society of America. Journal, 89(1):182
190, 1991.

[Jiang et al., 2013] Shan Jiang, Zhiliang Su, Xingji Wang, Sha Liu, and Yan Yu,
Development of a new tissue-equivalent material applied to optimizing surgical
accuracy, Materials Science and Engineering: C, 33(7):37683774, 2013.

[Jiang et al., 2011] Wei-Jian Jiang, Wengui Yu, Ning Ma, Bin Du, Xin Lou, and
Peter A Rasmussen, High resolution MRI guided endovascular intervention of
basilar artery disease, Journal of neurointerventional surgery, pages jnis2010,
2011.

[Kallel and Bertrand, 1996] F Kallel and M Bertrand, Tissue elasticity recon-
struction using linear perturbation method, Medical Imaging, IEEE Transac-
tions on, 15(3):299313, 1996.

[Kallel et al., 1999] Faouzi Kallel, R Jason Stafford, Roger E Price, Raffaella
Righetti, Jonathan Ophir, and John D Hazle, The feasibility of elastographic
visualization of HIFU-induced thermal lesions in soft tissues, Ultrasound in
medicine & biology, 25(4):641647, 1999.
 164

[Karaman et al., 1995] Mustafa Karaman, Pai-Chi Li, and Matthew ODonnell,
Synthetic aperture imaging for small scale systems, Ultrasonics, Ferroelectrics
and Frequency Control, IEEE Transactions on, 42(3):429442, 1995.

[Karimi et al., 2008] Reza Karimi, Ting Zhu, Brett E Bouma, and Mohammad
R Kaazempur Mofrad, Estimation of nonlinear mechanical properties of vas-
cular tissues via elastography, Cardiovascular Engineering, 8(4):191202, 2008.

[Katouzian et al., 2008] Amin Katouzian, Shashidhar Sathyanarayana, Babak

Baseri, Elisa E Konofagou, and Stephane G Carlier, Challenges in atheroscle-
rotic plaque characterization with intravascular ultrasound (IVUS): from data
collection to classification, Information Technology in Biomedicine, IEEE
Transactions on, 12(3):315327, 2008.

[Kawasaki et al., 1987] Takeshi Kawasaki, Shigetake Sasayama, Shin-Ichi Yagi,

Tetsuya Asakawa, and Tadakazu Hirai, Non-invasive assessment of the age
related changes in stiffness of major branches of the human arteries, Cardio-
vascular Research, 21(9):678687, 1987.

[Khalil et al., 2006] Ahmad S Khalil, Brett E Bouma, and Mohammad

R Kaazempur Mofrad, A combined FEM/genetic algorithm for vascular soft
tissue elasticity estimation, Cardiovascular Engineering, 6(3):93102, 2006.

[Khalil et al., 2005] Ahmad S Khalil, Raymond C Chan, Alexandra H Chau,

Brett E Bouma, and Mohammad R Kaazempur Mofrad, Tissue elasticity es-
timation with optical coherence elastography: toward mechanical characteriza-
tion of in vivo soft tissue, Annals of biomedical engineering, 33(11):16311639,
2005.

[Kim et al., 2004] Kang Kim, WF Weitzel, JM Rubin, Hua Xie, Xunchang Chen,
and M ODonnell, Vascular intramural strain imaging using arterial pressure
equalization, Ultrasound in medicine & biology, 30(6):761771, 2004.
 165

[Konofagou and Ophir, 1998] Elisa Konofagou and Jonathan Ophir, A new elas-
tographic method for estimation and imaging of lateral displacements, lateral
strains, corrected axial strains and Poissons ratios in tissues, Ultrasound in
medicine & biology, 24(8):11831199, 1998.

[Korukonda and Doyley, 2011a] Sanghamithra Korukonda and Marvin M Doyley,

3D ultrafast elastography imaging of the carotid artery using sparse arrays, In
Ultrasonics Symposium (IUS), 2011 IEEE International, pages 721724. IEEE,
2011.

[Korukonda and Doyley, 2011b] Sanghamithra Korukonda and Marvin M Doyley,

Estimating axial and lateral strain using a synthetic aperture elastographic
imaging system, Ultrasound in medicine & biology, 37(11):18931908, 2011.

[Korukonda and Doyley, 2012] Sanghamithra Korukonda and Marvin M Doyley,

Visualizing the radial and circumferential strain distribution within vessel
phantoms using synthetic-aperture ultrasound elastography, Ultrasonics, Fer-
roelectrics and Frequency Control, IEEE Transactions on, 59(8):16391653,
2012.

[Korukonda et al., 2013] Sanghamithra Korukonda, Rohit Nayak, Nancy Carson,

Giovanni Schifitto, Vikram Dogra, and Marvin M Doyley, Noninvasive vascular
elastography using plane-wave and sparse-array imaging., IEEE transactions
on ultrasonics, ferroelectrics, and frequency control, 60(2):332342, 2013.

[Krouskop et al., 1998] Thomas A Krouskop, Thomas M Wheeler, Faouzi Kallel,

Brian S Garra, and Timothy Hall, Elastic moduli of breast and prostate tissues
under compression, Ultrasonic imaging, 20(4):260274, 1998.

[Kubo et al., 2007] Takashi Kubo, Toshio Imanishi, Shigeho Takarada, Akio
Kuroi, Satoshi Ueno, Takashi Yamano, Takashi Tanimoto, Yoshiki Matsuo,
 166

Takashi Masho, Hironori Kitabata, et al., Assessment of Culprit Lesion Mor-

phology in Acute Myocardial InfarctionAbility of Optical Coherence Tomog-
raphy Compared With Intravascular Ultrasound and Coronary Angioscopy,
Journal of the American College of Cardiology, 50(10):933939, 2007.

[Kwon et al., 1998] Hyuck Moon Kwon, Giuseppe Sangiorgi, Erik L Ritman,
Charles McKenna, David R Holmes Jr, Robert S Schwartz, and Amir Ler-
man, Enhanced coronary vasa vasorum neovascularization in experimental
hypercholesterolemia., Journal of Clinical Investigation, 101(8):1551, 1998.

[Lally et al., 2004] C Lally, AJ Reid, and PJ Prendergast, Elastic behavior of

porcine coronary artery tissue under uniaxial and equibiaxial tension, Annals
of biomedical engineering, 32(10):13551364, 2004.

[Larsson et al., 2015] Matilda Larsson, Peter Verbrugghe, Marija Smoljkic, Jelle
Verhoeven, Brecht Heyde, Nele Famaey, Paul Herijgers, and Jan Dhooge,
Strain assessment in the carotid artery wall using ultrasound speckle tracking:
validation in a sheep model, Physics in medicine and biology, 60(3):1107, 2015.

[Laurent et al., 2006] Stephane Laurent, John Cockcroft, Luc Van Bortel, Pierre
Boutouyrie, Cristina Giannattasio, Daniel Hayoz, Bruno Pannier, Charalambos
Vlachopoulos, Ian Wilkinson, and Harry Struijker-Boudier, Expert consensus
document on arterial stiffness: methodological issues and clinical applications,
European heart journal, 27(21):25882605, 2006.

[Le Floch et al., 2010] Simon Le Floch, Guy Cloutier, Gerard Finet, Philippe
Tracqui, Roderic I Pettigrew, and Jacques Ohayon, On the potential of a new
IVUS elasticity modulus imaging approach for detecting vulnerable atheroscle-
rotic coronary plaques: in vitro vessel phantom study, Physics in medicine
and biology, 55(19):5701, 2010.
 167

[Le Floch et al., 2012] Simon Le Floch, Guy Cloutier, Yoshifumi Saijo, Gerard
Finet, Saami K Yazdani, Flavien Deleaval, Gilles Rioufol, Roderic I Pettigrew,
and Jacques Ohayon, A four-criterion selection procedure for atherosclerotic
plaque elasticity reconstruction based on in vivo coronary intravascular ultra-
sound radial strain sequences, Ultrasound in medicine & biology, 38(12):2084
2097, 2012.

[Le Floch et al., 2009] Simon Le Floch, Jacques Ohayon, Philippe Tracqui,
Gerard Finet, Ahmed M Gharib, Roch L Maurice, Guy Cloutier, Roderic I Pet-
tigrew, et al., Vulnerable atherosclerotic plaque elasticity reconstruction based
on a segmentation-driven optimization procedure using strain measurements:
theoretical framework, Medical Imaging, IEEE Transactions on, 28(7):1126
1137, 2009.

[Learoyd and Taylor, 1966] Brian M Learoyd and Michael G Taylor, Alterations
with age in the viscoelastic properties of human arterial walls, Circulation
research, 18(3):278292, 1966.

[Lee et al., 1992] Richard T Lee, S Geoffrey Richardson, Howard M Loree, Alan J
Grodzinsky, Sina A Gharib, Frederick J Schoen, and Natesa Pandian, Predic-
tion of mechanical properties of human atherosclerotic tissue by high-frequency
intravascular ultrasound imaging. An in vitro study., Arteriosclerosis, Throm-
bosis, and Vascular Biology, 12(1):15, 1992.

[Lehmann et al., 1998] Eldon D Lehmann, Kathleen D Hopkins, Ataullah

Rawesh, Ramina C Joseph, Kadiadia Kongola, Simon W Coppack, and Ray-
mond G Gosling, Relation between number of cardiovascular risk fac-
tors/events and noninvasive Doppler ultrasound assessments of aortic compli-
ance, Hypertension, 32(3):565569, 1998.

[Lerner et al., 1988] Robert M Lerner, Kevin J Parker, Jarle Holen, Raymond
 168

Gramiak, and Robert C Waag, Sono-elasticity: medical elasticity images de-

rived from ultrasound signals in mechanically vibrated targets, In Acoustical
imaging, pages 317327. Springer, 1988.

[Levinson et al., 1995] Stephen F Levinson, Masahiko Shinagawa, and Takuso

Sato, Sonoelastic determination of human skeletal muscle elasticity, Journal
of biomechanics, 28(10):11451154, 1995.

[Liang et al., 2008] Yun Liang, Hui Zhu, and Morton H Friedman, Estimation of
the transverse strain tensor in the arterial wall using IVUS image registration,
Ultrasound in medicine & biology, 34(11):18321845, 2008.

[Libby, 1995] Peter Libby, Molecular bases of the acute coronary syndromes,
Circulation, 91(11):28442850, 1995.

[Libby et al., 2002] Peter Libby, Paul M Ridker, and Attilio Maseri, Inflamma-
tion and atherosclerosis, Circulation, 105(9):11351143, 2002.

[Little et al., 1988] William C Little, M Constantinescu, RJ Applegate,

MA Kutcher, MT Burrows, FR Kahl, and WP Santamore, Can coronary
angiography predict the site of a subsequent myocardial infarction in patients
with mild-to-moderate coronary artery disease?, Circulation, 78(5):11571166,
1988.

[Loree et al., 1994] Howard M Loree, Alan J Grodzinsky, Susan Y Park, Lorna J
Gibson, and Richard T Lee, Static circumferential tangential modulus of
human atherosclerotic tissue, Journal of biomechanics, 27(2):195204, 1994.

[Loree et al., 1992] Howard M Loree, RD Kamm, RG Stringfellow, and RT Lee,

Effects of fibrous cap thickness on peak circumferential stress in model
atherosclerotic vessels., Circulation Research, 71(4):850858, 1992.
 169

[Lu et al., 2015] Jinqiu Lu, Wanying Duan, and Aike Qiao, Finite element analy-
sis of mechanics of neovessels with intraplaque hemorrhage in carotid atheroscle-
rosis, BioMedical Engineering OnLine, 14(Suppl 1):S3, 2015.

[Lubinski et al., 1996] Mark A Lubinski, Stanislav Y Emelianov, KR Raghavan,

Andrew E Yagle, Andrei R Skovoroda, and Matthew ODonnell, Lateral dis-
placement estimation using tissue incompressibility, IEEE Transactions on
Ultrasonics Ferroelectrics and Frequency Control, 43(2):247256, 1996.

[Lyshchik et al., 2007] Andrej Lyshchik, Tatsuya Higashi, Ryo Asato, Shinzo
Tanaka, Juichi Ito, Masahiro Hiraoka, Michael F Insana, Aaron B Brill, Tsu-
neo Saga, and Kaori Togashi, Cervical Lymph Node Metastases: Diagnosis at
SonoelastographyInitial Experience 1, Radiology, 243(1):258267, 2007.

[Ma et al., 2014] Jianguo Ma, K Heath Martin, Paul A Dayton, and Xiaon-
ing Jiang, A preliminary engineering design of intravascular dual-frequency
transducers for contrast-enhanced acoustic angiography and molecular imag-
ing, Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions
on, 61(5):870880, 2014.

[Maehara et al., 2009] Akiko Maehara, Gary S Mintz, and Neil J Weissman, Ad-
vances in intravascular imaging, Circulation: Cardiovascular Interventions,
2(5):482490, 2009.

[Maresca et al., 2012] David Maresca, Krista Jansen, Guillaume Renaud,

G Van Soest, X Li, Q Zhou, N De Jong, KK Shung, and AFW Van der
Steen, Intravascular ultrasound chirp imaging, Applied Physics Letters,
100(4):043703, 2012.

[Maresca et al., 2013] David Maresca, Guillaume Renaud, Gijs van Soest, Xiang
Li, Qifa Zhou, K Kirk Shung, Nico de Jong, and Antonius FW van der Steen,
 170

Contrast-enhanced intravascular ultrasound pulse sequences for bandwidth-

limited transducers, Ultrasound in medicine & biology, 39(4):706713, 2013.

[Maresca et al., 2014] David Maresca, Ilya Skachkov, Guillaume Renaud, Krista
Jansen, Gijs van Soest, Nico de Jong, and Antonius FW van der Steen, Imaging
Microvasculature with Contrast-Enhanced Ultraharmonic Ultrasound, Ultra-
sound in medicine & biology, 40(6):13181328, 2014.

[Marquardt, 1963] Donald W Marquardt, An algorithm for least-squares estima-

tion of nonlinear parameters, Journal of the Society for Industrial & Applied
Mathematics, 11(2):431441, 1963.

[Maurice et al., 2005a] Roch L Maurice, Elisabeth Brusseau, Gerard Finet, and
Guy Cloutier, On the potential of the Lagrangian speckle model estimator
to characterize atherosclerotic plaques in endovascular elastography: in vitro
experiments using an excised human carotid artery, Ultrasound in medicine
& biology, 31(1):8591, 2005.

[Maurice et al., 2005b] Roch L Maurice, Michel Daronat, Jacques Ohayon,

Ekatherina Stoyanova, F Stuart Foster, and Guy Cloutier, Non-invasive high-
frequency vascular ultrasound elastography, Physics in medicine and biology,
50(7):1611, 2005.

[Maurice et al., 2004] Roch L Maurice, Jacques Ohayon, Gerard Finet, and Guy
Cloutier, Adapting the Lagrangian speckle model estimator for endovascular
elastography: Theory and validation with simulated radio-frequency data, The
Journal of the Acoustical Society of America, 116(2):12761286, 2004.

[McCormick et al., 2012] M McCormick, T Varghese, X Wang, C Mitchell,

MA Kliewer, and RJ Dempsey, Methods for robust in vivo strain estimation
in the carotid artery, Physics in medicine and biology, 57(22):7329, 2012.
 171

[McGarry et al., 2013] Matthew McGarry, C Johnson, B Sutton, J Georgiadis,

E Van Houten, J Weaver, and K Paulsen, Including Spatial Information in
Nonlinear Inversion MR Elastography Using Soft Prior Regularization., Med-
ical Imaging, IEEE Transaction on, 2013.

[Mehrabian and Samani, 2009] Hatef Mehrabian and Abbas Samani, Con-
strained hyperelastic parameters reconstruction of PVA (Polyvinyl Alcohol)
phantom undergoing large deformation, In SPIE Medical Imaging, pages
72612G72612G. International Society for Optics and Photonics, 2009.

[Mojabi and LoVetri, 2009] Puyan Mojabi and Joe LoVetri, Overview and clas-
sification of some regularization techniques for the Gauss-Newton inversion
method applied to inverse scattering problems, Antennas and Propagation,
IEEE Transactions on, 57(9):26582665, 2009.

[Mol and Breddels, 1982] Chris R Mol and Paul A Breddels, Ultrasound velocity
in muscle, The Journal of the Acoustical Society of America, 71(2):455461,
1982.

[Mooney, 1940] M Mooney, A theory of large elastic deformation, Journal of

applied physics, 11(9):582592, 1940.

[Moore et al., 1998] MP Moore, T Spencer, DM Salter, PP Kearney, TRD Shaw,

IR Starkey, PJ Fitzgerald, R Erbel, A Lange, NW McDicken, et al., Char-
acterisation of coronary atherosclerotic morphology by spectral analysis of ra-
diofrequency signal: in vitro intravascular ultrasound study with histological
and radiological validation, Heart, 79(5):459467, 1998.

[Moritake et al., 1974] K Moritake, H Handa, A Okumura, K Hayashi, and H Ni-

imi, Stiffness in Cerebral Arteries, Neurologia, 14(1):47, 1974.

[Moselewski et al., 2004] Fabian Moselewski, Dieter Ropers, Karsten Pohle, Udo
Hoffmann, Maros Ferencik, Ray C Chan, Ricardo C Cury, Suhny Abbara, Ik-
 172

kyung Jang, Thomas J Brady, et al., Comparison of measurement of cross-

sectional coronary atherosclerotic plaque and vessel areas by 16-slice multide-
tector computed tomography versus intravascular ultrasound, The American
journal of cardiology, 94(10):12941297, 2004.

[Moulton et al., 1999] Karen S Moulton, Eric Heller, Moritz A Konerding, Evelyn
Flynn, Wulf Palinski, and Judah Folkman, Angiogenesis inhibitors endostatin
or TNP-470 reduce intimal neovascularization and plaque growth in apolipopro-
tein Edeficient mice, Circulation, 99(13):17261732, 1999.

[Naghavi et al., 2003] Morteza Naghavi, Peter Libby, Erling Falk, S Ward Cass-
cells, Silvio Litovsky, John Rumberger, Juan Jose Badimon, Christodoulos Ste-
fanadis, Pedro Moreno, Gerard Pasterkamp, et al., From vulnerable plaque
to vulnerable patient a call for new definitions and risk assessment strategies:
part I, Circulation, 108(14):16641672, 2003.

[Nair et al., 2001] Anuja Nair, Barry D Kuban, Nancy Obuchowski, and D Ge-
offrey Vince, Assessing spectral algorithms to predict atherosclerotic plaque
composition with normalized and raw intravascular ultrasound data, Ultra-
sound in medicine & biology, 27(10):13191331, 2001.

[Nieman et al., 2002] Koen Nieman, Filippo Cademartiri, Pedro A Lemos, Rolf
Raaijmakers, Peter MT Pattynama, and Pim J de Feyter, Reliable noninva-
sive coronary angiography with fast submillimeter multislice spiral computed
tomography, Circulation, 106(16):20512054, 2002.

[Nissen and Yock, 2001] Steven E Nissen and Paul Yock, Intravascular ultra-
sound novel pathophysiological insights and current clinical applications, Cir-
culation, 103(4):604616, 2001.

[Oberai et al., 2004] Assad A Oberai, Nachiket H Gokhale, Marvin M Doyley,

 173

and Jeffrey C Bamber, Evaluation of the adjoint equation based algorithm for
elasticity imaging, Physics in Medicine and Biology, 49(13):2955, 2004.

[Oberai et al., 2009] Assad A Oberai, Nachiket H Gokhale, Sevan Goenezen,

Paul E Barbone, Timothy J Hall, Amy M Sommer, and Jingfeng Jiang, Lin-
ear and nonlinear elasticity imaging of soft tissue in vivo: demonstration of
feasibility, Physics in medicine and biology, 54(5):1191, 2009.

[Ophir et al., 1991] J Ophir, I Cespedes, Hm Ponnekanti, Y Yazdi, and X Li,

Elastography: a quantitative method for imaging the elasticity of biological
tissues, Ultrasonic imaging, 13(2):111134, 1991.

[Owen et al., 2011] DRJ Owen, AC Lindsay, RP Choudhury, and ZA Fayad,

Imaging of atherosclerosis, Annual review of medicine, 62:25, 2011.

[Parker et al., 2011] KJ Parker, MM Doyley, and DJ Rubens, Imaging the elastic
properties of tissue: the 20 year perspective, Physics in medicine and biology,
56(1):R1, 2011.

[Parker et al., 1990] KJ Parker, SR Huang, RA Musulin, and RM Lerner, Tissue

response to mechanical vibrations for sonoelasticity imaging, Ultrasound in
medicine & biology, 16(3):241246, 1990.

[Patel and Janicki, 1970] Dali J Patel and Joseph S Janicki, Static elastic prop-
erties of the left coronary circumflex artery and the common carotid artery in
dogs, Circulation research, 27(2):149158, 1970.

[Pavan et al., 2012] Theo Z Pavan, Ernest L Madsen, Gary R Frank, Jingfeng
Jiang, Antonio AO Carneiro, and Timothy J Hall, A nonlinear elasticity
phantom containing spherical inclusions, Physics in medicine and biology,
57(15):4787, 2012.
 174

[Pazos et al., 2009] V Pazos, R Mongrain, and JC Tardif, Polyvinyl alcohol

cryogel: optimizing the parameters of cryogenic treatment using hyperelastic
models, Journal of the mechanical behavior of biomedical materials, 2(5):542
549, 2009.

[Perego et al., 2011] Mauro Perego, Alessandro Veneziani, and Christian Vergara,
A variational approach for estimating the compliance of the cardiovascular tis-
sue: An inverse fluid-structure interaction problem, SIAM journal on scientific
computing, 33(3):11811211, 2011.

[Prati et al., 2010] Francesco Prati, Evelyn Regar, Gary S Mintz, Eloisa Arbus-
tini, Carlo Di Mario, Ik-Kyung Jang, Takashi Akasaka, Marco Costa, Giulio
Guagliumi, Eberhard Grube, et al., Expert review document on methodology,
terminology, and clinical applications of optical coherence tomography: phys-
ical principles, methodology of image acquisition, and clinical application for
assessment of coronary arteries and atherosclerosis, European heart journal,
31(4):401415, 2010.

[Raj and Ramasamy, 1983] P.P. Raj and V. Ramasamy, Strength of Materials,
Dhanpat Rai and Sons, 1983.

[Rao et al., 2007] M Rao, Q Chen, H Shi, T Varghese, EL Madsen, JA Zagzebski,

and TA Wilson, Normal and shear strain estimation using beam steering on
linear-array transducers, Ultrasound in medicine & biology, 33(1):5766, 2007.

[Richards and Doyley, 2013] Michael S Richards and Marvin M Doyley, Non-
rigid image registration based strain estimator for intravascular ultrasound elas-
tography, Ultrasound in medicine & biology, 39(3):515533, 2013.

[Richards et al., 2015] Michael S Richards, Renato Perucchio, and Marvin M Doy-
ley, Visualizing the Stress Distribution Within Vascular Tissues Using In-
 175

travascular Ultrasound Elastography: A Preliminary Investigation, Ultrasound

in medicine & biology, 2015.

[Richards and Doyley, 2011] MS Richards and MM Doyley, Investigating the

impact of spatial priors on the performance of model-based IVUS elastography,
Physics in medicine and biology, 56(22):7223, 2011.

[Rivlin, 1948] RS Rivlin, Large elastic deformations of isotropic materials. IV.

Further developments of the general theory, Philosophical Transactions of
the Royal Society of London. Series A, Mathematical and Physical Sciences,
241(835):379397, 1948.

[Rossmann and Dym, 2008] Jenn Stroud Rossmann and Clive L Dym, An In-
troduction to Engineering Mechanics: A Continuum Approach, CRC Press,
2008.

[Saam et al., 2009] T Saam, MF Reiser, and K Nikolaou, Imaging of the carotid
atherosclerotic plaque with 3T MRI using dedicated 4-channel surface coils, J
Cardiovasc Magn Reson, 11:41, 2009.

[Samani et al., 2001] Abbas Samani, Jonathan Bishop, and Donald B Plewes, A
constrained modulus reconstruction technique for breast cancer assessment,
Medical Imaging, IEEE Transactions on, 20(9):877885, 2001.

[Schaar et al., 2003] Johannes A Schaar, Chris L de Korte, Frits Mastik, Chaylen-
dra Strijder, Gerard Pasterkamp, Eric Boersma, Patrick W Serruys, and An-
ton FW van der Steen, Characterizing vulnerable plaque features with in-
travascular elastography, Circulation, 108(21):26362641, 2003.

[Schmitt et al., 2007] Cedric Schmitt, Gilles Soulez, Roch L Maurice, Marie-
France Giroux, and Guy Cloutier, Noninvasive vascular elastography: toward
a complementary characterization tool of atherosclerosis in carotid arteries,
Ultrasound in medicine & biology, 33(12):18411858, 2007.
 176

[Schwan and FitzHugh, 1969] Herman P Schwan and Richard FitzHugh, Biolog-
ical engineering, volume 9, McGraw-Hill Companies, 1969.

[Shekhar et al., 2014a] Himanshu Shekhar, Ivy Awuor, Sahar Hashemgeloogerdi,

and Marvin M Doyley, Nonlinear intravascular ultrasound contrast imaging
with a modified clinical system, The Journal of the Acoustical Society of
America, 135(4):23092309, 2014.

[Shekhar et al., 2014b] Himanshu Shekhar, Ivy Awuor, Steven Huntzicker, and
Marvin M Doyley, Ultraharmonic intravascular ultrasound imaging with com-
mercial 40 MHz catheter: A feasibility study, The Journal of the Acoustical
Society of America, 136(4):23022302, 2014.

[Shekhar et al., 2014c] Himanshu Shekhar, Ivy Awuor, Keri Thomas, Joshua J
Rychak, and Marvin M Doyley, The Delayed Onset of Subharmonic and
Ultraharmonic Emissions from a Phospholipid-Shelled Microbubble Contrast
Agent, Ultrasound in medicine & biology, 40(4):727738, 2014.

[Shekhar and Doyley, 2012] Himanshu Shekhar and Marvin M Doyley, Improv-
ing the sensitivity of high-frequency subharmonic imaging with coded excita-
tion: A feasibility study, Medical physics, 39(4):20492060, 2012.

[Shekhar and Doyley, 2013] Himanshu Shekhar and Marvin M Doyley, The re-
sponse of phospholipid-encapsulated microbubbles to chirp-coded excitation:
Implications for high-frequency nonlinear imaging, The Journal of the Acous-
tical Society of America, 133(5):31453158, 2013.

[Simon et al., 1973] BR Simon, AS Kobayashi, CA Wiederhielm, and DE Strand-

ness, Deformation of the arterial vasa vasorum at normal and hypertensive
arterial pressure, Journal of biomechanics, 6(4):349359, 1973.
 177

[Sinkus et al., 2000] R Sinkus, J Lorenzen, D Schrader, M Lorenzen, M Dargatz,

and D Holz, High-resolution tensor MR elastography for breast tumour de-
tection, Physics in medicine and biology, 45(6):1649, 2000.

[Sinkus et al., 2005] R Sinkus, M Tanter, S Catheline, J Lorenzen, C Kuhl, E Son-

dermann, and M Fink, Imaging anisotropic and viscous properties of breast
tissue by magnetic resonance-elastography, Magnetic resonance in medicine,
53(2):372387, 2005.

[Skjaerpe et al., 1985] T Skjaerpe, L Hegrenaes, and L Hatle, Noninvasive esti-

mation of valve area in patients with aortic stenosis by Doppler ultrasound and
two-dimensional echocardiography., Circulation, 72(4):810818, 1985.

[Sridharan et al., 2013] Anush Sridharan, John R. Eisenbrey, Priscilla Machado,

Ebo D. deMuinck, Marvin M. Doyley, and Flemming Forsberg, Delineation
of atherosclerotic plaque using subharmonic imaging filtering techniques and a
commercial intravascular ultrasound system, Ultrasonic imaging, 35(1):3044,
2013.

[Srinivasan et al., 2004] R Srinivasan, S Chakrabarti, T Walsh, T Igarashi,

Y Takahashi, D Kleiner, T Donohue, R Shalabi, C Carvallo, AJ Barrett, et al.,
Improved survival in steroid-refractory acute graft versus host disease after
non-myeloablative allogeneic transplantation using a daclizumab-based strat-
egy with comprehensive infection prophylaxis, British journal of haematology,
124(6):777786, 2004.

[Staub et al., 2010] Daniel Staub, Mita B Patel, Anjan Tibrewala, David Lud-
den, Mahala Johnson, Paul Espinosa, Blai Coll, Kurt A Jaeger, and Steven B
Feinstein, Vasa vasorum and plaque neovascularization on contrast-enhanced
carotid ultrasound imaging correlates with cardiovascular disease and past car-
diovascular events, Stroke, 41(1):4147, 2010.
 178

[Surry et al., 2004] KJM Surry, HJB Austin, A Fenster, and TM Peters, Poly
(vinyl alcohol) cryogel phantoms for use in ultrasound and MR imaging,
Physics in medicine and biology, 49(24):5529, 2004.

[Taylor et al., 2000] LS Taylor, BC Porter, DJ Rubens, and KJ Parker, Three-

dimensional sonoelastography: principles and practices, Physics in medicine
and biology, 45(6):1477, 2000.

[Techavipoo et al., 2004] Udomchai Techavipoo, Quan Chen, Tomy Varghese,

and James A Zagzebski, Estimation of displacement vectors and strain ten-
sors in elastography using angular insonifications, Medical Imaging, IEEE
Transactions on, 23(12):14791489, 2004.

[Teng et al., 2012] Zhongzhao Teng, Jing He, Andrew J Degnan, Shengyong
Chen, Umar Sadat, Nasim Sheikh Bahaei, James HF Rudd, and Jonathan H
Gillard, Critical mechanical conditions around neovessels in carotid
atherosclerotic plaque may promote intraplaque hemorrhage, Atherosclero-
sis, 223(2):321326, 2012.

[Vaitkevicius et al., 1993] Peter V Vaitkevicius, Jerome L Fleg, James H Engel,

Frances C OConnor, Jeanette G Wright, Loretta E Lakatta, FC Yin, and
Edward G Lakatta, Effects of age and aerobic capacity on arterial stiffness in
healthy adults., Circulation, 88(4):14561462, 1993.

[Vale et al., 2012] Thiago Cardoso Vale, Ricardo Oliveira Horta Maciel, Debora
Maia, Rogerio Beato, and Francisco Cardoso, Takayasus Arteritis in a Pa-
tient with Sydenhams Chorea: is There an Association?, Tremor and Other
Hyperkinetic Movements, 2, 2012.

[Varghese and Ophir, 1997] Tomy Varghese and Jonathan Ophir, A theoreti-
cal framework for performance characterization of elastography: the strain fil-
 179

ter, Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on,

44(1):164172, 1997.

[Varghese et al., 2003] Tomy Varghese, Udomchai Techavipoo, Wu Liu, James A

Zagzebski, Quan Chen, Gary Frank, and Fred T Lee Jr, Elastographic mea-
surement of the area and volume of thermal lesions resulting from radiofre-
quency ablation: pathologic correlation, American journal of roentgenology,
181(3):701707, 2003.

[Vavuranakis et al., 2008] Manolis Vavuranakis, Ioannis A Kakadiaris, Sean M

OMalley, Theodore G Papaioannou, Elias A Sanidas, Morteza Naghavi,
Stephane Carlier, Dimitrios Tousoulis, and Christodoulos Stefanadis, A new
method for assessment of plaque vulnerability based on vasa vasorum imag-
ing, by using contrast-enhanced intravascular ultrasound and differential image
analysis, International journal of cardiology, 130(1):2329, 2008.

[Vengrenyuk et al., 2006] Yuliya Vengrenyuk, Stephane Carlier, Savvas Xanthos,

Luis Cardoso, Peter Ganatos, Renu Virmani, Shmuel Einav, Lane Gilchrist, and
Sheldon Weinbaum, A hypothesis for vulnerable plaque rupture due to stress-
induced debonding around cellular microcalcifications in thin fibrous caps,
Proceedings of the National Academy of Sciences, 103(40):1467814683, 2006.

[Versluis et al., 2006] Antheunis Versluis, Alan J Bank, and William H Douglas,
Fatigue and plaque rupture in myocardial infarction, Journal of biomechan-
ics, 39(2):339347, 2006.

[Virmani et al., 2003] Renu Virmani, Allen P Burke, Frank D Kolodgie, and An-
drew Farb, Pathology of the Thin-Cap Fibroatheroma, Journal of interven-
tional cardiology, 16(3):267272, 2003.

[Virmani et al., 2000] Renu Virmani, Frank D Kolodgie, Allen P Burke, Andrew
Farb, and Stephen M Schwartz, Lessons from sudden coronary death a com-
 180

prehensive morphological classification scheme for atherosclerotic lesions, Ar-

teriosclerosis, thrombosis, and vascular biology, 20(5):12621275, 2000.

[Virmani et al., 2005] Renu Virmani, Frank D Kolodgie, Allen P Burke, Aloke V
Finn, Herman K Gold, Thomas N Tulenko, Steven P Wrenn, and Jagat Narula,
Atherosclerotic plaque progression and vulnerability to rupture angiogenesis as
a source of intraplaque hemorrhage, Arteriosclerosis, thrombosis, and vascular
biology, 25(10):20542061, 2005.

[Vogel, 2002] Curtis R Vogel, Computational methods for inverse problems, vol-
ume 23, Siam, 2002.

[Wagner et al., 1983] Robert F Wagner, Stephen W Smith, John M Sandrik, and
Hector Lopez, Statistics of speckle in ultrasound B-scans., IEEE TRANS.
SONICS ULTRASONICS., 30(3):156163, 1983.

[Walsh et al., 2014] MT Walsh, EM Cunnane, JJ Mulvihill, AC Akyildiz, FJH Gi-

jsen, and Gerhard A Holzapfel, Uniaxial tensile testing approaches for charac-
terisation of atherosclerotic plaques, Journal of biomechanics, 47(4):793804,
2014.

[Wellman et al., 1999] Parris Wellman, Robert D Howe, Edward Dalton, and
Kenneth A Kern, Breast tissue stiffness in compression is correlated to his-
tological diagnosis, Harvard BioRobotics Laboratory Technical Report, pages
115, 1999.

[Wellman et al., 2001] Parris S Wellman, Edward P Dalton, David Krag, Ken-
neth A Kern, and Robert D Howe, Tactile imaging of breast masses: first
clinical report, Archives of surgery, 136(2):204208, 2001.

[Wenk et al., 2010] Jonathan F Wenk, Panayiotis Papadopoulos, and Tarek I Zo-
hdi, Numerical modeling of stress in stenotic arteries with microcalcifica-
 181

tions: a micromechanical approximation, Journal of biomechanical engineer-

ing, 132(9):091011, 2010.

[Widman et al., 2014] Erik Widman, Kenneth Caidahl, and Matilda Larsson,
In vivo radial and longitudinal carotid artery plaque strain estimation via
ultrasound-based speckle tracking, In Ultrasonics Symposium (IUS), 2014
IEEE International, pages 523526. IEEE, 2014.

[Wittek et al., 2013] Andreas Wittek, Konstantinos Karatolios, Peter Bihari,

Thomas Schmitz-Rixen, Rainer Moosdorf, Sebastian Vogt, and Christopher
Blase, In vivo determination of elastic properties of the human aorta based
on 4D ultrasound data, Journal of the mechanical behavior of biomedical ma-
terials, 27:167183, 2013.

[Yalavarthy et al., 2007a] Phaneendra K Yalavarthy, Brian W Pogue, Hamid De-

hghani, Colin M Carpenter, Shudong Jiang, and Keith D Paulsen, Structural
information within regularization matrices improves near infrared diffuse optical
tomography, Optics Express, 15(13):80438058, 2007.

[Yalavarthy et al., 2007b] Phaneendra K Yalavarthy, Brian W Pogue, Hamid De-

hghani, and Keith D Paulsen, Weight-matrix structured regularization pro-
vides optimal generalized least-squares estimate in diffuse optical tomography,
Medical physics, 34(6):20852098, 2007.

[Zhang et al., 2006] Yong Zhang, Lawrence O Hall, Dmitry B Goldgof, and
Sudeep Sarkar, A constrained genetic approach for computing material prop-
erty of elastic objects, Evolutionary Computation, IEEE Transactions on,
10(3):341357, 2006.

[Zhao et al., 2009] Xihai Zhao, Zachary E Miller, and Chun Yuan, Atheroscle-
rotic plaque imaging by carotid MRI, Current cardiology reports, 11(1):7077,
2009.
 182

[Zhdanov, 2002] Michael S Zhdanov, Geophysical inverse theory and regulariza-

tion problems, volume 36, Elsevier, 2002.

[Zhou and Fung, 1997] J Zhou and YC Fung, The degree of nonlinearity and
anisotropy of blood vessel elasticity, Proceedings of the National Academy of
Sciences, 94(26):1425514260, 1997.