Performance Analysis and Optimization of 2-D Cardiac Strain Imaging for Clinical Applications

Ethan Bunting

Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Graduate School of Arts and Science

COLUMBIA UNIVERSITY

2017

Heart disease has remained the deadliest disease in the United States for the past 100 years.

Imaging methods are frequently employed in cardiology in order to help clinicians diagnose the specific type of heart disease and to guide treatment decisions. Ultrasound is the most frequently used imaging modality in cardiology because it is inexpensive, portable, easy to use, and extremely safe for patients. Using a variety of imaging processing techniques, deformations exhibited by the cardiac tissue during contraction can be imaged with ultrasound and used as an indicator of myocardial health.

This dissertation will demonstrate the clinical implementation of two ultrasound-based strain estimation techniques developed in the Ultrasound and Elasticity Imaging Laboratory at

Columbia University. Each of the two imaging methods will be tailored for clinical applications using techniques for optimal strain estimation derived from ultrasound and imaging processing theory. The motion estimation rate (MER) used for strain estimation is examined in the context of the theoretical Strain Filter and used to increase the precision of axial strain estimation.

Diverging beam sequences are used to achieve full-view high MER imaging within a single heartbeat. At approximately 500 Hz, the expected elastographic signal-to-noise ratio (E(SNRe|ε)) of the axial strain becomes single-peaked, indicating an absence of “peak-hopping” errors which can severely corrupt strain estimation. In order to mediate the tradeoff in spatial resolution resulting from the use of diverging beams, coherent spatial compounding is used to increase the accuracy of the lateral strain estimation, resulting in a more physiologic strain profile. A sequence with 5 coherently compounded diverging waves is used at 500 Hz to improve the radial

SNRe of the strain estimation compared to a single-source diverging sequence at 500 Hz.

The first technique, Myocardial Elastography (ME), is used in conjunction with an intracardiac echocardiography (ICE) system to image the formation of thermal ablation lesions in vivo using a canine model (n=6). By comparing the systolic strain before and after the formation of a lesion, lesion maps are generated which allow for the visualization of the lesion in real-time during the procedure. A good correlation is found between the lesion maps and the actual lesion volume as measured using gross pathology (r2=0.86). The transmurality of the lesions are also shown to be in good agreement with gross pathology. Finally, the feasibility of imaging gaps between neighboring lesions is established. Lesion size and the presence of gaps have been associated with the success rate of cardiac ablation procedures, demonstrating the value of ME as a potentially useful technique for clinicians to help improve patient outcomes following ablation procedures.

The second technique, Electromechanical Wave Imaging (EWI), is implemented using a transthoracic echocardiography system in a study of heart failure patients (n=16) and healthy subjects (n=5). EWI uses the transient inter-frame strains to generate maps of electromechanical activation, which are then used to distinguish heart failure patients from healthy controls (p<.05).

EWI was also shown to be capable of distinguishing responders from non-responders to cardiac resynchronization therapy (CRT) on the basis of the activation time of the lateral wall. These results indicate that EWI could be used as an adjunct tool to monitor patient response to CRT, in addition to helping guide lead placement prior to device implantation.

TABLE OF CONTENTS

List of Figures iv

Acknowledgements viii

Chapter 1. Introduction 1

1.1 Motivation 1

1.2 Specific Aims 3

1.3 Overview and Significance 4

Chapter 2. The Heart 6

2.1 Anatomy and Function 6

2.2 Excitation-Contraction Coupling 10

2.2.1 Electrical Activation 10

2.2.2 Mechanical Activation 13

2.2.3 Electromechanical Activation 14

2.3 Heart Disease 15

2.3.1 Overview 15

2.3.2 Atrial Fibrillation 16

2.3.3 Heart Failure 17

Chapter 3. Cardiac Imaging Modalities 20

3.1 Mechanical Evaluation of the Heart 20

3.2 Myocardial Elastography 23

3.2.1 Introduction 23

3.2.2 Methods 24

3.3 Electrical Evaluation of the Heart 27

3.4 Electromechanical Wave Imaging 29

3.4.1 Introduction 29

3.4.2 Methods 31

Chapter 4. Optimizing Echocardiographic Strain Imaging 34

4.1 Introduction 34

4.2 Displacement Estimation 35

4.3 Strain Estimation 37

4.4 Study: Motion Estimation Rate 39

4.4.1 Introduction 39

4.4.2 Methods 42

4.4.3 Results 46

4.4.4 Discussion 50

4.5 Study: Coherent Spatial Compounding 57

4.5.1 Introduction 57

4.5.2 Methods 58

4.5.3 Results 62

4.5.4 Discussion 68

4.6 Conclusion: Specific Aim 1 72

Chapter 5. Myocardial Elastography for Cardiac Ablation Monitoring 74

5.1 Introduction 74

5.2 Methods 76

5.3 Results 80

5.3.1 Lesion Size and Transmurality 80

5.3.2 Lesion Gap Imaging 81

5.4 Discussion 84

5.4 Conclusion: Specific Aim 2 88

Chapter 6. Clinical Applications of Electromechanical Wave Imaging 89

6.1 Introduction 89

6.2 Methods 93

6.3 Results 97

6.4 Discussion 102

6.5 Conclusion: Specific Aim 3 107

Chapter 7. Summary of Work 109

References 111

List of Publications 125

iii

LIST OF FIGURES

Figure 2-2: Wiggers diagram depicting the pressure and volume of the atria and ventricles in two cardiac cycles. Figure licensed under the Creative Commons Attribution-Share Alike 2.5 Generic license.

Figure 2-3: (A) Typical waveform of the cardiac action potential. (B) The summation of the each individual cardiac action potential is measured by surface electrodes to form the electrocardiogram (ECG).

Figure 2-4: Typical ECG tracing of a healthy human adult. P wave indicates atrial depolarization, QRS complex indicates ventricular depolarization and atrial repolarization, and T wave indicates ventricular repolarization.

Figure 2-5: The LV geometry is often modeled as a prolate ellipsoid and referred to in a polar coordinate system with radial, circumferential, and longitudinal directions. Arrows indicate the general direction of contraction.

Figure 2-6: Electromechanical activation of the heart. Cardiac cells are first electrically activated by the action potential before experiencing mechanical activation by cell shortening. The delay between electrical and mechanical activation is the electromechanical delay.

Figure 3-1: Brief overview of the methods for Myocardial Elastography. (A) A diverging beam sequence to capture element data at high frame-rates. (B) Delay-and-sum beamforming to create RF image frames. (C) 1-D or 2-D displacement estimation by normalized cross-correlation. (D) Inter-frame motion tracked and accumulated over systole. (E) Least-squares algorithm to convert cumulative displacements to cumulative strains.

Figure 3-2: Typical ECG tracings for (A) normal sinus rhythm, (B) atrial fibrillation, and (C) left bundle branch block.

Figure 3-3: Brief overview of the methods for Electromechanical Wave Imaging. (A) A diverging beam sequence to capture element data at high frame-rates. (B) 1-D inter-frame displacement estimation using normalized cross-correlation. (C) Inter-frame strain estimation using a least squares strain estimator. (D) Selection of the temporal zero-crossings corresponding to electromechanical activation. (E) Mapping the timings of zero-crossings to the cardiac tissue.

Figure 4-1: Depiction of correlation-based displacement estimation at two separate locations within the same line. The relative displacement between the two points will be reflected by a change in strain. Figure 4-2: Overview of the sequences used for the MER study. (A) Acoustic beam profile for the TUAS (left) and DBS (right). Temporal profiles for (B) the conventional sequence, (C)

TUAS, and (D) DBS. Each line in the B-D corresponds to a single transmission of the corresponding beam shown in A.

Figure 4-3: Conditional pdfs of axial strains taken with the TUAS at 544 Hz (left) and DBS at 500 Hz (right) in a single patient. Horizontal lines denoting the transition zones which separate the CRLB from the BB have been drawn at the local minima of the pdfs. Figure 4-4: Comparison of axial and lateral inter-frame strain estimations for TUAS and DBS during LV contraction in the same subject. ECG tracing is shown below the images indicating the timing within the cardiac cycle. Axial and lateral strains appear similar at similar stages of contraction.

Figure 4-5: Expected SNRe curves for axial and lateral strain estimation using the TUAS and DBS at various MERs in 5 healthy subjects.

Figure 4-6: Expected SNRe curves for axial (left) and radial (right) strain estimation using the TUAS and DBS at similar MERs.

Figure 4-7: Description of the hybrid diverging beam acquisition sequence used for the compounding study. First, a series of 5 compounding waves are emitted using a reduced sub- aperture with equally spaced source locations along the transducer. Immediately following, a series of 5 diverging waves is emitted using the full aperture and the same source for each transmit.

Figure 4-8: Reconstructed B-mode images of the LV at end-diastole using the diverging and the compounding acquisitions. Anatomic regions of the LV wall are denoted (ant=anterior, lat=lateral, sept=septal, post=posterior).

Figure 4-9: Example cumulative end-systolic axial and lateral displacements for the diverging and compounding acquisitions in two subjects

Figure 4-10: Example cumulative end-systolic axial and lateral strains for the diverging and compounding acquisitions in two subjects.

Figure 4-11: Cumulative radial strain at end-systole for all subjects using the diverging and compounding sequences.

Figure 4-12: Comparison between the diverging and compounding sequences of radial SNRe over time for each of the four subjects.

Figure 5-1: ICE catheter position within the right ventricle.

Figure 5-2: Methods for determining lesion volume and transmurality based on strain computed from (A) intracardiac ME and (B) TTC staining.

Figure 5-3: End-systolic strain distribution obtained in the canine LV (A) before ablation and (B) after ablation. (C) Comparison of strain change before and after the lesion is used to generate a (D) lesion map. (E) Excised lesion with TTC staining and (F) trichrome staining (F).

Figure 5-4: Area of strain reduction (>8%) vs. excised lesion volume measured after TTC staining.

Figure 5-5: Bland-Altman agreement between lesion transmurality measured using ME strain and gross pathology.

Figure 5-6: End-systolic strain distribution obtained in the canine LV after the formation of a two-lesion line having no gaps. Cumulative end-systolic strain is shown (A) before ablation and (B) 2 minutes after ablation, along with (C) the thresholded lesion map in comparison to (D) gross pathology.

Figure 5-7: End-systolic strain distribution obtained in the canine LV after the formation of a three-lesion line having small gaps between lesions. Cumulative end-systolic strain is shown (A) before ablation and (B) 2 minutes after ablation, along with (C) the thresholded lesion map in comparison to (D) gross pathology.

Figure 6-1: Comparison of the strain computed using EWI and traditional strain imaging methods. EWI (blue) captures small inter-frame strains and can precisely identify the strain zero- crossing in response to electromechanical activation (orange circle), while traditional strain imaging (green) computes the global total strain and often relies on time to peak strain (red circle).

Figure 6-2: Ventricular activation isochrones from healthy volunteers. Activation times are mapped to the ECG displayed below the images (total time = 200 ms).

Figure 6-3: Ventricular activation isochrones from three HF patients identified as responders to CRT. Isochrones are shown for native rhythm (a) and biventricular pacing (b). Activation times are mapped to the ECG displayed below the images (total time = 200 ms). Patient I has been diagnosed with ischemic cardiomyopathy while Patients II and III have been diagnosed with non-ischemic cardiomyopathy.

Figure 6-4: Ventricular activation isochrones from HF patients identified as non-responders to CRT. Isochrones are shown for (a) native rhythm and (b) biventricular pacing. Activation times are mapped to the ECG displayed below the images (total time = 200 ms). Patients I and II have been diagnosed with ischemic cardiomyopathy while Patient III has been diagnosed with non- ischemic cardiomyopathy.

Figure 6-5: Cumulative statistics of LWAT for all normal subjects (n=5) and HF patients (n=15). HF patients have also been separated into responders (n=7) and non-responders (n=8). * indicates p<0.01 by unpaired or paired Student’s t test where appropriate.

Figure 6-6: Bland-Altman plot of LWAT reproducibility for consecutive EWI acquisitions of the same patient (n=6) in the same protocol.

Figure 6-7: Mean change in QRSd and LWAT within each patient as a result of CRT.

vii

ACKNOWLEDGEMENTS

The Ph.D. journey has been fairly long and there are many people to thank. Even though I will try to keep this section short, it will definitely not reflect that amount of support I have received from so many different people.

First and foremost, I would like to thank my advisor, Elisa Konofagou. She has not only taught me an immense amount about engineering, but also how to view research with good mix of optimism and practicality. It was a privilege to be a part of such a well-respected and international lab, and the many amazing experiences I have had abroad would not be possible without her. In a world of stuffy academics, she is always a breath of fresh air in every room she walks into. She is also one of the most hard-working PIs that I have seen at any institution, and I hope I can be as productive as her in my future career.

I would also like to thank my collaborators in the Ultrasound and Elasticity Imaging Laboratory, particularly within the cardiac group. Jean Provost and Clement Papadacci are two of the best engineers I have ever met and always took the time to patiently explain things to me when I was lost. Julien Grondin was my desk neighbor and frequent lunch companion, but what I will remember him most for is his unfailing willingness to help me, even when I was blatantly interrupting his work. Alexandre Costet will go down for sure as one of the most important people I have met at Columbia, since it was him who opened my eyes to many alternative viewpoints about the world. The basketball talk I had with Jason Apostolakis, Ronny Li, and

Thomas Payen were invaluable in keeping me sane throughout the entire process. Pierre Nauleau was a phenomenal addition to our little corner in the lab, and his steadfast patience with me during surgery was a big reason why we had so many successful experiments. Finally, I’d like to thank Vince Sayseng and Lea Melki, the young bucks in the cardiac group. They did an excellent

viii

job supporting Alex and I as we were finishing up and I can’t wait to see what they have in store in their remaining years.

When I came to Columbia, I knew I wanted to do research with actual human patients. I could not have done that without the help of several very talented cardiologists: Dr. Elaine Wan, Dr.

Litsa Lambrakos, and Dr. Alok Gambhir. All of them took significant time out of their busy schedules to help guide me through the intricacies of patient studies. Khady Fall and Fusako Sera were both gracious enough to help scan patients for me and provided me with excellent images.

Special thanks to Alok and Elaine for participating in multiple, day-long experiments and still coming back for more. Speaking of those experiments, I would also like to thank Dr. Peter

Danilo and Dr. Alex Romanov for teaching me how to perform surgery and keep a heart beating.

I would also like to express my gratitude to my dissertation committee members, Dr. Andrew

Laine, Dr. Christoph Juchem, Dr. Elaine Wan, and Dr. Dan Wang. All of them have provided me with extremely useful feedback throughout the dissertation process.

Outside of Columbia, I would like to thank all my friends for just being themselves. It was through their support that I was able to tough it out through the hard times of the Ph.D., when experiments aren’t working and there seemed to be no end in sight.

Finally, I would like to thank my parents, the WaveKing and WaveQueen. I think they are the best parents in the world. Suffice it to say that without them I would not have had the great life I have had so far, and I definitely would not have finished my Ph.D. and become the scientist I dreamed of as a kid.

This thesis is dedicated to the best parents in the world, Lee and Criss Bunting

Chapter 1 Introduction

1.1 Motivation

Heart disease remains the leading cause of death in the United States, accounting for approximately 1 in 3 deaths each year1. Approximately 38% of all Americans have some form of heart disease, a number which is expected to continue to increase with life expectancy. Heart disease also poses a severe economic burden and is projected to cost the U.S. healthcare system more than $800 million yearly by 20302. Technologies for cardiac imaging are uniquely involved in both the diagnosis and treatment of heart disease and are highly integrated into current clinical practice. For instance, a single interventional cardiology procedure such as radiofrequency ablation could simultaneously require fluoroscopy, electroanatomic (EA) mapping, electrocardiogram (ECG), and echocardiography to diagnose and treat an arrhythmia during surgery. Cardiac imaging technologies are thus expected to play an important role in the future of healthcare.

Of the major medical imaging modalities, only MRI and ultrasound do not require the use of potentially harmful ionizing radiation. In terms of financial cost to the patient, ultrasound is by far the least expensive imaging procedure3. Echocardiography is therefore a highly desirable imaging modality from a patient perspective. At the same time, the portability and ease-of-use of ultrasound machines have led to widespread adoption among cardiologists, establishing ultrasound as the most frequently used imaging modality in clinical cardiology. The main drawback of ultrasound is the apparent decrease in image quality and spatial resolution compared

to computed tomography (CT) or magnetic resonance imaging (MRI). However, ultrasound is still frequently used to quantify the cardiac structure, as ejection fraction and left-ventricular volume evaluated using echocardiography remain important markers of cardiac disease. As will be discussed in detail in this dissertation, there is also a great wealth of functional information captured using ultrasound that cannot be appreciated by visual inspection of the standard B-mode image.

A major goal of functional cardiac imaging is to identify the presence and location of abnormalities within the heart. This information can be used to make diagnoses and to guide future treatment approaches. Functional imaging of the heart is challenging due to the relationship between the initial electrical activation and subsequent mechanical contraction of the heart. Cardiac electrical activity is frequently a target for therapeutic approaches and abnormalities can be measured with a high degree of accuracy using the ECG. However, the

ECG provides only a global depiction of cardiac electrical activity which can be insufficient for targeted therapies such as radiofrequency ablation and cardiac resynchronization therapy. EA contact mapping is a more precise imaging technique widely used in interventional cardiology, but this procedure requires an invasive device, several hours of point-by-point acquisition, and integration with CT imaging. Cardiac mechanical activity is most often evaluated using qualitative inspection of the ultrasound B-mode cine-loop, i.e. wall motion scoring. Image processing techniques such as speckle tracking and tissue Doppler can be used to quantify the mechanical deformations experienced by the heart during contraction. Although many indices of mechanical function have been developed using these techniques, the clinical reliability of these indices is still unclear4,5.

In general, clinical utilization of ultrasonic strain imaging techniques has been slowed by reproducibility concerns4,6. However, there is vast potential for ultrasound imaging technologies due to the widespread use of echocardiography within cardiology. By establishing a link between ultrasound theory and clinical practice, ultrasound strain estimation techniques may be developed at higher precision and accuracy than what is currently available in commercial systems.

1.2. Specific Aims

The specific aims of this dissertation are as follows:

SPECIFIC AIM 1: Characterize the effects of motion estimation rate and spatial compounding on two-dimensional cardiac strain imaging in order to design optimal acquisition sequences for clinical use.

Hypotheses

 Increases in MER will lead to an increase in 2-D estimation precision assessed using

SNRe

 Spatial compounding will lead to an increase in 2-D estimation accuracy assessed using

the physiologic strain distribution in the healthy heart

SPECIFIC AIM 2: Investigate the utility of Myocardial Elastography (ME) in assessing lesion formation during RF ablation of cardiac tissue.

Hypotheses

 Myocardial Elastography will be able to detect the presence of an ablation lesion by

depicting the associated strain variation in the ablated region

 Significant variations in strain will correlate with larger lesion volume.

SPECIFIC AIM 3: Investigate the utility of Electromechanical Wave Imaging (EWI) in imaging heart failure and response to CRT pacing.

Hypotheses

 EWI is capable of detecting an increase in ventricular electromechanical activation in

heart failure patients compared to healthy volunteers.

 EWI is capable of detecting a reduction in electromechanical activation in heart failure

patients as a result of CRT pacing.

 Reduction in electromechanical activation is significantly higher in for responders to

CRT compared to non-responders.

1.3 Overview and Significance

The overall goal of this dissertation is to advance the techniques of Myocardial Elastography

(ME) and Electromechanical Wave Imaging (EWI) out of the laboratory and into the clinical environment of cardiology. This goal will be achieved in two ways; first, by optimizing the imaging sequences used for ME and EWI, and second by tailoring the various image processing elements to fit specific clinical applications. Imaging sequences refer to the characteristics of the acoustic waves transmitted by the ultrasound probe. The effects of the frame-rate, or temporal resolution, of the imaging sequence will be investigated. Furthermore, the effects of spatial compounding on the spatial resolution of imaging sequence will also be investigated. The

fundamental tradeoff that exists in ultrasound between temporal resolution and spatial resolution will be an important theme throughout the sequence optimization process. Although these optimization methods will serve to increase the quality of the ultrasound data and subsequent strain estimation, applications of these methods to real clinical problems is an entirely different pursuit. In this dissertation, the techniques of Myocardial Elastography and Electromechanical

Wave Imaging will be applied specifically to two important clinical conditions: atrial fibrillation and heart failure. Each of these disease states affects a very large patient population within the

United States and worldwide, allowing the work presented here to have far-reaching implications across many specialties of medicine.

Chapter 2

The Heart

2.1 Anatomy and Function

The heart is a major organ in the human body and is the driving force behind the cardiovascular system. Cardiac tissue is primarily composed of muscle cells called cardiomyocytes which undergo periodic shortening-lengthening cycles in order to provide mechanical contractions to pump blood throughout the body7. The pumping mechanism of the heart serves two key purposes. First, it allows deoxygenated blood entering the right side of the heart to pass through the pulmonary system into the lungs where oxygenation occurs. Secondly, it produces a cyclical, driving pressure gradient that forces the oxygenated blood out of the heart and into the rest of the body’s organs. All tissues in the body—including the brain, skin, and even the heart itself—rely on an oxygenated blood supply for metabolic purposes. Failure to receive adequate blood supply can rapidly lead to loss of function and possibly necrosis of the tissue.

The cardiac structure is composed of four chambers, each of which serves a particular function in the overall pumping mechanism (Figure 2-1). The two chambers located on the top of the heart are referred to as the atria, while the bottom two chambers are called the ventricles. The atria are responsible for receiving blood into the heart from the circulatory system and pumping this blood down into the ventricles. The ventricles are then responsible for the subsequent ejection of this blood out of the heart. Flows between the atria and ventricle are controlled by two atrioventricular valves, which open and close and regular intervals during normal cardiac

Figure 2-1: Anatomical depiction of the heart showing the Purkinje system (blue) and its subsequent branching in the ventricles. RA = Right Atrium, LA = Left Atrium, RV = Right Ventricle, LV =Left Ventricle function. Similarly, two semilunar valves also exist between the ventricles and the circulatory system. Due to the short distance and low resistance between the atria and ventricles, the atrial walls are relatively thin compared to the ventricular walls, which need to pump blood using a much greater force to overcome the high resistance of the long, cylindrical vessels of the circulatory system.

In addition to a top-bottom (atrial-ventricular) organization, the heart can also be classified on a right-left basis. The right side of the heart receives deoxygenated blood from the venous portion of the systemic circulation, while the left side of the heart handles the oxygenated blood received

from the pulmonary circulation. Deoxygenated blood enters the heart via the right atrium (RA) and is passed to the right ventricle (RV) through the tricuspid valve. The RV then pumps the blood through the pulmonary valve into the pulmonary circulation, where it participates in oxygen exchange at the capillary-alveolus interface in the lungs. The newly oxygenated blood then enters into the left atrium (LA) and is pumped through the mitral valve into the left ventricle

(LV). Finally, LV contraction forces the blood out through the aortic valve into the aorta, the first major vessel of the arterial system. Since the LV is the final chamber in the pipeline to push blood into the circulation, it possesses the thickest and strongest walls and is often considered to be the most clinically-important chamber. The inner walls of all chambers of the heart are referred to as the endocardium, while the outer walls are referred to as the epicardium.

The cyclical pumping mechanism of the heart can be divided into two distinct phases in time, systole and diastole. Systole refers to the active ventricular contraction of the cardiac that squeezes blood into the circulation. Systole begins with the closing of the atrioventricular valves, forming a closed system in each of the two ventricles (the semilunar valves are also closed at this point). Isovolumetric contraction of the ventricular tissue then occurs, causing a rapid increase in ventricular cavity pressure. The increased cavity pressure with respect to the circulation provides the pressure gradient responsible for blood flow out of the ventricles. Isovolumetric contraction occurs for less than 100 ms until the semilunar valves are opened and blood is forced into the circulation. The end of systole occurs with the closing of the semilunar valves after a certain fraction of blood has been expelled from the ventricles. The diastolic phase then follows, which immediately begins with the opening of the atrioventricular valves to allow blood to pass from the atria into the ventricles. During diastole, ventricular filling is achieved both by the suction

created by ventricular relaxation and by active contraction of the atrial tissue. Systole and diastole are typically defined with respect to ventricular mechanics, which is why systole is often thought of as the “active contraction” phase and diastole is thought of as the “passive filling” phase. However, it is important to note that instances of contraction and relaxation occur at alternating times in the atria compared to the ventricles. The time course of the various cardiac pressures and volumes is displayed in Figure 2-2 in a multifaceted plot known as a Wiggers diagram8.

2.2 Excitation-Contraction Coupling

The previous section introduced the concept of the heart as a pump that induces a pressure gradient to drive the flow of blood throughout the circulation. In this section, the origin of the cardiac pumping action is examined at a much smaller, cellular level. Like other muscle cells in the human body, cardiomyocytes must be activated by an external electrical signal called an action potential in order to contract. The cardiac action potential refers to a change in cardiomyocyte membrane voltage that causes the cell to contract. The electrical depolarization produced by the action potential leads directly to mechanical changes within the cell that cause contraction7.

2.2.1 Electrical Activation

Initiation of the cardiac action potential is performed by specialized cardiomyocytes called pacemaker cells. A cluster of these pacemaker cells is located in the right atrium and referred to as the sinoatrial (SA) node. Pacemaker cells possess the unique property of being able to spontaneously generate a cardiac action potential without requiring innervation from the nervous system. The major mechanism for propagation of the cardiac action potential throughout the heart is the specialized cardiac conduction system known as the Purkinje network9 (Figure 2-1).

The Purkinje network contains specialized conducting cells that allow for rapid propagation of the cardiac action potential. The main pathway of the Purkinje network begins in the SA node and travels to the atrioventricular junction at the Bundle of His, which slows conduction for a brief moment. Exiting the Bundle of His into the ventricles, the Purkinje network then separates into the right and left bundle branches, which serve to innervate the right and left ventricles, respectively. There are many smaller Purkinje fibers branching off of the main network, which

allows for activation of the entire myocardium. Action potential propagation is also possible outside the Purkinje system via cellular gap junctions10.

Changes in the membrane voltage of cardiomyocytes form the action potential (AP) waveform shown in Figure 2-3a. The cardiac AP is composed of five phases, each of which is defined by the opening and closing of sodium, potassium, and calcium channels located within the cell membrane11:

Phase 0: rapid membrane depolarization caused by opening of fast sodium channels Phase 1: initial repolarization caused by closing of fast sodium channels Phase 2: plateau phase sustained by calcium channels and potassium channels Phase 3: rapid repolarization caused by closing of calcium channels Phase 4: resting membrane potential dominated by potassium channels

Compared to APs in skeletal muscle, the cardiac AP lasts for a much longer duration (~200 ms).

The absolute refractory period for the cardiac action potential, which lasts from phase 0 to phase

3, is thus relatively long, forming the basis for many types of arrhythmia12,13. Indeed, the

mechanisms of many common anti-arrhythmic drugs are related to the conduction and

propagation of the cardiac AP14.

The electrocardiogram (ECG) provides a high-level picture of cardiac electrical activity by

recording the global summation of the individual cardiac APs using body surface electrodes

(Figure 2-3b)15.

The ECG is an extremely important tool in cardiology and emergency medicine and more than

100 arrhythmias have been categorized based on their appearance on the ECG16. The normal

waveform of the ECG for a full cardiac cycle is displayed in Figure 2-4. Each deflection of the

signal corresponds to particular activation sequence within the heart and is labeled using a letter.

The first deflection is referred to as the P wave and represents the depolarization of the atria. The

high amplitude signal defined by the Q, R, and S waves is collectively referred to as the QRS complex and corresponds to depolarization of the ventricles in addition to the repolarization of the atria. Finally, the T wave corresponds to the repolarization of the ventricles.

2.2.2 Mechanical Activation

Mechanical contraction of the cardiomyocytes begins when the cardiac action potential causes a potential increase from the resting membrane potential (-90 mV) above the threshold value of approximately -40 mV. Intracellular proteins (myosin and actin) interact to pull the longitudinal ends of the cardiomyocyte towards each other, serving to both “shorten” and “thicken” the individual cells. Because each cardiomyocyte contracts in a similar fashion, the alignment of cardiac cells is an important determinant of mechanical contraction. In the LV, cardiomyocytes are generally aligned depending on their location within the myocardial wall17.

The shape of the LV is often modeled as half of a prolate spheroid with a polar coordinate system (Figure 2-5). In the cardiac polar coordinate system, the radial and circumferential directions are defined relative to the center of a circular slice through the LV cavity, while the longitudinal direction refers to the long axis of the heart from apex to base. Cardiomyocytes are generally aligned in the longitudinal direction near the epicardium and undergo a gradual rotation towards circumferentially-aligned fibers at the endocardium17. As a result, the global contraction of the LV manifests as shortening in the longitudinal and circumferential directions and lengthening in the radial direction.

The volume of the LV cavity at peak contraction is referred to as the LV end-systolic volume

(LVESV). This volume is decreased compared to resting volume of the LV, which is defined as the LV end-diastolic volume (LVEDV), as a result of ventricular contraction. The magnitude of this reduction in volume is referred to as the LV ejection fraction (LVEF) and is an important marker of cardiac health. In the clinic, echocardiography is frequently used to measure the LV dimensions in order to estimate LVEDV, LVESV, and LVEF18. In a healthy human adult, LVEF is approximately 60%19.

2.2.3 Electromechanical Activation

The interaction between the electrical and mechanical aspects of the heart is critical to its functionality. Electrical activation of the cardiomyocytes always occurs before mechanical activation, with the intervening time interval being known as the electromechanical delay. The electromechanical delay has been estimated at 20-50 ms20,21.

2.3 Heart Disease

2.3.1 Overview

Heart disease is a very general term that refers to a wide variety of conditions that can disrupt the normal function of the heart. Diseases of the heart represent a significant health burden and are the number one cause of death in the United States. Furthermore, heart disease is major precursor to stroke, the fifth leading cause of death in the United States. The origin and progression of heart disease is different in each patient and results from a combination of factors, including age, diet, exercise, genetics, substance abuse, and environment. In their 2016 annual report, the

American Heart Association (AHA) identified family history, high blood cholesterol, high blood

pressure, and diabetes as the most important risk factors in heart disease. Significant classifications of heart disease mentioned in this report include both atrial fibrillation and heart failure1.

2.3.2 Atrial Fibrillation

Atrial fibrillation (AF) is defined by chaotic electrical activation of the atria, which prevents efficient contraction and can lead to impaired cardiac function22,23. The direct symptoms of AF are generally mild and include chest palpitations and presyncope (lightheadedness). More importantly, AF is linked to the formation of thromboembolisms (blood clots) which can travel to the brain and cause stroke. Patients with AF are five times more likely to experience a stroke and are at increased risk for all-cause mortaliy23,24. Each year, more than 400,000 people in the

United States are hospitalized as a result of AF, making it the most common arrhythmia which requires treatment23,25.

Initial treatment for AF is usually confined to antiarrhythmic drugs such as amiodarone; however, these are ineffective in approximately half of all patients26. As a result, a procedure known as radiofrequency (RF) ablation is often used to surgically correct the arrhythmia. RF ablation involves placing a catheter within the heart to generate thermal lesions in regions of the myocardium thought to contribute to the chaotic activation. Most often, thermal lesions are generated around the pulmonary veins in the left atrium in a technique known as pulmonary vein isolation. Electroanatomic mapping systems are often used to identify targets for ablation27; however, this process involves point-by-point contact mapping using a separate catheter which can add hours to the procedure. Success rates for ablation procedures have been reported in the

range of 53-57% for a single procedure and 71-80% after multiple procedures26,28. The size, spacing, and depth of the lesions in the tissue has been shown to be critical to the success of an ablation procedure and preventing AF recurrence29–32. Conventionally, lesion size has been controlled using temperature, amplitude and duration of the RF energy provided by the ablation system33,34. Contact force of the ablation catheter has also recently been proposed as a method to control lesion size35,36. However, these methods rely on indirect feedback from the catheter system instead of direct assessment of the lesion size. There is currently no method in the clinic able to directly image the lesion formation in real time.

2.3.3 Heart Failure

Systolic heart failure (HF) refers to a state of impaired ventricular ejection of blood. As a result, heart failure is often accompanied by a reduced ejection fraction (<40%) and may result in fatigue, pulmonary edema, and eventually sudden cardiac death. There are a vast amount of potential causes for HF, including obesity, hypertension, infarction, and LV hypertrophy. A common indication of HF is a conduction abnormality known as left bundle branch block

(LBBB) which is currently diagnosed using the ECG. LBBB involves a partial or complete block of the left bundle branch of the Purkinje network which innervates the LV. As a result, activation of the LV is delayed, resulting in a dyssynchrony between the right- and left-sided contractions of the heart. Diagnosis of systolic HF is normally performed using the ECG in addition to conventional ultrasound. Heart failure, and particularly LBBB, manifest as a prolongation of the

QRS complex on the ECG, while a reduction in ejection fraction can be observed using a standard echocardiogram.

The most commonly-used classification system for HF is the New York Heart Association

(NYHA) classification system37. The NYHA divides HF into four classifications based on the symptomatic presentation of patients. Stage I HF is often simply treated with exercise and diet recommendations. These recommendations continue into Stage II and can be accompanied by some pharmacological therapy. Stage III and IV HF are accompanied by severe symptoms that often cannot be managed with lifestyle changes and drugs alone. In these patients, cardiac resynchronization therapy (CRT) may be recommended to limit the progression of the disease to avoid the end-stage treatment of cardiac transplant. In fact, CRT is the only currently- recommended treatment that is able to actually reverse the progression of HF and lead to improvements in symptoms37. CRT requires the use of an implanted biventricular pacemaker having leads located in both ventricles in addition to the standard atrial lead. Using simultaneous pacing of the RV and LV, CRT hopes to resynchronize the ventricles contractions of the heart.

Although CRT has been shown to lead to extensive benefits in patients38,39, approximately 30-

50% of CRT patients receive no benefit and are considered non-responders4,40. This high rate of non-response is especially concerning considering the invasive nature and high cost of CRT.

There are various estimates for the true non-response rate to CRT, although most of the literature places this number at approximately 30-50%4,40. This rate of non-response is particularly troublesome due to the high cost and invasiveness of the treatment, and in light of expanded indications to a larger group of HF patients40,41. Electrical measures of dyssynchrony obtained using the ECG are now used to select patients for CRT, but their relationship to CRT response may be limited. QRS duration (QRSd) provides only a general prediction of CRT response and clinical trials have shown poor correlation between QRSd and outcome42. Mechanical

dyssynchrony parameters have shown promise in determining response in small observational studies, although large, multi-center trials have failed to validate the predictive power of these approaches. The largest prospective trial (PROSPECT) showed poor predictive value of several echocardiographic parameters derived using tissue Doppler4. Speckle-tracking echocardiography has been used to quantify dyssynchrony associated with response to CRT43, although recent studies using speckle tracking dyssynchrony measures to predict response to CRT have shown mixed results44,45.

Recent advancements in CRT device technology have motivated the investigation of new pacing strategies for CRT. Leads with multiple implanted electrodes allow for the selection of different pacing locations and for multi-point pacing. Several studies have reported the utility of targeted

LV lead placement, using the site of latest electrical activation46–48 or a region that is remote from scar49. Other studies have shown that the offset between LV and RV pacing can be optimized for the individual patient to improve hemodynamics and functional characteristics of the LV50,51. Multi-point pacing of the LV may also be useful over the long-term to improve the rate of response to CRT52.

Chapter 3

Cardiac Imaging Modalities

Because all body tissues require oxygenated blood to survive, the maintenance of cardiac function is critical to ensure patient quality of life and longevity. As such, evaluation of cardiac function is routinely performed by medical professionals in patients with both acute and chronic conditions. Since the heart functions as an electromechanical pump—using electrical activation to induce mechanical contraction—traditional methods to assess cardiac behavior are mainly targeted at either the electrical or the mechanical side. In this chapter, we will discuss the most commonly-used diagnostic equipment in the clinic for both mechanical and electrical behavior.

In addition, we will present two novel ultrasound-based technologies, Myocardial Elastography

(ME) and Electromechanical Wave Imaging (EWI), which carry great potential for future use in cardiology as adjunct tools to assess the mechanical and electrical behaviors of the heart respectively.

3.1 Mechanical Evaluation of the Heart

The mechanical function of the heart is closely related to the ability to pump blood to the rest of the body. One of the most common symptoms of heart disease is fatigue, which is in many cases related to the fact that the heart cannot provide enough blood to sufficiently oxygenate vital body tissues53. Thus, the mechanical functionality of the heart is of great interest to clinicians and patients.

Ultrasound applied to the heart is referred to as echocardiography and is used frequently in the clinic due to its ease of use, low cost, and portability. Echocardiography provides clinicians with a convenient tool to image both the structure and function of the cardiac tissue in real time, allowing for visual evaluation of important features such as chamber size, motion abnormalities, and valvular disorders. For more quantitative analysis, most commercial ultrasound systems provide specialized software to easily compute the dimensions of various structures within the image. In this way, various quantitative measurements such as LVEF can be standardized and tracked over time54. Doppler flow techniques are also frequently used in the clinic to image blood flow within the chambers of heart, particularly for screening of valve-related disorders55.

Cardiac strain and strain rate imaging techniques have been developed using ultrasound for various clinical applications56–58, although integration of these techniques into routine clinical practice has been slow due to reproducibility concerns4,6. In the heart, strain and strain rate measurements are often preferred to displacement and velocity measurements in order to reduce the influence of rigid motion artifacts caused by natural physiologic processes such as breathing58. These techniques are able to image deformation experienced by the myocardium by tracking characteristics of the ultrasound speckle pattern throughout the cardiac cycle. Time- domain techniques such as normalized cross-correlation and optical flow can be applied to the ultrasound data to estimate the deformation of structures within the image over time59. Further details of the normalized cross-correlation algorithm will be explored in Chapter 4. Tissue deformation can also be imaged using frequency-domain processing such as in tissue Doppler60; however, these techniques can be limited by angle dependency issues61.

Other developing ultrasound-based methods for cardiac mechanical evaluation include acoustic radiation force impulse (ARFI) imaging62 and shear wave elastography (SWE)63. These techniques use an acoustic beam to “push” the tissue and measure the resulting deformations.

Since the acoustic force of the push beam is known, these techniques allow for direct computation of the elastic modulus of various regions of the myocardium. However, the use of acoustic push beams introduces other challenges such as limitations in penetration depth and more complex safety considerations.

Other imaging modalities are used far less frequently in clinical cardiology. Nuclear imaging techniques including positron emission tomography (PET) and single-photon emission computed tomography (SPECT) are becoming increasingly used in the clinic for perfusion imaging to detect and localize myocardial ischemia in patients with coronary artery disease (CAD)64,65.

Similarly, cardiac computed tomography (CT) and magnetic resonance imaging (MRI) have been used with contrast angiography to identify occlusions located within the coronary vessels66,67. In recent years, hybrid technologies combining nuclear imaging with CT or MRI have been developed in order to overlay perfusion information with high-resolution images of tissue structure68,69. However, many of these techniques are expensive and/or involve the exposure of patients to ionizing radiation. In light of this, joint guidelines established by the

AHA and the American College of Cardiology Fellows (ACCF) have noted that these imaging technologies are not recommended for repeat testing or general screening70,71. Thus, longitudinal evaluation of patients using these imaging technologies may not be feasible. Other drawbacks of nuclear, CT, and MRI technologies include limited temporal resolutions64,72, long procedure times, and use of injected tracers/contrast agents.

3.2 Myocardial Elastography

3.2.1 Introduction

Myocardial Elastography is an ultrasound-based strain estimation technique developed for cardiac applications73. The concept of strain is highly relevant to the function of the heart since muscle contraction, i.e. shortening, by definition involves a change in strain. Therefore, strain that occurs within the cardiac tissue is likely to be linked to its structure and function. A large number of studies have been published on the applications of strain estimation in cardiology, which include ischemia/infarct localization, dyssynchrony assessment, and quantification of LV torsion56,58,74,75.

The idea behind ME was developed as an extension of the quasi-static elastography technique developed for, among others, applications in breast lesion imaging76. In the breast, an area of reduced strain following compression may be indicative of cyst or tumor formation. In the heart however, an area of reduced strain during contraction may indicate a region of decreased contractility due to ischemia or as a result of radiofrequency ablation. The identification and localization of abnormalities in myocardial contractility is of high clinical interest for several medical specialties, including cardiology, emergency medicine, and surgery. ME has previously been validated in simulations77,78, animals models79, and in human subjects80,81. Recent studies have been more clinically-oriented and have demonstrated the ability of ME for detecting ischemia following coronary artery occlusion82 and for lesion monitoring during RF ablation procedures83,84.

3.2.2 Methods

In this dissertation, two different ultrasound hardware systems are used to implement the ME technique. Both hardware systems are able to obtain the raw signals received by each element of their respective transducers, so the data output of each system is effectively equivalent from a post-processing standpoint. The Verasonics Vantage (Redmond, Washington) system is designed for use with a 2.5 MHz phased-array transthoracic probe, while the St. Jude Medical ViewMate

Z (St. Paul, Minnesota) is designed for use with a 6.0 MHz phased-array intracardiac echocardiography (ICE) probe. A general overview of the methods involved in ME is displayed in Figure 3-1. Subjects are scanned using a phased-array probe attached to either acquisition system. A custom, high frame-rate diverging beam sequence (DBS) is programmed into the system and used to obtain element RF data for 2 seconds, which is a sufficient amount of time to capture the entirety of a single cardiac cycle (Figure 3-1a). The DBS refers to a transmission that is capable of imaging the entire field of view at a high frame-rate using a single transmit and is discussed further in Chapter 4. The ultrasound system also triggers an external module to acquire

ECG data in parallel with the ultrasound data. Following the DBS acquisition, a standard acquisition is performed at 30 Hz using focused beams. This sequence provides a high resolution image of the heart and is used for segmentation and display purposes only. Strain estimation is performed entirely using the data collected from the high frame-rate DBS.

Each of the two systems enables the capture of “element” RF data; that is, the raw RF signals received by each individual element within the transducer. In order to generate readable images from the element data, beamforming techniques must be used along with a priori knowledge of

the acquisition sequence (Figure 3-1b). A standard beamforming strategy is the delay-and-sum algorithm which is one of the more computationally-efficient methods85. The delay-and-sum algorithm computes the amount of time it would take for the acoustic wave to reflect back to each transducer element from each point within the image. The contributions from all transducer elements are summed together at each point in space to generate intensity values throughout the image.

1-D or 2-D displacements are estimated between consecutive RF frames using normalized cross- correlation algorithms (Figure 3-1c)86,87. For the 1-D estimation, axial displacements are estimated within each line using a 1-D kernel with a high degree of overlap of (>80%) between consecutive estimation windows, allowing for highly-sampled strain estimations in the both directions. The size of the correlation kernel may vary based on the application. In general, a larger window size leads to a smoother distribution of displacement estimation throughout the image. However, a correlation kernel that is too large will be affected by border artifacts and may not produce the displacement gradient necessary for the strain computation. A good starting point for selecting the correlation kernel is a 10-wavelength window size, which has been shown to offer a good tradeoff between resolution and noise suppression for strain estimation using cross-correlation88. For 2-D estimation, the 1-D correlation kernel is used within a 2-D search window. In order to compensate for the limited RF resolution in the lateral direction, linear interpolation is used to upsample the RF data in the lateral direction87. In both the 1-D and 2-D estimations, cosine interpolation is used to increase the available resolution of the displacement estimation by allowing for the estimation of sub-sample displacements89.

The initial displacements estimated by 1-D or 2-D cross-correlation refer the motion between two frames only and are termed inter-frame displacements. In Myocardial Elastography, inter- frame displacements must be accumulated throughout systole in order to capture a mechanical characterization of the entirety of systole. As such, the estimated inter-frame displacements are tracked through time and accumulated over the systolic phase (Figure 3-1d). To reduce processing time, the ECG is used to truncate the displacement data to a time window consisting only of systole. The systolic phase of the heart begins with end-diastole, identified by the peak of the R-wave in the ECG, and ends with end-systole, identified by the peak of the T wave in the

ECG.

After the displacements have been tracked and accumulated over systole, a least-squares strain estimator is used to compute the strain throughout every systolic frame, resulting in an image stack containing cumulative systolic strains (Figure 3-1e). The magnitude and distribution of the strains occurring at end-systole are then analyzed in the context of the clinical problem of interest.

3.3 Electrical Evaluation of the Heart

The electrocardiogram (ECG) is one of the most important tools available to a physician due to its ease-of-use and familiarity throughout clinical practice. The 12-lead ECG uses 12 surface electrodes to capture the global summation of all action potentials produced by cardiac cells at high temporal resolution. Since the ECG has a long history of use in clinical practice, it is currently the most commonly-used diagnostic procedure for cardiovascular conditions and is capable of diagnosing more than 100 different cardiac pathologies15,16. Although the ECG is

Figure 3-2: Typical ECG tracings for (A) normal sinus rhythm, (B) atrial fibrillation, and (C) left bundle branch block. fundamentally designed to detect electrical abnormalities such as atrial fibrillation and left bundle branch block, it can also be used to suggest certain mechanical abnormalities such as ischemia and left ventricular hypertrophy (Figure 3-2).

Despite its prevalent use in the clinic, the ECG does have its shortcomings. Most notably, since the ECG ultimately measures a global electrical signal, the spatial resolution of the ECG is fairly poor. Although signals obtained by different leads can be generally correlated with different regions of the heart16, the ECG remains unable to provide a comprehensive spatial mapping of cardiac electrical activity. Localization of cardiac electrical activity may assist in diagnosis and treatment planning. Currently, clinicians must resort to EA mapping techniques in order to image the cardiac electrical activity27. These methods involve a catheter placed inside the heart to record the electrical signals at various points in the myocardium, allowing for the generation of the spatial map of electrical conduction. Contact mapping using various commercial systems is routinely done during clinical interventions, although this process requires the use of

fluoroscopic imaging and is fairly inconvenient, since point-by-point mapping of the heart can take up to several hours to complete. Non-contact mapping methods have also been developed90, but their use in the clinic is yet to be established.

3.4 Electromechanical Wave Imaging

3.4.1 Introduction

Electromechanical Wave Imaging (EWI) was conceived after the initial development of ME with the goal of imaging cardiac arrhythmia. Although the same basic methods of ultrasonic strain estimation apply to both EWI and ME, the clinical nature of each application leads to slightly different acquisition and processing methods. First, it is important to understand that EWI attempts to map the cardiac electromechanical function—the transient deformations that occur in response to electrical activation—as opposed to the mechanical function studied by ME.

Electromechanical activation, which is discussed in detail in Chapter 2.3.3, is closely linked to electrical activation91, providing a useful platform to study cardiac arrhythmia. Ultimately, the

EWI technique is able to produce an electromechanical activation map, which depicts the timings of electromechanical activation throughout the cardiac tissue. EWI electromechanical activation maps have recently demonstrated high correlation with electrical activation measured with EA mapping in all four chambers of the heart92,93. Thus, EWI carries immense clinical potential due to its ability to map arrhythmia using a non-invasive, non-ionizing, and real-time imaging technology. Current clinical techniques to assess arrhythmia either do not provide spatial mapping (ECG) or require time-consuming procedures (EA mapping).

In addition to validation by EA mapping, EWI has also been validated using a 3-D electromechanical model94 and implanted electrodes95. Applications of EWI have extended to both the atria and ventricles of the heart. In the ventricles, EWI has previously demonstrated feasibility in mapping pacing origins94–97, regions of ischemia98, and left bundle branch block97.

In the atria, EWI has shown feasibility in detecting atrial fibrillation cycle length99 and in differentiating between three common atrial arrhythmias (fibrillation, flutter, and focal tachycardia)100.

The major difference between EWI and ME is the use of inter-frame strains in EWI compared to the use of cumulative strains in ME. While changes in cardiac contractility are often revealed over the entire length of systole, electromechanical activation may occur very rapidly, within the first 50 ms of systole. Thus, high frame-rate imaging is even more critical for EWI compared to

ME. The effects of high frame-rate imaging on strain estimation are discussed further in Chapter

4. The other major difference between EWI and ME is the use of 1-D strain estimation for EWI compared to 2-D estimation used by ME. Since EWI is more concerned with the timing of activation rather than the strain magnitude, axial strain estimation alone is thought to provide sufficient information to capture this phenomenon. The angle independence of 1-D EWI has been confirmed previously, with similar activation maps being generated regardless of the imaging orientation101. 1-D axial strain estimation lends itself well to the use of the apical imaging views, which allow visualization of all four chambers. In apical views, the walls of the heart are approximately aligned with the axial direction of the image, allowing for a consistent and simultaneous calculation of longitudinal strain in all four chambers of the heart.

3.4.2 Methods

Figure 3-3: Brief overview of the methods for Electromechanical Wave Imaging. (A) A diverging beam sequence to capture element data at high frame-rates. (B) 1-D inter-frame displacement estimation using normalized cross-correlation. (C) Inter-frame strain estimation using a least squares strain estimator. (D) Selection of the temporal zero- crossings corresponding to electromechanical activation. (E) Mapping the timings of zero- crossings to the cardiac tissue.

Figure 3-3 describes the basic workflow for the EWI acquisition and processing. A 2.5 MHz phased-array probe attached to a Verasonics system is used to transmit a DBS at 2000 Hz in the apical view of the heart (Figure 3-3a). Similar to ME, the DBS is followed by a conventional focused acquisition for 2 seconds at 30 Hz to be used for display and segmentation purposes

only. The positioning of the apical view requires approximately 16-20 cm of imaging depth to be able to visualize both the atria and ventricles in a single image. Element RF data is collected by each of the 64 channels of the probe and beamformed offline using a standard delay-and-sum algorithm. The beamformed RF data for EWI consists of 64 lines at an axial sampling rate of 20

MHz. 1-D axial displacements are estimated using the normalized cross-correlation function described in the previous section (window size = 5.85 mm, window overlap = 90%) (Figure 3-

3b). Axial inter-frame strains are then estimated using a least-squares strain estimator.

Segmentation of the cardiac tissue is performed using a B-mode image at end-diastole obtained from the conventional sequence (Figure 3-3c). The progression of the axial inter-frame strains within the myocardium over time is then used to detect electromechanical activation (Figure 3-

3d). Electromechanical activation is defined by the sharp transition in strain that occurs just after the electrical activation shown by the ECG. For atrial mapping, electrical activation is defined by the onset of the P-wave, while ventricular electrical activation is defined by the onset of the QRS waveform. In order to be able to compute a very precise time of electromechanical activation, inter-frame strains are used for EWI instead of the cumulative strains used in ME. Even though the magnitude of the inter-frame strains is extremely small (~0.05%), we will see from studies presented in the next chapter that our strain estimation technique performs well within this range.

Figure 3-3d displays an example of how inter-frame strain is used to compute electromechanical activation. The strain curve displayed in this figure represents the inter-frame strain of a single point within the ventricles and how it changes in time in relation to the surface ECG. A sharp transition from nearly zero to negative strain is noted immediately following the first deflection of the QRS complex. In the apical views of the heart, negative strain indicates the longitudinal

shortening that accompanies ventricular contraction. The zero-crossing of this transition to negative strain is identified as the electromechanical activation time, which can be computed for every point within the view of the image. Previous EWI studies have used manual selection of the “zero-crossings” of inter-frame strain signals to generate an electromechanical activation map97,102. However, methods to automatically identify and select the correct zero-crossing have been developed and are discussed in Chapter 6. The zero-crossing times are then mapped onto the cardiac tissue segmentation obtained from the conventional B-mode image (Figure 3-3e).

Chapter 4

Optimizing Echocardiographic Strain Imaging

4.1 Introduction

The modern approach to ultrasonic strain imaging was introduced by Jonathan Ophir in 1991103 who coined the term “elastography” to refer to the measurement of the elastic modulus in tissues using ultrasound-derived strain measurements. In this paper, Ophir describes how the application of an external compression to a tissue sample will induce a change in tissue strain that can be measured using ultrasound signals. Over the years, more advanced techniques to measure and track the strain-induced signal change have been developed, including methods for 2-D and 3-D strain estimation. In this dissertation, 2-D strain estimation is performed using normalized cross- correlation with a 1-D window in a 2-D search87. Other modern algorithms for 2-D strain estimation include 2-D coarse-to-fine cross-correlation104, 2-D block matching105, and

Lagrangian displacement tracking using a polar grid106. However, each of these techniques are fundamentally based on the fact that ultrasound speckle is not simply random noise, but is characteristic of the underlying tissue and can be tracked over time. The technique of Myocardial

Elastography was in fact developed as an extension of compression-based elastography, relying on the deformations induced by the natural contraction of the heart instead of those induced via external compression73.

4.2 Displacement Estimation

Displacement estimation in the time domain is often computed using correlation-based techniques, which rely on the use of a correlation function to track the motion of RF signals.

Although the type and implementation of the correlation function can differ between techniques, the most commonly-used correlation function is the normalized cross-correlation (NCC) function86. This function computes the degree of similarity between two signals f(n) and g(n) and takes the form

푢+푊−1 ∑푛=푢 푓(푛)푔(푛+휏) 푅푁퐶퐶(푢, 휏) = , (휏1 ≤ 휏 ≤ 휏2), 푢+푊−1 2 푢+푊−1 2 √∑푛=푢 푓 (푛)∗∑푛=푢 푔 (푛+휏) where u is the location of the signal window of interest, W is the window size, and τ is the shift between the two signals. The output of this function, R, is referred to as correlation coefficient and describes the amount of similarity between two signals on a normalized scale (from 0 to 1).

In 1-D motion tracking applications, the two signals are obtained from the same ultrasound line in consecutive frames in time. Since the ultrasound speckle pattern does not change significantly over time, the NCC function is able to identify where different regions of the signal at time t moved to at time t+Δt. This tracking is performed for several small windows within a particular line, providing multiple estimations of displacement per line. Figure 4-1 provides a general depiction of the outcome of this process for two windowed locations within a single line.

The NCC function considers each of the highlighted windows in frame i and is able to find their

“best matches” in frame i+1. The best match is determined by the maximum value of the NCC function obtained when sliding the windows in frame i over a specified search range in frame i+1. The location of this maximum value have be further refined using a cosine interpolation to allow the estimation of sub-sample displacements89. The correlation coefficient also serves as a

Figure 4-1: Depiction of correlation-based displacement estimation at two separate locations within the same line. The relative displacement between the two points will be reflected by a change in strain. benchmark for the reliability of the displacement estimation. In effect, high quality tracking results in correlation coefficients close to 1.0, while poor estimation results in lower correlation coefficients.

In ultrasound applications, the NCC function can be used with either RF signals or B-mode signals to track displacements. Although B-mode signals often appear less noisy upon visual inspection, RF signals offer superior precision in displacement estimation applications involving the NCC function107,108, largely due to the presence of phase information within the RF signal

that is removed during the Hilbert transform to B-mode data. The NCC algorithm is sensitive to fluctuations in both amplitude and phase, both of which are preserved in RF signals.

Since the NCC algorithm uses a sliding window in the comparison frame to compute displacements, the direction of this sliding window can be changed to estimation displacements in different directions. Axial displacement can be estimated by a search in the axial direction along a single line, while lateral displacement can be estimated by a search in the lateral direction between lines. In this way, the same NCC function can be used to estimate displacements in either direction. However, since RF lines are reconstructed in the axial direction independently of neighboring lines, the amount of available information is much greater in the axial direction compared to the lateral direction. Attempts to overcome the limitations in the lateral direction have included interpolation between beams87 and novel transmission/beamforming techniques109,110.

4.3 Strain Estimation

As stated previously, estimations of strain can be derived directly from displacement estimates computed using time-domain correlation functions such as NCC. For 1-D applications, strain can be computed using a simple gradient operator applied in the axial direction to the displacement field111. In computing 2-D strains, the entire strain tensor must be considered and calculations of the axial and lateral strain values involve several more operations. However, the spatial gradient of the displacement in either the axial or lateral direction remains the most significant determinant of the strain in that particular direction.

The earliest holistic investigation into the factors involved in ultrasonic strain estimation quality occurred in 1997112. Like many pursuits in the imaging field, the authors attempted to first use a simulation model to investigate a relationship between the ultrasound acquisition and the quality of axial strain estimation. A theoretical expression for the error of cross-correlation-based strain estimation was developed for various magnitudes of strain. The estimation error for the smallest magnitude strains estimated by ultrasound is governed by the Cramer-Rao Lower Bound

(CRLB)113:

3 1 1 휎2 ≅ [ (1 + )2 − 1] 퐶푅퐿퐵 2 3 2 2 2 2휋 푇(퐵 + 12퐵푓0 ) 휌 푆푁푅푠 where f0 is the transducer center frequency, B is the transducer bandwidth, T is the window size of the strain estimator, ρ is the correlation coefficient, and SNRs is the sonographic signal-to- noise ratio term that encompasses RF data quality. For larger strains, the Barankin Bound (BB) is used, which raises the CRLB by a constant factor:

푓 2 휎2 = 12 ( 0) 휎2 퐵퐵 퐵 퐶푅퐿퐵

The cutoff strain value between the two error functions is governed by a separate equation, but usually occurs between 1-10% strain for cardiac imaging transducers.

The two parameters in the CRLB to be more closely examined in this chapter are correlation coefficient (p) and the signal-to-noise ratio of the RF data (SNRs). These two parameters represent two distinct focus areas for optimization of strain estimation. The correlation coefficient is a function of the displacement estimation, while SNRs is a function of the acquisition used to obtain RF signals.

4.4 Study: Motion Estimation Rate

4.4.1 Introduction

One way to increase the correlation coefficient of the displacement estimation is to reduce the amount of motion that occurs. Motion tracking using cross-correlation is based on the assumption that the signal does not significantly change from frame to frame. However, tissue motion does cause slight changes in the RF signals and corresponding speckle pattern, leading to decorrelation of the signals from frame to frame107,114. High strains causing large amounts of signal decorrelation lead to increased errors in motion estimation and consequently strain estimation115. Further support for the desirability of small strains comes from the previous discussion of the CRLB and BB; the BB, which is used to define the error for the estimation of large strains, exceeds the magnitude of the CRLB for estimation of small strains.

Decreasing the time between frames used for motion estimation is one way to reduce the amount of motion that exists from frame to frame, allowing for more precise displacement and strain estimation116–118. However, a fundamental tradeoff exists in ultrasound physics between depth, beam density, and the acquisition frame-rate. This tradeoff is described by the classic ultrasound range equation, which states that acquisition framerate is proportional to the speed of sound (c) and inversely proportional to imaging depth (d) and the number of lines (N):

푐 퐹푅 = 푚푎푥 2푑푁

Cardiac imaging requires a minimum depth (typically between 12-20 cm) to capture the entire heart structure. In most cases, the imaging depth is set as low as possible to increase acquisition frame-rate. Thus, increasing frame-rate comes at the cost of decreasing the number of lines for a given depth; i.e., increasing the temporal resolution decreases the spatial resolution. Indeed, early

attempts to increase frame-rate used smaller fields of view73,119 or sparse sector scans120. Not surprisingly, increasing frame-rate while maintaining a high spatial resolution is a challenging yet relevant task for many cardiac strain estimation techniques. ECG gating118 and motion matching98 techniques have exploited the periodic rhythm of the heart to combine smaller field- of-view acquisitions at higher frame-rates. However, these techniques result in unfavorable stitching artifacts and require imaging over multiple heart cycles, which precludes the study of aperiodic arrhythmias.

Several techniques have been developed in our lab to enable single-heartbeat strain estimation at both high frame rates and high spatial resolution97,121. These techniques rely on specialized transmission sequences that differ significantly from the standard acquisitions used in clinical practice today. The first technique, referred to as temporally unequispaced acquisition sequence

(TUAS), alters the order by which beams are emitted from the transducer to create an effective increase in frame-rate throughout the entire view. The second technique, referred to as diverging beam sequence (DBS), uses a novel, diverging beam interrogate a large amount of space using only a single transmission. Further details of these techniques are displayed in Figure 4-2.

The first major difference between the two techniques can be seen in the differeing beam profiles

(Figure 4-2a). While TUAS uses a series of conventionally focused beams, DBS uses a single diverging beam to image the entire view at once. Conventional acquisitions (Figure 4-2b) construct an image by transmitting multiple focused beams across the entire view one after

Figure 4-2: Overview of the sequences used for the MER study. (A) Acoustic beam profile for the TUAS (left) and DBS (right). Temporal profiles for (B) the conventional sequence, (C) TUAS, and (D) DBS. Each line in the B-D corresponds to a single transmission of the corresponding beam shown in A.

another. TUAS also uses focused transmit beams, but changes the order that these beams are fired in order to increase the effective frame-rate used for motion estimation, i.e. the motion estimation rate (MER). Instead of imaging an entire frame all at once, TUAS takes two acquisitions of a small sector of the view very rapidly. By increasing MER, TUAS causes a corresponding drop in the temporal sampling rate of motion estimation, i.e. the motion sampling rate (MSR). The difference between MER and MSR may be confusing at first, but is futher explained in Figure 4-2. In short, the MER refers to the time between frames where motion estimation is performed, while MSR refers to the time between motion estimations of the same point. In conventional sequences where MER and MSR are the same, each frame is used for two displacement calculations—both the previous frame and next frame in the sequence. In TUAS, each frame is paired with only one other frame to perform motion esimation at the MER, while these pairs of frames are separated from each other at the MSR.

The objective of the MER study will be to examine the effects of MER on inter-frame axial and lateral strain estimation in human subjects (n=5). Since both TUAS and DBS will be used to achieve high MER imaging in all subjects, we will also compare the inter-frame strain distributions estimated by each sequence.

4.4.2 Methods

Using a protocol approved by the Institutional Review Board of Columbia University, five, healthy male volunteers aged 23-30 were scanned in the lateral decubitus position by a trained cardiac sonographer. All patients were scanned with a 2.5 MHz phased-array probe connected to a Verasonics V-1 system (Redmond, WA) using both TUAS and DBS. For each sequence,

acquisition lasted for 2 seconds to ensure the capture of an entire cardiac cycle. The same echocardiographic view was taken in all patients; namely, the parasternal short-axis view at the papillary muscle level. Because the study objective was to analyze and compare both axial and lateral strain estimations, the short-axis view was chosen so that the strain magnitudes would be similar in each direction.

Three TUAS acquisitions of different sector sizes were acquired per patient, which allowed for the construction of RF data at four effective MERs (65, 272, 544, 897 Hz). A single DBS acquisition at 2000 Hz was also obtained in each patient. Lower MERs of 250, 500, and 1000 Hz were achieved by skipping frames during motion estimation. All DBS strain estimations were performed at a MSR of 65 Hz to match the temporal sampling frequency of the TUAS estimations.

For each sequence, element data received by the transducer was sampled at 10 MHz for each of the 64 individual traducer elements. Delay-and-sum beamforming was used to reconstruct 120 lines of RF data per frame at an axial sampling rate of 20 MHz. For the TUAS, this corresponds to the construction of 10 RF lines per transmit beam; while in the DBS, all 120 lines were constructed from a single transmit. A 90° field of view at a depth of 20 cm was used to mimic standard clinical settings. ECG measurements were triggered by the ultrasound system and acquired in parallel with the ultrasound measurements.

Strain estimation was performed as described by the 2-D ME processing methods described in

Chapter 3.2. However, only inter-frame strains were considered in this study, such that no

motion tracking or accumulation was performed. The window size for the correlation kernel was selected at 5.85 mm (approximately 10 wavelengths) and the window overlap was 90%.The size of the strain kernel was selected to be 10.3 x 10.5 mm. 2-D spatial filtering of the displacements and strains was performed only for the strain images presented in the first section of the results.

In order to investigate the utility of these two sequences and the effect of the corresponding increase in MER, a metric must be developed to evaluate strain estimation quality. Implantable sonomicrometry beads, which can measure the change in strain over time between two markers, have been employed previously for this purpose122, although use of sonomicrometry is cumbersome and clearly not suitable for human subjects. Furthermore, sonomicrometry can only compute strain values at discrete points within the heart, while most ultrasonic strain imaging techniques will estimate strain throughout the entirety of the tissue. Alternatively, the elastographic signal-to-noise ratio (SNRe) may be helpful to evaluate strain estimation quality, especially within a continuous field of strain estimates. SNRe is computed as the ratio between the mean and standard deviation of strain existing within a small (3x3 mm) region of interest:

µ(휀) 푆푁푅 = . 푒 휎(휀)

This region of interest should be sufficiently small so that the expectation of strain uniformity within the region is reasonable. SNRe has been used for years to evaluate strain estimation and is an excellent metric to assess estimation precision112, although it is less related to the accuracy of the measurements. Although a single, fixed window may be used to study SNRe, comparisons between different sequences may be inhibited by slightly the slightly different views that are acquired with each sequence. Thus, the use of multiple MER windows located within the cardiac tissue for every frame should facilitate a more fair comparison121.

Since both strain and SNRe were computed for each point in the myocardium throughout systole, a large number (>600,000) of strain-SNRe pairs were generated for each sequence. This paired data can be used to generate a two-dimensional histogram (f(SNRe,ε)) that displays the probability density function (pdf) of SNRe for a set of discrete strain values. However, since the strain distribution measured in the heart is non-uniform, the 2-D pdf will be biased towards strain values that occur more frequently. To remove this bias, the conditional pdf was calculated according to

푓(푆푁푅 , 휀) 푓(푆푁푅 |휀) = 푒 , 푒 푓(휀) where f(SNRe, ε) is the 2-D pdf and f(ε) is the 1-D strain distribution. It is important to keep in mind that f(SNRe| ε), f(SNRe, ε), and f(ε) will be affected by the quality of strain estimation and thus be dependent on the MER. Figure 4-3 shows an example plot of the conditional pdf of axial strains taken with TUAS and DBS at 544 Hz and 500 Hz respectively. To simplify the information obtained from the 2-D conditional pdf, which manifests as the 3-D plot in Figure 4-

3, the conditional expected value of SNRe for each strain was calculated as

+∞

퐸(푆푁푅푒|휀) = ∫ 푆푁푅푒푓(푆푁푅푒|휀)푑푆푁푅푒 . 0

E(SNRe| ε) curves are generated for each sequence, which allows for a relatively easy comparison to be performed between different sequences for a wide range of strain values. An example E(SNRe| ε) curve can also be seen in Figure 4-3.

However, since incremental strains are being used, it is important to keep in mind that the same deformation will be measured differently by sequences at different MERs. For instance, a 1%

strain measured at 1000 Hz would be measured as a 2% strain at 500 Hz. Therefore, in order to compare strain magnitudes between sequences, strains have been multiplied by MER for normalization. Final magnitudes of normalized strain are presented in decibels, with the reference of 0 dB being a 1% strain measured at 100 Hz:

푀퐸푅 ∗ 휀 푛표푟푚푎푙푖푧푒푑 휀 = 20 log ( ). 100 퐻푧 ∗ 1%

4.4.3 Results

Figure 4-3 displays incremental strain images of the left ventricle (LV) during contraction taken with the TUAS and DBS at 544 Hz and 500 Hz respectively. Images were synchronized in time using ECG recordings, which are displayed below the strain images. The exact point in the cardiac cycle from which each image was taken is denoted by the red dot on the ECG tracing.

Figure 4-4: Comparison of axial and lateral inter-frame strain estimations for TUAS and DBS during LV contraction in the same subject. ECG tracing is shown below the images indicating the timing within the cardiac cycle. Axial and lateral strains appear similar at similar stages of contraction.

The first image in each series was taken towards the end of diastole, directly before the P-wave.

Both sequences show relatively low strains all directions at this time. The following two images in each series were taken during systole following the initiation of ventricular contraction.

Positive axial strains can be seen in the anterior and posterior walls of the LV, with negative axial strain occupying the septal and lateral walls. In the lateral direction, the distributions of positive and negative strains are reversed, with negative lateral strains occurring in the anterior and posterior walls and positive lateral strains occurring in the septal and lateral walls. Radial and circumferential strains appeared more uniform, especially in the late systolic frame. Positive radial strain and negative circumferential strain are clearly occurring in much of the LV wall during late systole. In the early systolic frame, the radial and circumferential strain distributions are less uniform, especially in the septum. In all cases, the magnitudes of the strains increase from early systole to late systole, which are the second and third images in each series respectively.

E(SNRe|) curves are shown in Figure 4-5 for both sequences at various MERs. In the axial case, the curves exhibit a two-peaked, bimodal distribution. The locations of both peaks shift towards larger values of strains for higher MERs. This shift becomes less apparent at MERs >544 Hz, in which case the locations of the first peak are clustered at approximately 10 dB. This phenomenon was observed for both the TUAS and DBS, although the shifting behavior is more apparent using

TUAS since the MERs tested are lower. The amplitudes of the first peaks also increase with

MER while the amplitudes of the second peaks decrease, causing the curve to approach a single- peaked distribution at higher MERs. The maximum value of E(SNRe|) for any sequence is

Figure 4-5: Expected SNRe curves for axial and lateral strain estimation using the TUAS and DBS at various MERs in 5 healthy subjects.

5.71, occurring for the DBS at an MER of 2000 Hz and a strain of 8.91 dB. This magnitude corresponds to an incremental strain magnitude of 0.14% measured at 2000 Hz. Figure 4-6 plots curves for both sequences at similar MERs—272 and 544 Hz for TUAS, along with 250 and 500

Hz for DBS. Both sequences produce E(SNRe|) curves with approximately the same shape and magnitude at similar MERs.

Figure 4-6: Expected SNRe curves for axial (left) and radial (right) strain estimation using the TUAS and DBS at similar MERs.

In contrast to the bimodal distributions observed for the axial direction at low MERs, all

E(SNRe|) curves appear unimodal for the lateral direction regardless of MER. Also, the rightward shift with increasing MER observed in the axial case is not as clear in the lateral direction. At higher MERs, the locations of the peaks do not change significantly and are clustered around 40 dB. Unlike in the axial direction, the magnitudes of the peaks of the curves do not seem to significantly change with MER; the maximum magnitude of each curve is approximately 4.5. However, the 65 Hz TUAS curve is shifted by approximately 15 dBs to the left of the other curves. Again, at similar MERs, the TUAS and DBS curves appear similar in magnitude and shape.

4.4.4 Discussion

The precision of cardiac strain estimation using RF cross-correlation is often reduced due to RF signal decorrelation caused by the large incremental strains that occur during contraction. 50

Increasing the MER is a well-accepted method of reducing this decorrelation, thus increasing precision. However, it is difficult to achieve high MER sequences while maintaining a high spatial resolution. It is even more difficult to achieve both of these goals by imaging single cardiac cycle. The two sequences presented here can be used within a single heart cycle at higher

MERs without compromising spatial resolution, and have been applied to 2-D strain imaging.

The series of strain images in Figure 4-4 indicate that both sequences can reliably be used to estimate two-dimensional cardiac strains in vivo. The strain distributions imaged by each sequence agree with the expectations derived from knowledge of cardiac mechanics74,118. At the onset of the P-wave, the atria are just beginning to contract while the ventricles are at rest. The strains imaged at this time were very low as expected. During contraction, the myocardium of the left ventricle begins to thicken in the radial direction and shorten in the circumferential direction. Radial thickening would manifest as positive axial strain in the anterior and posterior walls, along with negative axial strain in the septal and lateral walls. The lateral strain distribution would be reversed since the axial and lateral directions are relatively orthogonal at each point. The expected behavior is observed in all directions using either sequence.

Furthermore, the strains imaged by the TUAS and DBS at the same point in time seem to be fairly consistent in terms of both magnitude and distribution. The worst agreement between sequences occurs during early systole when the myocardium is just beginning to contract, creating incremental strains that are changing very rapidly in both space and time. Here, it is very difficult to observe the exact same strain image in both sequences, especially considering the limited MSR. Towards the end of systole when the contractile behavior is more uniform, the strain distributions imaged by both sequences become more similar. There is worse agreement

for radial and circumferential strains, which is expected. Since radial and circumferential strains are generated by combining both axial and lateral strains, errors from each estimation direction are combined and propagated. Furthermore, unlike axial and lateral strains, radial and circumferential strains are largely influenced by the myocardial segmentation, since the centroid for the coordinate transformation is defined by the manually drawn mask.

A previous study on TUAS has noted the two-peaked nature of the axial SNRe distribution, noting the switch to a single-peaked distribution above an MER of 452 Hz121. This particular shape is explained to be a result of the transition zone that occurs in the Strain Filter presented by

Varghese and Ophir112. The transition zone occurs at high strains when decorrelation becomes too significant and motion is estimated on the basis of RF signal amplitude only, i.e. the signal envelope, instead of using both the amplitude and the phase information. Smaller strains below the transition zone are estimated using both amplitude and phase information, with estimation quality being governed by a modified version of the Cramer-Rao lower bound (CRLB) for time- delay estimators113. For larger strains above the transition zone, phase information is lost due to significant signal decorrelation, and the estimation quality is instead governed by the Barankin bound (BB). For strains near the transition zone, the method of motion estimation is ambiguous and inconsistent, causing a local minimum of SNRe121. In this work using E(SNRe|ε), similar, two-peaked results are reported, with the disappearance of the second peak above an MER of

544 (Figure 4-5). The disappearance of the second peak is due to the fact that smaller incremental strains will be measured at high MERs. If MER is sufficiently high, the measured strain values will be small enough to fall below the transition zone, leading to a disappearance of

the second peak. Figure 4-3 shows the approximate location of the transition zone, along with the locations of the CRLB and BB within the strain magnitude spectrum.

Motion estimation using the RF signal envelope has consistently been shown to be less precise compared to techniques which use both the RF signal phase and amplitude108,123. Thus, any axial motion estimation technique should aim to minimize the amount of strains estimated using only the envelope. The work presented here indicates that sequences with a single-peaked axial distribution of E(SNRe|) should be preferred over a sequence with a two-peaked distribution.

The E(SNRe|) distribution of the lateral strain estimation is single-peaked for all MERs (Figure

4-5). A single-peaked distribution is expected for the lateral direction due to the difference method of motion estimation compared to the axial direction. Although axial phase information is used to increase the accuracy of the lateral motion estimation, there is no lateral phase information inherent in a collection of 1-D RF lines. Therefore, the lateral cross-correlation function is only sensitive to variations in the lateral envelope. It is somewhat surprising that the amplitude and shape of the lateral curves do not change significantly with MER, although the width of the curve at 65 Hz appears significantly smaller than the other curves. However, it has been previously reported that lateral estimation quality is much less influenced by MER than axial estimation124. Others have reported that the quality of lateral estimation is largely dependent on number of beams, element size, and element spacing125. The locations of the peaks of the lateral curves occur at higher strain values compared to the axial curves. An explanation for this phenomenon is provided later in this section.

There is also a discernible rightward shift of E(SNRe|) curve peaks in the axial direction as

MER increases (Figure 4-5). This trend is much more noticeable at the lower MERs of TUAS

(MER ≤ 544 Hz) and is fairly insignificant at the higher MERs (MER > 544 Hz) of both sequences. To explain this shift, recall that strain magnitudes are normalized by the multiplication of MER. For the 815 Hz sequence, a maximum occurs at an axial normalized strain of 10 dB, which corresponds to a strain of 0.4% estimated at 815 Hz. This strain magnitude reflects the optimal deformation amount for a motion estimation performed at 815 Hz.

However, for the same value of normalized strain, estimation with a 65 Hz MER will experience significantly more decorrelation compared to an MER of 815 Hz, since the estimation interval is longer. Therefore, the peaks corresponding to optimal motion estimation for the 65 Hz sequence must exist at a lower normalized strain value, where there will be less decorrelation. The same explanation can be applied to the lateral, radial, and circumferential E(SNRe|) curves as well, although this shift is much less apparent. It is clear that the 65 Hz TUAS curve is shifted leftward by approximately 25 dB for axial/lateral E(SNRe|) and 17 dB for radial/circumferential E(SNR- e|) compared to all other curves due to the reasons provided above. It is unclear why this shift does not persist at the higher MERs in the lateral direction.

Considering the explanation of the axial curve shift above, the absence of a significant shift in the lateral direction provides an important insight into the reliability of the 2-D strain estimator used. The similarity of the lateral E(SNRe|) curves indicates that the lateral motion estimator used in this study is only tailored to estimate precisely over a small range of normalized strain.

This range of normalized strain occurs at much higher values compared to axial strain due to the difficulty of estimating small strains using the lateral estimator. Even though the lateral RF data

is interpolated by a factor of 10 to increase resolution, the lateral direction still lacks the phase information that allows for the precise axial estimation of small strains around 10 dB. Also, since decorrelation seems to affect lateral E(SNRe|) equally at 250 Hz and 2000 Hz, it is clear that increasing MER is not the best solution to increase the precision of the lateral strain estimation.

The high MERs used in this study were achieved by the use of two different sequences, the

TUAS and DBS. The differing nature of the beam used by each sequence was thought to potentially have an impact on the quality of strain estimation, especially in the lateral direction.

Indeed, previous studies using simulations and ex vivo phantoms have explicitly shown the effect of the beam width, among other parameters, on the quality of lateral motion estimation125–127. A comparison between the two sequences can be performed at similar MERs, since the parameters of depth, field of view, and beam density are held constant in the data processing phase (Figure

4-6). In comparing the TUAS at 272 and 544 Hz with the DBS at 250 and 500 Hz, it was clear that each sequence produced fairly similar curves for all strain directions. The similarities in the curves extend to both magnitude and shape. This observation indicates that, in terms of our evaluation metric, the 2-D strains estimated by each sequence are equivalently precise, assuming that imaging and strain estimation parameters are held constant.

To better qualify this result, it is important to remember that the maximum pressure transmitted by each sequence is similar. Because DBS uses a diverging beam, the voltage supplied to the transducer was able to be increased above that of the TUAS in order to reach the desired ISPTA value. Previous investigations into the effect of beam width on strain estimation did not normalize the pressure intensity between sequences of different beam widths. Furthermore, these

previous studies compared the effect of widening the transmit beams in a conventional sequence; there are few studies which compare a conventionally-focused sequence consisting of multiple transmits per frame to a diverging sequence containing a single transmit per frame128. The

TUAS and DBS are characterized both by different beam widths and different transmit densities—the TUAS used 12, narrow transmits per frame while the DBS used a single, wide transmit per frame (Figure 4-2). Previous work has shown that lateral SNRe seems to be influenced both by beam width and beam density; optimizing the combination of these parameters using the advanced hardware available in newer systems is thus an interesting area for future study. It is also noteworthy that synthetic focusing was used for beamforming and allowed the creation of the same number of RF lines per frame for both TUAS and DBS, which may have contributed to their apparent similarity.

It is difficult to assess the quality of an in vivo data set because there is typically no ground truth measurement available. E(SNRe|) was shown to be able to differentiate sequences based on

MER and may be a useful tool to stochastically evaluate the precision of an in vivo data set.

However, the use of SNRe does have some limitations, especially for axial and lateral strains which have alternating positive and negative values in the short axis view of the heart. A small kernel size (3.0 x 3.2 mm) was used for the SNRe computation to limit the issues caused by this inhomogeneity. Although some SNRe samples may still have been affected by the alternating axial and lateral strain distributions, a large number of SNRe samples (>600,000) were obtained for each sequence so that the overall behavior of E(SNRe|) would not be overly influenced by the border zones between the positive and negative strains. Also, it is important to emphasize that although SNRe is a good measure of precision, it does not reflect the accuracy of these

strains. Feasibility of 2-D strain estimation using TUAS and DBS was observed qualitatively by inspecting the strain images (Figure 4-4), which provided reasonable results. The accuracy of the

2-D strain estimation performed here has been previously validated in simulations77 and using

MR tagging81.

4.5 Study: Spatial Compounding

4.5.1 Introduction

Attempts to increase MER are motivated by the decrease in decorrelation that accompanies a high temporal sampling rate. From the perspective of the Strain Filter, MER and decorrelation are mostly covered by the correlation coefficient (ρ). As stated previously, the sonographic signal-to-noise ratio (SNRs) term within the Strain Filter equation suggests that RF data quality is also an important area to consider for optimization. This area is of particular interest for high

MER applications that utilize diverging beam sequences, since the diffuse transmit profiles have been shown to exhibit lower spatial resolution and contrast129,130. Thus, techniques to increase

SNRs have experienced a resurgence over the past few years due to the widespread use of diverging transmits to increase MER.

Spatial compounding is one such technique that is able to increase the quality of the RF data obtained using diverging transmits while maintaining relatively high MERs131,132. In short, this technique involves the use of multiple diverging beams per frame to form an image. For a single diverging beam transmit, a “virtual focus” is placed behind the probe in order to calculate the piezoelectric element delays to create the diverging beam profile (Figure 4-7). Spatial compounding employs multiple diverging beams per frame having virtual sources located at

different positions along the probe (Figure 4-7). A single RF frame is comprised of the summation of the received data from the various virtual source transmits. Since diverging waves from different virtual sources interrogate each tissue location at slightly different angles, the summation of the reflected signals at each point will be better defined117,129,130,133. Spatial compounding has been applied to motion estimation techniques and has been shown to improve

1-D displacement estimation for in vivo shear wave imaging 130,132 and 2-D strain estimation using in vitro vessel phantoms134. However, the effect of spatial compounding on 2-D cardiac strain imaging in vivo is yet to be investigated.

4.5.2 Methods

The performance of the compounding and diverging sequences was investigated in vivo using acquisitions obtained in four healthy human volunteers. All imaging was performed using a

Verasonics Vantage acquisition system with a 2.5 MHz phased-array probe. Informed consent was acquired in each patient prior to the study. Using a custom acquisition sequence described in the next section, each volunteer was imaged by a trained cardiac sonographer in the parasternal short-axis view at mid-level at a depth of 12 cm. This particular view was chosen since the physiologic strains in the axial and lateral directions occur at approximately the same magnitude over systole. The acquisition also included a trigger to simultaneously acquire electrocardiogram

(ECG) signals for later identification of the systolic phase.

In order to closely compare the strain estimation of each imaging technique, a custom acquisition sequence was designed to perform compounding and diverging imaging simultaneously (Figure

4-7). The custom sequence was composed of a repeated set of 10 diverging wave transmissions;

Figure 4-7: Description of the hybrid diverging beam acquisition sequence used for the compounding study. First, a series of 5 compounding waves are emitted using a reduced sub-aperture with equally spaced source locations along the transducer. Immediately following, a series of 5 diverging waves is emitted using the full aperture and the same source for each transmit.

the first 5 composing the compounding sequence and the final 5 being used for the diverging sequence. Although both the compounding and diverging transmits utilized virtual foci placed behind the probe, the exact locations of the virtual foci and the element apodization differed between each transmit.

For the compounding acquisition, the 5 consecutive transmits corresponded to the 5 virtual foci placed along the x-axis of the transducer, evenly spaced from -3.3 mm to 3.3 mm. The x - locations of each foci were aligned to the center of a 21 element sub-aperture, which was moved along with the foci between transmits. This aperture size was picked so that each of the 5 foci would occupy the center of the aperture all the way along the array. The location of the foci in

the z direction was set at 3.3 mm ensuring an angular aperture of 90°. The number of compounding emissions (5) was chosen to ensure a good tradeoff between a high number of transmits and framerate. The pulse repetition frequency (PRF) was set at 5000 Hz, corresponding to an effective framerate of 500 Hz when considering the 10 total transmits used for the full sequence.

The diverging acquisition consisted of a 5 diverging wave transmits acquired immediately following the 5 compounding transmits. For the diverging transmit, the full 64-element aperture of the probe was used, with the virtual focus placed 10.1 mm behind the center of the probe (Fig.

1a). The PRF between the compounding and diverging acquisitions was also set at 5000 Hz to allow for an effective framerate of 500 Hz for both compounding and diverging imaging.

The total length of the hybrid acquisition was 2 seconds to ensure the uninterrupted capture of the systolic phase of cardiac contraction. ECG signals acquired in parallel with the ultrasound acquisition were used in post-processing to select the frames corresponding to systole. The peak of the QRS complex on the ECG was used to define end-diastole. A displacement M-mode was also used to remove the strains corresponding to the very beginning of systole when rapid variations in strain are occurring. The peak of the T-wave was used to define end-systole.

Radiofrequency signals for each of the 64 transducer elements were recorded throughout the entire acquisition sequence at a sampling rate of 10 MHz. Delay-and-sum beamforming was performed off-line to reconstruct 180 RF lines per acquisition at an axial sampling rate of 10

MHz. Linear interpolation was used to increase the resolution of the RF data in the axial

direction to 10 MHz. The 5 transmits for each transmission type were then combined coherently to form single RF frames, resulting in the same total yield in number of frames for the compounding and the diverging imaging.

Strain estimation was performed as described by the 2-D ME procedure described in Chapter 3.2.

Two-dimensional displacements were estimated between frames using a normalized cross- correlation algorithm having a 1-D correlation kernel (size = 6.1 mm, overlap = 90%) in a 2-D search (axial search range = 0.77 mm; lateral search range = 1 beam)86,87. Spatiotemporal filtering was not performed on either the displacements or strains during processing. Inter-frame

th displacements were filtered in time using a low-pass filter (4 -order Butterworth filter, fc = 500

Hz) and in space using a 2-D median filter (4.55 x 4.0 mm). The 2-D Lagrangian strain tensor was then applied to the cumulative displacements to compute 2-D cumulative strains throughout systole. The 2-D cumulative strains were also filtered in space using a 2-D averaging filter (6.5 x

5.7 mm). Axial and lateral strains were then converted to radial strains by assuming the center of mass of the LV cavity to be the centroid for coordinate transformation.

The compounding and diverging sequence were compared using the elastographic signal-to- noise ratio (SNRe) of the radial strain.

휇 푆푁푅푒 = 휀,푟푎푑푖푎푙 휎휖,푟푎푑푖푎푙

Radial SNRe was computed by considering the mean and standard deviation of the radial strain throughout the entire myocardium. Both inter-frame and cumulative radial strains were used for

SNRe calculations. Cumulative radial strain was used to compare the end-systolic strain distribution estimated by each sequence. Since myocardial velocity changes rapidly throughout

the systolic phase, it is also of interest to compare the effect of the compounding sequence at various time points during contraction. For this comparison, inter-frame radial strains were used to calculate SNRe.

Although the “true” cardiac strain values were not known for any of the volunteers, the wide range of cardiac imaging literature can be exploited to establish a general set of expectations for cardiac strain values in the healthy human heart. Since healthy human volunteers were used for this study, the LV is assumed to contract normally, i.e. positive radial strain occurring throughout the short-axis view. Although there is no consensus on the normal values of cardiac radial strain in humans, echocardiographic studies have reported radial strains between 35.1% and 59.0%135.

Direct comparison of displacements and strains between each sequence is possible due to the hybrid acquisition allowing capture of the same cardiac cycle for both sequences.

4.5.3 Results

Envelope detection by the Hilbert transform is used to generate readable B-mode images from the RF data, which can be used to establish an overview of the RF data quality. B-mode images generated from two subjects using the diverging and compounding acquisitions are displayed in

Figure 4-8. The compounding B-modes exhibit better contrast, especially within the cavity and along the endocardial and epicardial borders. In particular, the compounding B-mode for Subject

4 displays significantly increased resolution of the lateral wall compared to the diverging acquisition.

End-systolic axial and lateral displacements for the two subjects displayed in Figure 4-8 are overlaid onto the compounding B-modes and presented in Figure 4-9. Axial displacements appear fairly similar in the diverging and compounding acquisitions for both patients. However, the lateral displacements estimated by the compounding acquisition show more significant differences compared to the diverging acquisition in both subjects. Differences in lateral displacement are especially noted in the lateral wall for Subject 4, which is the same region that

Figure 4-9: Example cumulative end-systolic axial and lateral displacements for the diverging and compounding acquisitions in two subjects.

Figure 4-10: Example cumulative end-systolic axial and lateral strains for the diverging and compounding acquisitions in two subjects.

exhibited better contrast in the B-mode images in Figure 4-8. In the diverging sequence, the lateral displacements in the lateral and septal walls of Subject 4 are fairly close to zero, while in the compounding acquisition they are more indicative of physiologic displacement, i.e. leftward and rightward motion respectively.

End-systolic axial and lateral strains for the two subjects are displayed in Figure 4-10.

Myocardial thickening is shown as positive strain (red) while myocardial thinning is displayed as negative strain (blue). Again, axial strains appear fairly similar between the diverging and compounding sequences, although larger differences are noted in the strains compared to the displacements. In particular, the compounding acquisition results in more homogenous distribution of positive axial strains in the posterior wall, especially in Subject 4. Lateral strains differed more significantly between the two acquisitions, although many regional features remained consistent. Differences in end-systolic lateral strain can be especially noted in the lateral and septal walls of both subjects, where the compounding sequence estimation gives rise to more consistent distribution of positive strain.

Cumulative end-systolic radial strains for all four subjects used in this study are displayed in

Figure 4-11. Quantitative statistics regarding the strain cumulative end-systolic strain estimation performed by each sequence is listed in Table 4-1. In general, the expected distribution of positive radial strain at end-systole was obtained more consistently using the compounding sequence compared to the diverging sequence. This observation is reflected by the increase in the radial SNRe at end-systole for the compounding sequence as presented in Table 4-1.

Furthermore, regions of negative end-systolic radial strain, which are likely due to estimation

Figure 4-11: Cumulative radial strain at end-systole for all subjects using the diverging and compounding sequences.

Table 4-1: Quantitative comparison of end-systolic radial strain (%) for all subjects using the diverging and compounding acquisitions.

artifacts based on the assumption of positive radial strain, were significantly decreased in the compounding sequences compared to the diverging sequences.

Plots of SNRe over time for inter-frame radial strains are displayed in Figure 4-12. For all patients, radial SNRe was low at the beginning of systole and increased throughout the cardiac cycle before falling off at end-systole. Although the diverging and compounding acquisitions followed the same general trends within each patient, the compounding acquisition exhibited higher SNRe throughout the vast majority of the cardiac cycle for all patients. The SNRe improvements gained from the compounding sequence are the most significant at the beginning of systole during the rapid onset of contraction. Towards end-systole, radial SNRe for the two sequences tend to converge to a similar value.

Figure 4-12: Comparison between the diverging and compounding sequences of radial SNRe over time for each of the four subjects.

4.5.4 Discussion

The goal of this study was to examine the effects of coherent spatial compounding on cardiac strain estimation in the in vivo setting. Although previous investigations have established the utility of coherent compounding in other areas of ultrasound research, this study is the first to present the effects of compounding for two-dimensional cardiac strain estimation in vivo. In general, we found that coherent spatial compounding leads to an increase in strain estimation performance in vivo. This observation is reflected by a reduction in radial strain artifacts and is quantitatively captured by an increase in both end-systolic and inter-frame radial SNRe.

Subjects 2 and 4 were chosen for display in order to demonstrate both moderate and high levels of improvement in strain estimation as a result of coherent spatial compounding. Fig. 4-8 provides a visual depiction of the effect of compounding at the level of the RF data. It is clear

from visual inspection that coherent spatial compounding increases the contrast between myocardium and the surrounding environment, which has been demonstrated previously. For

Subject 4, the lateral wall of is can be resolved much better in the compounding acquisition, which is reflected by the marked differences in lateral displacement and strain in this location.

The poor contrast of the lateral wall in the diverging acquisition could be the result of shadowing imposed by the ribs and/or lungs. Imaging artifacts can be extremely disruptive to strain estimation since the speckle pattern does not accurately reflect the underlying tissue structure and may disrupt accurate motion estimation by cross-correlation. Because strains are computed from the spatial gradient of the displacement field, concentrated regions of poor displacement estimation can lead to even larger errors in the strain calculation.

The in vivo end-systolic strains obtained using each acquisition appear more similar in the axial direction compared to the lateral direction (Figure 4-9). Of course, it is important to note that lateral strain estimation is in general characterized by higher noise than axial strain estimation due to the lack of phase information in the lateral direction136. This effect is observed in the noisier lateral images in Figures 4-9 and 4-10. Because lateral strain estimation relies primarily on the amplitude of the RF data, the contrast and resolution of the various tissue structures contained within the RF data are extremely important to ensure accurate lateral strain estimation.

Subject 4 in particular demonstrates significant improvements in lateral displacement and strain estimation due to the increased resolution of the LV tissue using the compounding acquisition.

Thus, the improvement in lateral strain estimation in vivo presented in this study aligns well with previous investigations.

Improvements in lateral strain estimation may explain why end-systolic radial strain estimations are particularly improved in the lateral and septal walls, since positive lateral strain is the primary component of radial strain in these locations. Regions of negative radial strain, which are unlikely to occur in healthy subjects, are significantly reduced in the compounding acquisitions shown in Figure 4-11. For example, the end-systolic radial strains for Subject 1 appear extremely similar in Figure 4-11, except for the regions of negative strain located in the septal and lateral walls. The negative amplitude of the strain in these regions is significantly reduced in the compounding acquisition, leading to a more homogenous and physiologic distribution of end-systolic radial strain.

The quantitative metric of radial SNRe was chosen to avoid the use of arbitrary ROIs for axial and lateral strain. Unlike axial and lateral strain, radial strain should manifest as a relatively uniform distribution throughout the short axis view of the LV myocardium, especially in healthy subjects. Both end-systolic radial SNRe (Table 4-1) and inter-frame radial SNRe (Figure 4-12) are increased using the compounding acquisition. In all subjects, the compounding sequence improves radial SNRe the most during early systole, when myocardial motion is high. The rapid deformations experienced by the heart at early systole lead to large inter-frame strains and thus high decorrelation. At end-systole when motion is significantly less, the benefits gained from coherent spatial compounding are reduced.

It is important to discuss the limitations of the in vivo data collection used in this study.

Simulation studies are commonly used to investigate the effect of new acquisition techniques; however, it can be difficult to simulate the various sources of noise inherent to the in vivo setting.

In particular, shadowing and high reflection artifacts which can occur often in vivo are preserved in this study as can be seen from Figure 4-8. Furthermore, the out-of-plane motion experienced by the twisting LV is difficult to replicate in simulation. Therefore, the use of in vivo data was preferred in this study in order to specifically investigate the effects of compounding within the clinical setting. By employing a hybrid sequence, each volunteer was able to be imaged nearly simultaneously using the compounding and diverging acquisitions, ensuring that the segmentation of the myocardium as well as the choice of systolic phase is kept consistent.

Because each frame of the compounding acquisition is directly paired with a corresponding frame of the diverging acquisition, a relative comparison of the strain estimations can be performed with high confidence.

4.6 Conclusion: Specific Aim 1

In these two studies, the effects of MER and coherent spatial compounding for in vivo strain estimation were explored. We have shown that increasing MER is especially beneficial for axial strain estimation, while spatial compounding can help increase the single-to-noise ratio of the RF data with particular benefits to lateral strain estimation. In the context of the Strain Filter predicting the optimal parameters for strain estimation, MER is related to the correlation coefficient (ρ), while the technique of spatial compounding will affect SNRs.

Increasing MER has been well-established by the literature as being beneficial for strain estimation. Results from the MER study indicate that around approximately 500 Hz, the peak in

E(SNRe|ε) corresponding to the more precise Cramer-Rao lower bound begins to dominate. At an MER of 2000 Hz, estimation governed by the less precise Barankin bound is almost

completely eliminated. However, the use of diverging beams to increase MER leads to a tradeoff in spatial resolution. In the second study, we have shown that coherent spatial compounding can be used to recover some of the lost spatial resolution leading to more accurate strain estimation.

Lateral strain estimation in particular is significantly aided by the increase in spatial resolution provided by the compounding technique, leading to a more accurate estimation of radial strain.

It is important to keep in mind the tradeoff between temporal resolution and spatial resolution when designing ultrasound acquisition sequences for strain estimation purposes. Although a more rigorous parametric evaluation was performed for MER in the first study, it is clear that both high MER imaging and spatial compounding should be implemented for clinical applications. The hybrid sequence used in the second study consisted of two separate sequences at 500 Hz each using 5 transmits per frame. Thus, it is possible to design a sequence to image at a 14 cm depth with an MER of 1000 Hz and 5 coherently compounded waves per frame, or with an MER of 500 Hz and 10 coherently compounded waves per frame.

Chapter 5

Myocardial Elastography for Cardiac Ablation Monitoring

5.1 Introduction

Radiofrequency (RF) ablation procedures aim to correct various types of arrhythmia by thermally ablating select regions of the cardiac tissue thought to contribute to the abnormal rhythm. Success rates for ablation procedures have been reported in the range of 53-57% for a single procedure and 71-80% after multiple procedures26,28. The size, spacing, and depth of the lesions in the tissue has been shown to be critical to the success of an ablation procedure29–32, motivating the development of new methodologies to characterize lesion formation in real-time.

Conventionally, lesion size has been controlled using temperature, amplitude and duration of the

RF energy provided by the ablation system33,34. Contact force of the ablation catheter has also recently been proposed as a method to control lesion size35,36. However, these methods mainly rely on indirect feedback derived from the ablation catheter instead of direct assessment of myocardial tissue properties and function. As a result, several imaging techniques have been developed to provide more direct mapping of lesion formation within the myocardium. Magnetic resonance imaging137,138 has been used to characterize lesion size; however, the specialized equipment and long acquisition times required do not allow real-time monitoring during the procedure. On the other hand, echocardiography has advantages of being fast, low-cost, and already highly integrated into the field of cardiology. Intracardiac Echocardiography (ICE) is often used during cardiac ablation procedures in order to visualize and monitor the cardiac motion during surgery. The ICE probe may be introduced into the heart using the same route as

the ablation catheter, allowing for a convenient imaging view of the ablation site. ICE catheters integrated with high-frequency ultrasound have been used to map lesion formation in real time, using the change in tissue contrast to identify lesion formation139,140. Other ultrasound methods use the fact that stiffening of the cardiac tissue occurs as a result of lesion formation. Shear wave elastography (SWE) uses the shear wave propagation induced by a “push” beam to map tissue stiffness and has been applied to the detection of ablation lesions in beating hearts using an ICE catheter141–144. Acoustic radiation force imaging (ARFI) also uses an acoustic “push” beam to evaluation the displacement at the push location to provide a qualitative stiffness estimation and was applied to lesion detection in several studies 145,146. However, the significant attenuation experienced by the high-frequency probe and by the push beam in the case of SWE and ARFI has thus far required that the imaging transducer to be located within a few centimeters of the lesion. In clinical practice, it may prove challenging to manipulate and align the imaging and ablation catheters in close proximity during the procedure. ICE has also been used for mechanical characterization of the myocardium using strain 83,147 and strain rate imaging148,149.

Recent advancements in plane wave and diverging beam transmissions have been applied to ICE as well83,143. Diverging beam transmissions allow for the simultaneous interrogation of a large field of view, allowing for high frame-rate imaging which has been linked to superior motion and strain estimation116,150. Intracardiac Myocardial Elastography has been developed to perform real-time strain imaging using a diverging beam transmission and a fast method of normalized cross-correlation86. By taking advantage of the phase of the ultrasound RF data, strain estimation using normalized cross-correlation has been shown to provide for high precision displacement and strain estimation108,123. In regards to cardiac ablation, ME using ICE has previously

demonstrated feasibility in detecting the formation of ablation lesions in the atria of humans and canines using a clinical ultrasound scanner83,84. However, correlation between lesion size and strain change has not yet been established. In order to investigate this relationship, a canine model (n=5) of radiofrequency ablation was used to compare lesion size estimated from intracardiac ME with the true lesion size measured following tetrazolium chloride (TTC) staining of the excised tissue. This study was conducted in the canine LV to allow for greater flexibility in lesion generation since the LV walls are larger and thicker compared to the atria.

5.2 Methods

Using a protocol approved by the Columbia University Institutional Animal Care and Use

Committee, five mongrel dogs weighing approximately 25 kg each were anesthetized using 0.15 mg/kg morphine and sustained under 2-5% inhaled isoflurane. A lateral thoracotomy procedure was used to expose the heart for placement of the ablation catheter (TactiCath, St. Jude Medical) on the epicardial surface of the left ventricle (LV) (Figure 5-1). An 6-MHz ICE catheter

(ViewFlex, St. Jude Medical) was introduced into the external jugular vein and advanced through the superior vena cava into the right ventricle of the heart, where it was positioned to have a view of the lateral wall of the LV. Alignment between the ICE plane and ablation catheter was achieved by noting the presence of the high reflection produced by the ablation catheter in the ICE image (green arrow in Figure 5-1). The ICE catheter was connected to a clinical ultrasound machine (z.one, Zonare Medical Systems) programmed to emit a specialized sequence for ME imaging which is described in the next section.

Figure 5-1: ICE catheter position within the right ventricle.

Ablation lesions were generated at various locations in the epicardium of the left ventricle after confirming alignment between the ablation catheter and ICE view as described above. Ten lesions of different sizes were generated in three dogs by varying power settings (10-30W), contact force (10-30 g) and application time (60-90 s). In two dogs, lesion lines were formed having multiple lesions next to each other within the same ICE view. Ultrasound acquisitions were performed before and ~2 minutes after the induction of each individual ablation lesion.

The ME acquisition sequence was designed for high frame-rate imaging and consisted of repeated diverging wave emissions at 600 Hz at a 10 cm depth for 2 seconds to ensure capture of the entire cardiac cycle. For each diverging wave, delay-and-sum reconstruction was used offline to beamform an image consisting of 193 lines of ultrasound RF data, resulting in a field of view of 90° at an axial spatial sampling rate of 18 MHz. ECG measurements were also obtained in synchrony with ultrasound imaging. A fast method for normalized cross-correlation was used to

estimate axial displacements between consecutive frames (window size = 6.16 mm, overlap =

90%). These inter-frame displacements were tracked throughout systole and accumulated over time to form cumulative displacements. Cumulative strain was derived from the spatial gradient of the cumulative displacement measurements using a 1-D axial kernel (window size = 0.515 mm). Selection of the systolic phase was guided by the ECG and an M-mode image of the displacements of the center line of the image stack.

Figure 5-2: Methods for determining lesion volume and transmurality based on strain computed from (A) intracardiac ME and (B) TTC staining.

The cumulative end-systolic strain distribution estimated using ME was used to quantify the mechanical changes caused by the ablation lesion and to generate a lesion map. Differences in end-systolic cumulative strain before and after the induction of each lesion were calculated.

Probe motion between acquisitions was corrected for using an image registration method based on gradient descent to compare the B-mode images at end-diastole151. A threshold of an absolute

8% reduction in strain was applied to filter out smaller differences in strain not indicative of the mechanical changes caused by ablation. This threshold allowed a visualization of the lesion area in the region of the image near the ablation location, i.e. a lesion map. Both the area and transmurality of the lesion were computed from the lesion maps as shown in Figure 5-2. Lesion area was computed by counting the number of pixels contained within the thresholded region of the lesion. Lesion transmurality was calculated as the ratio of the lesion depth (DL) to the wall depth (DW), with each depth measured perpendicular to the epicardial surface. These two values were computed from the lesion map for all lesions and compared to the actual lesion volume and transmurality measured ex vivo.

Following the experiment, the lesions were excised from the cardiac tissue and subjected to tetrazolium chloride (TTC) staining, trichrome staining, and measurement. TTC staining highlights healthy myocardium using the dehydrogenase enzyme, causing healthy regions to appear red and necrotic regions to appear white-gray152. Lesion volume was calculated from gross pathology by assuming a half ellipsoid shape and measuring the three appropriate axes in the TTC-stained tissue (L, W, DL) as shown in Figure 5-2. Lesion transmurality was similarly computed as the ratio between the lesion depth (DL) and the wall depth (DW).

5.3 Results

5.3.1 Lesion Size and Transmurality

Figure 5-3 depicts the end-systolic strain in the LV before and after the application of ablation for a single isolated lesion. Before ablation, end-systolic strain is positive in both the septum (top of image) and lateral wall (bottom of image) in each case. After ablation, a significant decrease in strain is observed near the epicardium of the lateral wall. This difference in strain is further highlighted by the strain difference image (Figure 5-3c) and lesion map image (Figure 5-3d). The regions of the myocardium remote from the lesion site also show similar end-systolic strains before and after the ablation. The size and depth of the lesion delineated by the lesion map appears to be similar to the appearance of the lesion in gross pathology.

Figure 5-4 plots the area of strain reduction against the actual lesion volume measured ex vivo after TTC staining for all ten isolated lesions. A good correlation is obtained between ME- measured strain changes and lesion size (r2 = 0.86).

Figure 5-4: Area of strain reduction (>8%) vs. excised lesion volume measured after TTC staining.

Figure 5-5 presents a Bland-Altman plot of the correlation in lesion transmurality assessed using

ME versus TTC staining. Differences in lesion transmurality ranged from -16.2% to 8.9%, with a mean difference of 4.2%. All values fell within the 95% confidence interval established by 1.96 standard deviations from the mean.

Figure 5-5: Bland-Altman agreement between lesion transmurality measured using ME strain and gross pathology.

5.3.2 Lesion Gap Imaging

Figures 5-6 and 5-7 depict the end-systolic strain before and after the generation of lesion lines consisting of several individual lesions. The line shown in Fig. 6-6 was generated using two lesions with no gap in between. Accordingly, the strain difference caused by the lesion is relatively homogenous and the two individual lesions cannot be distinguished from each other.

The absence of a gap between the lesions can also be confirmed using images of the epicardial surface and the ex vivo tissue with TTC staining (Fig. 6-6d). The line shown in Fig. 6-7 was generated using three lesions with small gaps between each lesion, as can be seen from the images of the epicardial surface (Fig. 6-7d). Compared to the previous lesion line, the end- systolic strain after ablation for this line appears less homogenous due to the presence of gaps

within the lesion line. This heterogeneity is reflected in the lesion map image, which depicts the

presence of gaps in between the individual lesions.

Figure 5-6: End-systolic strain distribution obtained in the canine LV after the formation of a two- lesion line having no gaps. Cumulative end-systolic strain is shown (A) before ablation and (B) 2 minutes after ablation, along with (C) the thresholded lesion map in comparison to (D) gross pathology.

5.4 Discussion

In this paper, a strain-based method for characterizing the size and transmurality of ablation

lesions using an intracardiac echocardiography system was described which can be used during

interventional procedures in real-time. We have demonstrated that a local reduction in end-

systolic strain is associated with the formation of an ablation lesion, thereby allowing for lesion

mapping to be performed within the ICE image. We have demonstrated that the region affected

by a significant decrease in strain is well-correlated to both lesion volume and lesion transmurality. Gross pathology with TTC staining of the excised lesion tissue was used to calculate the actual lesion volume and transmurality and formed the basis of the gold standard comparison in each case. We have also demonstrated the initial feasibility of detecting gaps in lesion lines using our technique.

Strain estimated using ME can be used to provide a real-time mechanical characterization of the target region of ablation procedures and the surrounding tissue. Before ablation, cardiac strain is consistently positive throughout the septal and lateral walls, corresponding to the mechanical contraction and thickening that occur throughout systole (Figure 5-3a). As can be seen from

Figures 5-2, 5-3, 5-6, and 5-7, cardiac strain exhibits significant, localized changes following ablation, while remote regions remain relatively unchanged. In general, the center of the lesion exhibits negative strain, while the surrounding tissue continues to exhibit positive strain. A transition zone between the negative and positive strain is also noted, which may be related to the gradient in thermal energy experienced by the tissue during ablation. The threshold of an 8% strain reduction was based on the average lesion strain decrease estimated in order to objectively quantify the lesion size. However, the proper selection of this threshold and its relationship to lesion border zones, especially in the context of arrhythmia prevention, warrants further investigation.

Using a diverging beam acquisition, we have achieved an imaging frame-rate of 600 Hz to increase the precision of the 1-D strain estimation. However, as discussed in Chapter 4, diverging beam acquisitions suffer a trade-off in spatial resolution, particularly in the lateral direction130. In this study, strain estimation was performed in the axial direction only to avoid the

inherently noisier estimations of lateral strain. Accordingly, the ICE views were intentionally positioned so that the LV walls were approximately perpendicular to axial direction of the ultrasound beam, allowing for lesion depth to be aligned with direction of axial strain estimation.

In this image configuration, axial strain is equivalent to the radial strain component of cardiac contraction. However, the location of the lesions within the ultrasound image will likely be more varied in the clinical setting. Thus, 2-D strain estimation may be warranted in order to compute radial strain in clinical images where wall alignment is difficult to control. The use of 3-D imaging would increase the field of view for lesion mapping and eliminate any alignment ambiguities that exist as a result of the imaging plane; however, 3-D imaging requires a specialized probe to collect volumetric data. Recently, we have demonstrated in vivo lesion detection in 3-D using a transthoracic matrix array probe153. Intracardiac systems with matrix array probes have also recently been developed to enable volumetric intracardiac imaging154,155.

Since the depths of the lesions in this study were aligned with the axial direction of the ultrasound beam, lesion transmurality was well-correlated with gross pathology. In the clinical setting, knowing the transmurality of a lesion is critical in order to achieve permanent arrhythmia termination32. However, the resolution of the lesion detection in the transverse direction is also important, especially in the context of identifying gaps in lesion lines. Methods such as coherent spatial compounding could be used to increase resolution in the lateral direction while still maintaining a high frame-rate130. Although the initial feasibility of detecting gaps was demonstrated in this study, there was not sufficient data to precisely characterize the resolution of ME in detecting gaps in the lateral direction. Future studies should aim to investigate the lateral resolution of lesion detection using ME.

The widespread use of echocardiography in interventional cardiology is in part related to its low cost, ease of use, portability, and safety. Echocardiographic techniques to provide lesion mapping capabilities have been the topic of several reports in the literature. Unlike radiation force-based techniques, ME exploits the natural contraction of the myocardium place of a push beam, rendering direct stiffness mapping more challenging since the tissue stress is unknown.

However, ME enjoys other advantages related to the absence of the push beam, namely depth penetration and temporal resolution. Since ME is not limited by the penetration depth and location of the acoustic push beam, it enables lesion mapping throughout the entire view of the heart within a single cardiac cycle. Furthermore, the location of the lesion does not need to be precisely known before imaging, as long as it is capable of being captured within the ICE view.

This may be especially important when imaging multiple lesions or lesion lines, as the transmission settings and ICE view need not be adjusted depending on the location of the ablation site within the image. The high temporal resolution achieved in this study (600 Hz) is also sufficient enough to perform EWI analysis using the same data. The propagation of the electromechanical wave may be disrupted by the induced lesion, with zero-crossing analysis serving as a different technique to perform lesion mapping. Previous studies have indicated that cardiac electrical conduction velocity is altered in ischemia156,157 and EWI has recently been used to image electromechanical wave speed and to identify regions of infarct and corresponding border zones158. Thus, there remain several possibilities for the inclusion of intracardiac EWI in the realm of RF lesion mapping.

5.5 Conclusion: Specific Aim 2

The results of this study support the use of Myocardial Elastography to monitor the formation and assess the size of ablation lesions in vivo. As expected, the presence of an ablation lesion within the myocardium leads to decreased end-systolic cumulative strain as assessed by ME.

Furthermore, this study has demonstrated that the size of the area affected by the reduction in cumulative systolic strain is directly correlated to lesion size. Initial feasibility in lesion gap identification was also demonstrated. Since ICE imaging already plays an important role in conventional clinical practice during RF ablation procedures, use of imaging techniques such as

ME may serve to augment the role of ICE during ablation procedures and provide a means for lesion visualization, currently absent from the standard of care. The lesion mapping capabilities of ME may serve to increase the success rates of ablation procedures by allowing assessment of the size, spacing, and depth of lesions in real-time during the procedure.

Chapter 6

Electromechanical Wave Imaging for Monitoring of Cardiac

Resynchronization Therapy

6.1 Introduction

Heart failure (HF) affects more than 5 million Americans each year and is projected to cost the

U.S. healthcare system over $40 billion by 2020159. Cardiac resynchronization therapy (CRT) is an established treatment for HF patients with left ventricular (LV) systolic dysfunction and QRS prolongation, especially resulting from left bundle branch block (LBBB). Multiple large, randomized clinical trials of HF patients have demonstrated that CRT may decrease mortality and improve LV systolic function and patient quality of life38,39. However, there remains a large subset of patients (30-40%) that are considered non-responders and do not derive significant benefits from CRT following device implantation4,41. This rate of non-response is particularly troublesome due to the high cost and invasiveness of the treatment, and in light of expanded indications to a larger group of HF patients40,41.

Accordingly, research groups and clinicians have been highly motivated to uncover mechanisms to increase the response rate of CRT. Recent research has introduced the idea of “patient- specific” optimization of CRT that aims to tailor the therapy to the condition of each individual patient. Several studies have reported the utility of targeted LV lead placement, using the site of latest electrical activation46–48 or a region that is remote from scar49. Other studies have shown that the offset between LV and RV pacing can be optimized for the individual patient to improve

hemodynamics and functional characteristics of the LV50,51. As pacing lead technology continues to evolve, multi-site pacing of the LV may also be useful over the long-term to improve the rate of response to CRT52.

The variety of approaches to address the problem of non-response to CRT highlights the uncertainties regarding the exact physiological benefits of CRT. For instance, although the concept of synchrony is central to CRT, it remains unclear whether intraventricular (within the

LV) or interventricular (LV-RV) synchrony provides the largest benefit to patients160,161.

Furthermore, assessment of dyssynchrony can be performed on the basis of either electrical or mechanical parameters. Electrical measures of dyssynchrony obtained using the ECG are now used to select patients for CRT, but their relationship to CRT response may be limited. QRS duration (QRSd) provides only a general prediction of CRT response and clinical trials have shown poor correlation between QRSd and outcome42. Mechanical dyssynchrony parameters have shown promise in determining response in small observational studies, although large, multi-center trials have failed to validate the predictive power of these approaches. The largest prospective trial (PROSPECT) showed poor predictive value of several echocardiographic parameters derived using tissue Doppler4. However, echocardiography still remains a favorable imaging platform to assess HF due to its low cost, portability, and widespread integration in clinical practice. Speckle-tracking echocardiography has been used to quantify dyssynchrony associated with response to CRT43, although recent studies using speckle tracking dyssynchrony measures to predict response to CRT have shown mixed results44,45.

Figure 6-1: Comparison of the strain computed using EWI and traditional strain imaging methods. EWI (blue) captures small inter-frame strains and can precisely identify the strain zero-crossing in response to electromechanical activation (orange circle), while traditional strain imaging (green) computes the global total strain and often relies on time to peak strain (red circle).

Much of the previous research has ignored the fundamental relationship that exists between the electrical and mechanical aspects of the heart. Because the heart functions as an electromechanical pump—using electrical activation to induce mechanical contraction—it is highly likely that some combination of these two factors is required to promote optimal therapy.

Although traditional imaging techniques tend to assess one side or the other, examination of the coupling between the electrical and mechanical behaviors of the heart may provide useful information. This relationship is reflected by the electromechanical wave, i.e., the depolarization wave that initiates cardiac contraction162. Mapping of the electromechanical wave requires high temporal resolution due to the short time course of this behavior (<50 ms). Conventional

echocardiography operating at low frame-rates (<100 Hz) does not provide the required temporal sampling to precisely capture the strain corresponding to the electromechanical wave, which can occur in less than 50 ms (Figure 6-1).

Electromechanical wave imaging (EWI) is a non-invasive, ultrasound-based technique that uses the transient initiation of myocardial strain to depict the electromechanical activation of the heart102. EWI differs from conventional strain imaging in two major ways. First, EWI uses a transmission sequence that allows for extremely high frame rates (up to 2000 Hz) to capture the cardiac contraction at very small time increments (0.5 ms) in order to decrease motion-induced artifacts that are known to corrupt the strain estimation. High frame-rate ultrasound has consistently been shown to be of particular use in cardiac applications to capture the rapid deformations experienced by the heart during contraction117,150. Secondly, EWI uses a strain estimation technique based on cross-correlation of radiofrequency (RF) signals instead of conventional B-mode block matching to improve the precision of strain estimation. RF-based motion estimation also allows for the detection of extremely small displacements (on the order of

10 microns) that exist from frame-to-frame when imaging at extremely high frame-rates. Several ultrasound-based research groups have reported the superiority of cross-correlation techniques compared to B-mode-based speckle tracking, especially for small deformations108,123. The detection of such small motion at such a high frame-rate allows EWI to map the rapid onset of strain that occurs in response to electrical activation, i.e. the electromechanical wave (Figure 6-

1).

Previous studies have shown the initial feasibility of EWI in mapping the activation origins in atrial tachyarrhythmia and ventricular pacing97,102. In this study, we use EWI as a novel technique to characterize the electromechanical behavior in five healthy volunteers and 16 HF patients that have been retrospectively identified as responders or non-responders. Differences between the two response groups in native rhythm and with CRT pacing are quantified on the basis of a parameter derived from ventricular EWI maps. The reproducibility of EWI is also investigated in six patients using consecutive acquisitions of the same pacing protocol.

6.2 Methods

Using a protocol approved by the Columbia University Institutional Review Board, HF patients with CRT devices were recruited for the study from November 2012 to November 2013 during routinely scheduled device interrogations in the device clinic at New York Presbyterian Hospital.

Implantation of CRT devices was undertaken prior to the time of the study according to

ACCF/AHA standard guidelines163. The LV lead position was selected during the procedure on the basis of coronary sinus anatomy and pacing threshold and was not guided by any electrical or echocardiographic parameters. All patients were programmed for biventricular pacing. Inclusion criteria for recruitment consisted of a diagnosis of LBBB or IVCD by 12-lead electrocardiogram

(ECG) with minimum QRSd of 130 ms, a sustainable underlying rhythm (non-pacemaker dependent), and the absence of atrial fibrillation.

Among 24 recruited subjects, nine patients were excluded from the final analysis on the basis of poor echocardiogram quality (n=3) and lack of response data (n=6), resulting in a study sample of 16 HF patients (10 non-ischemic, 6 ischemic). CRT response was defined according to ≥5%

increase in LV ejection fraction (LVEF) comparing pre- and post-implant clinically-acquired

echocardiograms (n=8). The mean time from CRT implant to EWI acquisition was 1216±977

days in responders compared to 1062±924 days in non-responders, which was found to be non-

significant by two-tailed Student’s t-test (p=0.75). Finally, a second control group of five healthy

volunteers without CRT was also recruited. Baseline statistics of all subjects included in the final

analysis are listed in Table 6-1.

Table 6-1: Clinical characteristics of healthy subjects and HF patients involved in the study. * indicates a significant change (p<0.01) in either QRSd or LVEF as a result of CRT by paired Student’s t-test.

EWI acquisitions were obtained in patients by a trained echocardiographer in the standard apical

4-chamber view. CRT pacing settings were adjusted using telemetry, allowing for imaging of

several different pacing protocols for each patient. For all patients, acquisitions were obtained

during pre-programmed biventricular (BiV) pacing and during native rhythm with the CRT

device set to a sensing only mode.

Ultrasound acquisitions were performed at a depth of 20 cm using a Verasonics V-1 system with a P4-2 phased array probe (f0 = 2.5 MHz). A diverging beam was used to simultaneously interrogate a full 90° field of view with a single transmit, allowing for extremely high frame- rates up to 2000 Hz102. The diverging beam acquisition lasted a full two seconds to ensure the uninterrupted capture of a full cardiac cycle. RF signals were acquired throughout the entire diverging beam sequence for each of the 64 individual transducer elements. Following the diverging beam acquisition, a conventional, focused acquisition was performed for two seconds at a framerate of 30 Hz, bringing the total acquisition length to four seconds. On-line beamforming was used to reconstruct RF signals from the conventional sequence into high quality B-mode images, which were solely used for visualization and segmentation purposes and not used for any strain calculations. Recordings of Lead II in the standard 12-lead electrocardiogram (ECG) were also obtained by the system in parallel with the ultrasound acquisitions.

EWI processing of the ultrasound data was performed blinded to the CRT response status of each patient in accordance with the methods outlined in Chapter 3.4. A delay-and-sum algorithm was used offline to reconstruct the signals obtained with the diverging beam sequence into 64 RF lines per frame at an axial sampling frequency of 20 MHz97. Using the ECG recordings, RF frame sequences were truncated to a time window of 700 ms around the peak of the R-wave for displacement estimation. A normalized cross-correlation algorithm (window size = 5.85 mm) was used to estimate inter-frame axial displacements throughout the entire image 86. Axial displacement estimates were obtained at an axial spacing of 0.36 mm and a lateral spacing of

approximately 0.19 mm. Inter-frame strains were then estimated from the displacements using a least square strain estimator with a window size of 7.83 mm111.

Previous EWI studies have used manual selection of the “zero-crossings” of inter-frame strain signals to generate an electromechanical activation map97,102. The existence of a zero-crossing, defined as a sharp transition from positive to negative strain, reflects myocardial shortening that occurs in the walls of the heart in the apical 4-chamber view in response to electrical activation

(Figure 6-1). In this study, a robust algorithm was used to automatically identify and select the zero-crossings corresponding to electromechanical activation for each point in the tissue.

Zero-crossings were initially selected by identifying the clear negative peaks of the inter-frame strains over time and searching backwards in time to find the corresponding zero crossing.

“Clear” negative peaks were constrained to only include peaks associated with:

1. a negative magnitude ≥20% of the global minimum of the strain and

2. a negative first derivative ≥20% of the global maximum.

These criteria allowed for filtering of zero-crossings associated with progressive shortening that are not indicative of electromechanical activation. For many strain signals, a single negative peak existed and thus computation of the zero-crossing was relatively simple. However, some strain signals exhibited more than one negative peak following the onset of the QRS complex. In these instances, the algorithm searched the surrounding region for a strain signal containing a single, clear peak and chose the zero-crossing closest in time. Once all zero-crossings had been identified, the activation time of each zero-crossing was computed with respect to a reference point, selected to be the time-point of the first deflection on the ECG following the P-wave, i.e.

the onset of the QRS complex. Activation times were only computed within the ventricular tissue as determined from segmentation. A spatial rendering of these activation times is referred to as the EWI activation map.

A quantitative parameter, the lateral wall activation time (LWAT) was also calculated from the activation maps by computing the mean activation time occurring in the LV lateral wall.

Selection of the LV lateral wall was manually performed on the basis of anatomical structures observed on the B-mode images.

6.3 Results

Figure 6-2: Ventricular activation isochrones from healthy volunteers. Activation times are mapped to the ECG displayed below the images (total time = 200 ms).

EWI ventricular activation maps for two healthy subjects in sinus rhythm are displayed in Figure

6-2. Early activation is designated in blue, while late activation is marked in red. The exact time of activation with respect to the onset of the ECG is indicated by the color scale superimposed

over each patient’s ECG tracing below the image. Ventricular electromechanical activation is initiated at the mid-level of the interventricular septum before propagation to RV free wall and

LV lateral wall. Although slightly varying patterns of activation are observed between subjects, activation of the LV lateral wall, septum, and RV free wall occurs nearly simultaneously during the early portion of the QRS complex. Activation of the entire lateral wall progresses quickly, and mean LWAT for healthy subjects was 61.1±24.0 ms.

Mean baseline LVEF in 8 responders was 18.9±7.3%, which increased significantly to

42.8±9.5% following CRT (p<0.01). In non-responders, mean LVEF decreased slightly from

25.7±7.2% to 20.0±6.7% after CRT, although this change was not significant (p>0.05).

Responders also exhibited a statistically significant reduction in QRSd as a result of CRT, while non-responders again showed no significant change (Table 6-1).

Activation maps for representative HF patents in native rhythm are shown in Figure 6-3 (a-c) for

CRT responders and Figure 6-4 (a-c) for CRT non-responders. Early electromechanical activation is again frequently observed in the mid-septal region. Early activation near the RV apex also occurs in most patients. However, in all patients, electromechanical activation in the

LV lateral wall is delayed with varying levels of severity. The delay in LV lateral wall activation in HF patients in native rhythm can be qualitatively observed in Figures 6-3 and 6-4 (a-c) and is also reflected by the increased mean patient LWAT (113.2±25.2 ms) compared to healthy subjects as shown in Figure 6-5. The difference in mean LWAT between all HF patients in native rhythm and healthy controls was 52.1 ms, which was found to be statistically significant

(p<0.01).

Corresponding EWI isochrones for the same HF patients during biventricular pacing are also shown in Figures 6-3 (d-f) and 6-4 (d-f). For all patients, the spatiotemporal sequence of LV activation is markedly different compared to native rhythm. Differences are especially noted in the LV lateral wall, where several patients exhibit significantly earlier activation sequences.

Mean LWAT was decreased by 29.5±40.9 ms in all HF patients during CRT (Figure 6-5), although separation by treatment response yielded more significant differences. Patients in

Figure 6-3, which have been retrospectively identified as responders to CRT, qualitatively

showed more homogenous activation sequences and reduced delay in lateral wall activation in response to CRT. Responders experienced a corresponding mean reduction in LWAT of

56.4±28.9 ms (Figure 6-5), which was found to be statistically significant using a paired comparison (p<0.01). Although non-responders also experienced changes in electromechanical activation in response to CRT, the resulting activation was highly variable, especially in lateral wall region. Non-responders still experienced largely delayed electromechanical activation even with CRT pacing and as a whole exhibited very little change in LWAT (0.7±24.9 ms) compared to native rhythm (Figure 6-5).

100

The reproducibility of EWI and the computed LWAT parameter was investigated in six patients in biventricular pacing. Consecutive EWI acquisitions were obtained of each patient in the same pacing protocol, allowing for two similar activation maps to be generated. Less than a minute of elapsed time occurred between the consecutive acquisitions. Figure 6-6 depicts the intra-patient variability of LWAT for all six patients where consecutive acquisitions were obtained. A high correlation (r2=0.92) was obtained between the intra-patient LWAT computations for all six patients.

101

Figure 6-6: Bland-Altman plot of LWAT reproducibility for consecutive EWI acquisitions of the same patient (n=6) in the same protocol.

6.4 Discussion

In this study, we demonstrate the feasibility of Electromechanical Wave Imaging in mapping the electromechanical activation in HF patients and demonstrating the differences according to response to CRT. A parameter derived from EWI activation maps, LWAT, was used to quantify the local electromechanical behavior of the LV lateral wall and differentiate HF patients from healthy controls and responders from non-responders to CRT.

In healthy subjects, EWI depicts relatively uniform activation of the ventricular tissue, including both interventricular and intraventricular synchrony. In contrast, HF patients in native rhythm all experienced significantly delayed activation in the lateral wall. Electromechanical activation sequences for HF patients were also more variable compared to healthy subjects, especially in the region of the LV lateral wall. In some cases, areas of early activation were located in close

102

proximity to areas of late activation. This phenomenon has been observed in several previous studies using contact and non-contact mapping techniques and is referred to in these studies as

“conduction block”46,164. Furthermore, the existence of ischemic or infarcted regions of tissue is known to cause fractionated electrical signals and mechanical decoupling165, which may also contribute to the discontinuous activation patterns observed in HF patients.

CRT pacing was shown by EWI to initiate a different electromechanical activation pattern within the ventricular tissue of HF patients compared to their underlying rhythm. Indeed, it is clear from

Figures 6-3 and 6-4 that both the origins of activation and propagation pathways are changed in

CRT pacing for the same patient. In native rhythm, several HF patients exhibited activation origins in the mid-septal region of the LV (Figure 6-2). However, with CRT pacing, activation origins consistently appeared in the RV apex and in the LV lateral wall (Figure 6-3). Notably, the locations of these early activated regions correlate well with the standard biventricular lead locations used for CRT.

CRT pacing also induced different behavior in the lateral wall in responders compared to non- responders. In responders, early activation of the lateral wall was widespread (Figure 6-3), while non-responders showed only a small, contained area of early activation (Figure 6-4). EWI was therefore capable of imaging the initiation of lateral wall contraction by the LV CRT lead, but also indicated that this activation did not quickly propagate throughout the entirety of the lateral wall in non-responders. The differences in lateral wall electromechanical activation between responders and non-responders may help explain some of the factors that determine CRT response. The results presented herein indicate that the regional electromechanical behavior of

103

the cardiac tissue of each patient during CRT may be linked in some way to response. Indeed, several publications have discussed the importance of regional myocardial electromechanics in the context of CRT and have developed their own techniques to interpret this behavior47,166. This idea aligns well with recent studies that have advocated for “patient-specific” optimization of various aspects of CRT to increase the likelihood of response49,51,164,166.

Figure 6-7: Mean change in QRSd and LWAT within each patient as a result of CRT.

Clinical trials have failed to demonstrate a strong correlation between initial QRS duration and

CRT response, especially for QRSd < 150 ms42,167. Although ECG parameters including QRSd remain important in the evaluation of HF and CRT, it may be of significant clinical benefit to identify other methods for evaluation of cardiac electromechanics. Ultimately, the QRS duration reflects the electrical behavior of both ventricles at a macroscopic level, which may not sufficiently capture the cardiac excitation-contraction coupling as it pertains to CRT. The

104

parameter investigated in this study (LWAT) is an electromechanical measurement localized to the LV, which may be affected differently by tissue geometry, myocardial scar, and patterns of electrical conduction compared to QRSd. In this study, QRSd was slightly increased in responders compared to non-responders, as was LWAT. This observation is not surprising, since it is generally accepted that patients with prolonged ventricular activation will benefit most from

CRT. Following CRT, responders exhibited significant reductions in both QRSd and LWAT; however, the differences observed in LWAT were of greater magnitude compared to QRSd

(Figure 6-7). It is also of note that the standard deviations of the LWAT measurements are increased compared to the QRSd measurements. As such, clinical use of an electromechanical parameter such as LWAT and its correlation to QRSd requires further validation in a large patient trial.

Cardiac strain has previously been used to assess dyssynchrony in HF patients with mixed success43–45. However, these studies mainly used B-mode-based speckle tracking techniques to map the heart on the basis of peak strain rather than the onset of strain. While peak strain may be closely correlated to the mechanical behavior of the heart, the rapid onset of strain, i.e. the strain

“zero-crossing”, may better reflect the underlying electromechanical characteristics of the tissue

(Figure 6-1). As can be seen in Figure 6-1, this zero-crossing occurs very rapidly (<50 ms) and would be extremely difficult to detect using low frame-rate imaging. By precisely identifying electromechanical activation on the basis of the inter-frame zero-crossing, EWI is capable of capturing and mapping the interaction between the electrical and mechanical aspects of the heart.

This approach stands to be especially useful in applications involving CRT, which by nature affects both the electrical and the mechanical behavior of the heart.

105

The reproducibility of results obtained using novel imaging techniques is always important, especially for those involving clinical applications. Although EWI has previously been validated using simulations and electrical mapping94,102, a small investigation into the reproducibility of the LWAT parameter obtained from EWI in HF patients was performed in this study. As seen in

Figure 6-6, LWAT values are fairly consistent within the same patient. These results indicate that

EWI and its subsequent generation of LWAT can be used reliably to characterize the activation sequence of HF patients with biventricular pacing.

It is important to note that this study primarily concerns confirming the existence of a response rather than predicting the likelihood of a patient to respond to CRT. The demonstrated capability for EWI in characterizing the electromechanical activation in HF and CRT pacing lends itself to several clinical applications concerning HF diagnosis and CRT treatment planning/monitoring, especially given the recent advancements in pacemaker technology. The ability of EWI to non- invasively identify regions of late activation may help guide implant procedures since EA mapping is typically not performed for these procedures. Furthermore, as CRT pacing protocols become more flexible with the introduction of multi-polar leads, EWI may assist clinicians in optimizing the CRT device for each patient in real-time by imaging the electromechanical response to various pacing settings.

Future investigations using EWI to study HF and CRT should include a prospective study design to build off the retrospective study performed here. Currently, response to CRT is assessed several months following device implantation to allow time for remodeling of the myocardium in response to CRT. Although we have demonstrated that EWI is able to image the instantaneous

106

change in electromechanical activation resulting from CRT, it is unclear if improvements in electromechanical activation will be indicative of future improvements in mechanical indices of function (such as LVEF and LVESV) or clinical composite score. Future investigations should also aim to integrate additional image windows besides the apical 4-chamber view to allow for more dense EWI mapping, particularly in the anterior and posterior walls of the heart. Additional views may provide a more complete view of LV electromechanical activation and could expose additional metrics for quantifications beyond the LWAT parameter computed here. Finally, additional ECG recordings could help improve the selection of the origin of electrical activation, since the earliest deflection from baseline may not be reflected in Lead II.

6.5 Conclusion: Specific Aim 3

The primary goal of this study was to demonstrate the utility of EWI in studying the electromechanical activation of HF patients and the changes that accompany CRT. We have shown that EWI can be used to map the ventricular electromechanics of both healthy and HF patients and is capable of distinguishing between these two groups both quantitatively and qualitatively. We have also reported the differences in activation that occur in response to CRT for responders versus non-responders and that this information can be localized to the LV and quantified. Finally, our parameter derived to quantify LV electromechanical behavior (LWAT) was shown to be reproducible within the same patient, reinforcing the findings and trends reported here.

However, the ability of clinicians to improve patient selection and outcomes after CRT is highly relevant. This problem is fairly complex and has been the subject of extensive study in the past.

107

Some previous studies have indicated that ultrasound-based parameters may be poor predictors of response to CRT4,45. However, the parameters used in these studies were mostly computed either using Doppler-based approaches, which inherently suffer from angle dependence, or conventional ultrasound operating at low framerates (<100 Hz). The angle-independent, high frame-rate technique of EWI can provide precise information at a much higher temporal resolution. Considering the short activation time of cardiac tissue, a high frame-rate acquisition is extremely important to capture the relevant motion, especially in pathologies such as heart failure that give rise to non-homogenous activation sequences. Furthermore, EWI is able to capture the local activation of strain (i.e. inter-frame strain zero-crossing), in contrast to conventional, low frame-rate speckle tracking techniques that are limited to measurements of time to peak strain. These unique characteristics of EWI make it highly promising as an investigative tool for HF patients and CRT.

108

Chapter 7

Summary of Work

The work presented in this dissertation has focused on the clinical translation of two high frame- rate echocardiographic strain imaging techniques—Myocardial Elastography and

Electromechanical Wave Imaging—using ultrasound beamforming and ultrasound image processing techniques.

High motion estimation rate imaging leads to increased precision in axial strain estimation.

Using diverging waves, ultrasound imaging sequences have been designed to enable motion estimation rates as high as 2000 Hz to perform 2-D cardiac strain estimation in vivo. However, diverging wave acquisitions are accompanied by a tradeoff of a fairly poor speckle signature which inhibits 2-D strain estimation. To improve the quality of the lateral strain estimation, the acquisition sequence can be modified to include coherent spatial compounding at only a moderate cost to framerate. An imaging sequence consisting of 5 compounded diverging waves was used at 500 Hz to improve the quality of the radial strain estimated by Myocardial

Elastography in healthy subjects.

Myocardial Elastography was also implemented using an intracardiac echocardiography (ICE) probe to monitor the formation of thermal ablation lesions in vivo. By imaging the changes in total systolic axial strain, lesion maps were generated which demonstrated good correlation with lesion volume and lesion transmurality as measured by gross pathology. Intracardiac Myocardial

Elastography was also used to identify the presence of gaps between neighboring ablation

109

lesions, which could be used to help improve the success of ablation procedures for atrial fibrillation.

An imaging sequence for Electromechanical Wave Imaging was used at 2000 Hz in a small clinical study of 16 heart failure patients and 5 healthy subjects. In order to avoid the tedious process of manual selection for zero-crossing detection, an automated method to detect the electromechanical zero-crossing was implemented. Electromechniacal activation maps were generated and used to compute a quantitative parameter, lateral wall activation time (LWAT), which characterizes the electromechanical activation of the lateral wall of the left ventricle.

LWAT was found to be able to discriminate between healthy subjects and heart failure patients and to image the clinical response of cardiac resynchronization therapy.

110

References

1. Mozaffarian D, Benjamin EJ, Go AS, et al.: Executive Summary: Heart Disease and Stroke Statistics--2016 Update: A Report From the American Heart Association. Circulation 2016; 133:447–454.

2. Heidenreich PA, Trogdon JG, Khavjou OA, et al.: Forecasting the Future of Cardiovascular Disease in the United States A Policy Statement From the American Heart Association. Circulation 2011; 123:933–944.

3. Sistrom CL, McKay NL: Costs, charges, and revenues for hospital diagnostic imaging procedures: differences by modality and hospital characteristics. J Am Coll Radiol JACR 2005; 2:511–519.

4. Chung ES, Leon AR, Tavazzi L, et al.: Results of the Predictors of Response to CRT (PROSPECT) Trial. Circulation 2008; 117:2608–2616.

5. Pellikka PA, Douglas PS, Miller JG, et al.: American Society of Echocardiography Cardiovascular Technology and Research Summit: A Roadmap for 2020. J Am Soc Echocardiogr 2013; 26:325–338.

6. Abraham TP, Dimaano VL, Liang H-Y: Role of Tissue Doppler and Strain Echocardiography in Current Clinical Practice. Circulation 2007; 116:2597–2609.

7. Fuster V, O’Rourke RA, Walsh RA, Poole-Wilson P: Hurst’s The Heart, 12th Edition. 12th Edition. McGraw-Hill Professional, 2007,.

8. Sarazan RD: Cardiovascular pressure measurement in safety assessment studies: Technology requirements and potential errors. J Pharmacol Toxicol Methods 2014; 70:210–223.

9. Abramson DI, Margolin S: A Purkinje Conduction Network in the Myocardium of the Mammalian Ventricles. J Anat 1936; 70:250–259.

10. Rohr S: Role of gap junctions in the propagation of the cardiac action potential. Cardiovasc Res 2004; 62:309–322.

11. Luo CH, Rudy Y: A model of the ventricular cardiac action potential. Depolarization, repolarization, and their interaction. Circ Res 1991; 68:1501–1526.

12. Burton FL, Cobbe SM: Dispersion of ventricular repolarization and refractory period. Cardiovasc Res 2001; 50:10–23.

13. Kuo CS, Munakata K, Reddy CP, Surawicz B: Characteristics and possible mechanism of ventricular arrhythmia dependent on the dispersion of action potential durations. Circulation 1983; 67:1356–1367.

111

14. Singh BN, Hauswirth O: Comparative mechanisms of action of antiarrhythmic drugs. Am Heart J 1974; 87:367–382.

15. Kligfield P, Gettes LS, Bailey JJ, et al.: Recommendations for the Standardization and Interpretation of the Electrocardiogram Part I: The Electrocardiogram and Its Technology: A Scientific Statement From the American Heart Association Electrocardiography and Arrhythmias Committee, Council on Clinical Cardiology; the American College of Cardiology Foundation; and the Heart Rhythm Society Endorsed by the International Society for Computerized Electrocardiology. Circulation 2007; 115:1306–1324.

16. Mason JW, Hancock EW, Gettes LS: Recommendations for the Standardization and Interpretation of the ElectrocardiogramPart II: Electrocardiography Diagnostic Statement List A Scientific Statement From the American Heart Association Electrocardiography and Arrhythmias Committee, Council on Clinical Cardiology; the American College of Cardiology Foundation; and the Heart Rhythm Society Endorsed by the International Society for Computerized Electrocardiology. J Am Coll Cardiol 2007; 49:1128–1135.

17. Freeman GL, LeWinter MM, Engler RL, Covell JW: Relationship between myocardial fiber direction and segment shortening in the midwall of the canine left ventricle. Circ Res 1985; 56:31–39.

18. Folland ED, Parisi AF, Moynihan PF, Jones DR, Feldman CL, Tow DE: Assessment of left ventricular ejection fraction and volumes by real-time, two-dimensional echocardiography. A comparison of cineangiographic and radionuclide techniques. Circulation 1979; 60:760–766.

19. Pfisterer ME, Battler A, Zaret BL: Range of normal values for left and right ventricular ejection fraction at rest and during exercise assessed by radionuclide angiocardiography. Eur Heart J 1985; 6:647–655.

20. Cordeiro JM, Greene L, Heilmann C, Antzelevitch D, Antzelevitch C: Transmural heterogeneity of calcium activity and mechanical function in the canine left ventricle. Am J Physiol - Heart Circ Physiol 2004; 286:H1471–H1479.

21. Prinzen FW, Augustijn CH, Allessie MA, Arts T, Delhass T, Reneman RS: The time sequence of electrical and mechanical activation during spontaneous beating and ectopic stimulation. Eur Heart J 1992; 13:535–543.

22. Haíssaguerre M, Jaís P, Shah DC, Gencel L, Pradeau V, Garrigues S, Chouairi S, Hocini M, Le Métayer P, Roudaut R, Clémenty J: Right and Left Atrial Radiofrequency Catheter Therapy of Paroxysmal Atrial Fibrillation. J Cardiovasc Electrophysiol 1996; 7:1132– 1144.

23. Roy D, Talajic M, Dorian P, Connolly S, Eisenberg MJ, Green M, Kus T, Lambert J, Dubuc M, Gagné P, Nattel S, Thibault B: Amiodarone to prevent recurrence of atrial fibrillation. Canadian Trial of Atrial Fibrillation Investigators. N Engl J Med 2000; 342:913–920.

112

24. Wolf PA, Abbott RD, Kannel WB: Atrial fibrillation as an independent risk factor for stroke: the Framingham Study. Stroke 1991; 22:983–988.

25. Patel NJ, Deshmukh A, Pant S, et al.: Trends of Hospitalization for Atrial Fibrillation in the United States, 2000 Through 2010: Implications for Healthcare Planning. Circulation 2014; :CIRCULATIONAHA.114.008201.

26. Calkins H, Reynolds MR, Spector P, Sondhi M, Xu Y, Martin A, Williams CJ, Sledge I: Treatment of Atrial Fibrillation With Antiarrhythmic Drugs or Radiofrequency Ablation Two Systematic Literature Reviews and Meta-Analyses. Circ Arrhythm Electrophysiol 2009; 2:349–361.

27. Bhakta D, Miller JM: Principles of Electroanatomic Mapping. Indian Pacing Electrophysiol J 2008; 8:32–50.

28. Ganesan AN, Shipp NJ, Brooks AG, Kuklik P, Lau DH, Lim HS, Sullivan T, Roberts- Thomson KC, Sanders P: Long-term Outcomes of Catheter Ablation of Atrial Fibrillation: A Systematic Review and Meta-analysis. J Am Heart Assoc 2013; 2:e004549.

29. Cappato R, Negroni S, Pecora D, Bentivegna S, Lupo PP, Carolei A, Esposito C, Furlanello F, Ambroggi LD: Prospective Assessment of Late Conduction Recurrence Across Radiofrequency Lesions Producing Electrical Disconnection at the Pulmonary Vein Ostium in Patients With Atrial Fibrillation. Circulation 2003; 108:1599–1604.

30. Melby SJ, Lee AM, Zierer A, Kaiser SP, Livhits MJ, Boineau JP, Schuessler RB, Damiano Jr. RJ: Atrial fibrillation propagates through gaps in ablation lines: Implications for ablative treatment of atrial fibrillation. Heart Rhythm 2008; 5:1296–1301.

31. Miller MA, d’Avila A, Dukkipati SR, Koruth JS, Viles-Gonzalez J, Napolitano C, Eggert C, Fischer A, Gomes JA, Reddy VY: Acute electrical isolation is a necessary but insufficient endpoint for achieving durable PV isolation: the importance of closing the visual gap. Europace 2012; 14:653–660.

32. Kobza R, Hindricks G, Tanner H, Schirdewahn P, Dorszewski A, Piorkowski C, Gerds-Li J-H, Kottkamp H: Late recurrent arrhythmias after ablation of atrial fibrillation: Incidence, mechanisms, and treatment. Heart Rhythm 2004; 1:676–683.

33. Nath S, Haines DE: Ablative Therapy for Cardiac Arrhythmias. I.Biophysics and pathology of catheter energy delivery systems. Prog Cardiovasc Dis 1995; 37:185–204.

34. Skrumeda LL, Mehra R: Comparison of Standard and Irrigated Radiofrequency Ablation in the Canine Ventricle. J Cardiovasc Electrophysiol 1998; 9:1196–1205.

35. Shah DC, Lambert H, Nakagawa H, Langenkamp A, Aeby N, Leo G: Area Under the Real-Time Contact Force Curve (Force–Time Integral) Predicts Radiofrequency Lesion Size in an In Vitro Contractile Model. J Cardiovasc Electrophysiol 2010; 21:1038–1043.

113

36. Thiagalingam A, D’avila A, Foley L, Guerrero JL, Lambert H, Leo G, Ruskin JN, Reddy VY: Importance of Catheter Contact Force During Irrigated Radiofrequency Ablation: Evaluation in a Porcine Ex Vivo Model Using a Force-Sensing Catheter. J Cardiovasc Electrophysiol 2010; 21:806–811.

37. Yancy CW, Jessup M, Bozkurt B, et al.: 2013 ACCF/AHA Guideline for the Management of Heart Failure A Report of the American College of Cardiology Foundation/American Heart Association Task Force on Practice Guidelines. Circulation 2013; 128:e240–e327.

38. Abraham WT, Fisher WG, Smith AL, et al.: Cardiac Resynchronization in Chronic Heart Failure. N Engl J Med 2002; 346:1845–1853.

39. Cleland JGF, Daubert J-C, Erdmann E, Freemantle N, Gras D, Kappenberger L, Tavazzi L: The Effect of Cardiac Resynchronization on Morbidity and Mortality in Heart Failure. N Engl J Med 2005; 352:1539–1549.

40. Gold MR, Daubert C, Abraham WT, Ghio S, St. John Sutton M, Hudnall JH, Cerkvenik J, Linde C: The effect of reverse remodeling on long-term survival in mildly symptomatic patients with heart failure receiving cardiac resynchronization therapy: Results of the REVERSE study. Heart Rhythm 2015; 12:524–530.

41. Auricchio A, Prinzen FW: Non-Responders to Cardiac Resynchronization Therapy. Circ J 2011; 75:521–527.

42. Gorcsan III J, Prinzen FW: Understanding the cardiac substrate and the underlying physiology: Implications for individualized treatment algorithm. Heart Rhythm 2012; 9:S18–S26.

43. Suffoletto MS, Dohi K, Cannesson M, Saba S, Gorcsan J: Novel Speckle-Tracking Radial Strain From Routine Black-and-White Echocardiographic Images to Quantify Dyssynchrony and Predict Response to Cardiac Resynchronization Therapy. Circulation 2006; 113:960–968.

44. Klimusina J, De Boeck BW, Leenders GEH, Faletra FF, Prinzen F, Averaimo M, Pasotti E, Klersy C, Moccetti T, Auricchio A: Redistribution of left ventricular strain by cardiac resynchronization therapy in heart failure patients. Eur J Heart Fail 2011; 13:186–194.

45. Ghani A, Delnoy PPH m., Adiyaman A, Ottervanger JP, Ramdat Misier AR, Smit JJJ, Elvan A: Response to Cardiac Resynchronization Therapy as Assessed by Time-Based Speckle Tracking Imaging. Pacing Clin Electrophysiol 2015; 38:455–464.

46. Auricchio A, Fantoni C, Regoli F, Carbucicchio C, Goette A, Geller C, Kloss M, Klein H: Characterization of Left Ventricular Activation in Patients With Heart Failure and Left Bundle-Branch Block. Circulation 2004; 109:1133–1139.

47. Gold MR, Birgersdotter-Green U, Singh JP, Ellenbogen KA, Yu Y, Meyer TE, Seth M, Tchou PJ: The relationship between ventricular electrical delay and left ventricular remodelling with cardiac resynchronization therapy. Eur Heart J 2011; 32:2516–2524.

114

48. Singh JP, Fan D, Heist EK, Alabiad CR, Taub C, Reddy V, Mansour M, Picard MH, Ruskin JN, Mela T: Left ventricular lead electrical delay predicts response to cardiac resynchronization therapy. Heart Rhythm 2006; 3:1285–1292.

49. Bilchick KC, Kuruvilla S, Hamirani YS, et al.: Impact of Mechanical Activation, Scar, and Electrical Timing on Cardiac Resynchronization Therapy Response and Clinical Outcomes. J Am Coll Cardiol 2014; 63:1657–1666.

50. Bordachar P, Lafitte S, Reuter S, Sanders P, Jaïs P, Haïssaguerre M, Roudaut R, Garrigue S, Clementy J: Echocardiographic parameters of ventricular dyssynchrony validation in patients with heart failure using sequential biventricular pacing. J Am Coll Cardiol 2004; 44:2157–2165.

51. Chan W y. W, Blomqvist A, Melton IC, Norén K, Crozier IG, Benser ME, Eigler NL, Gutfinger D, Troughton RW: Effects of AV Delay and VV Delay on Left Atrial Pressure and Waveform in Ambulant Heart Failure Patients: Insights into CRT Optimization. Pacing Clin Electrophysiol 2014; 37:810–819.

52. Pappone C, Ćalović Ž, Vicedomini G, et al.: Improving cardiac resynchronization therapy response with multipoint left ventricular pacing: Twelve-month follow-up study. Heart Rhythm 2015; 12:1250–1258.

53. Lele SS, Thomson HL, Seo H, Belenkie I, McKenna WJ, Frenneaux MP: Exercise Capacity in Hypertrophic Cardiomyopathy. Circulation 1995; 92:2886–2894.

54. Quinones MA, Waggoner AD, Reduto LA, Nelson JG, Young JB, Winters WL, Ribeiro LG, Miller RR: A new, simplified and accurate method for determining ejection fraction with two-dimensional echocardiography. Circulation 1981; 64:744–753.

55. Miyatake K, Okamoto M, Kinoshita N, Izumi S, Owa M, Takao S, Sakakibara H, Nimura Y: Clinical applications of a new type of real-time two-dimensional doppler flow imaging system. Am J Cardiol 1984; 54:857–868.

56. D’hooge J, Heimdal A, Jamal F, Kukulski T, Bijnens B, Rademakers F, Hatle L, Suetens P, Sutherland GR: Regional Strain and Strain Rate Measurements by Cardiac Ultrasound: Principles, Implementation and Limitations. Eur J Echocardiogr 2000; 1:154–170.

57. Gilman G, Khandheria BK, Hagen ME, Abraham TP, Seward JB, Belohlavek M: Strain rate and strain: A step-by-step approach to image and data acquisition. J Am Soc Echocardiogr 2004; 17:1011–1020.

58. Sutherland GR, Di Salvo G, Claus P, D’hooge J, Bijnens B: Strain and strain rate imaging: a new clinical approach to quantifying regional myocardial function. J Am Soc Echocardiogr 2004; 17:788–802.

59. Hein IA, O’Brien WD: Current time-domain methods for assessing tissue motion by analysis from reflected ultrasound echoes-a review. IEEE Trans Ultrason Ferroelectr Freq Control 1993; 40:84–102.

115

60. Heimdal A, Støylen A, Torp H, Skjærpe T: Real-Time Strain Rate Imaging of the Left Ventricle by Ultrasound. J Am Soc Echocardiogr 1998; 11:1013–1019.

61. Castro PL, Greenberg NL, Drinko J, Garcia MJ, Thomas JD: Potential pitfalls of strain rate imaging: angle dependency. Biomed Sci Instrum 2000; 36:197–202.

62. Nightingale K, Soo MS, Nightingale R, Trahey G: Acoustic radiation force impulse imaging: in vivo demonstration of clinical feasibility. Ultrasound Med Biol 2002; 28:227– 235.

63. Sebag F, Vaillant-Lombard J, Berbis J, Griset V, Henry JF, Petit P, Oliver C: Shear Wave Elastography: A New Ultrasound Imaging Mode for the Differential Diagnosis of Benign and Malignant Thyroid Nodules. J Clin Endocrinol Metab 2010; 95:5281–5288.

64. Rahmim A, Zaidi H: PET versus SPECT: strengths, limitations and challenges: Nucl Med Commun 2008; 29:193–207.

65. Di Carli MF, Davidson M, Little R, et al.: Value of metabolic imaging with positron emission tomography for evaluating prognosis in patients with coronary artery disease and left ventricular dysfunction. Am J Cardiol 1994; 73:527–533.

66. Albert TSE, Kim RJ, Judd RM: Assessment of no-reflow regions using cardiac MRI. Basic Res Cardiol 2006; 101:383–390.

67. Knez A, Becker CR, Leber A, Ohnesorge B, Becker A, White C, Haberl R, Reiser MF, Steinbeck G: Usefulness of multislice spiral computed tomography angiography for determination of coronary artery stenoses. Am J Cardiol 2001; 88:1191–1194.

68. Gaemperli O, Schepis T, Valenta I, Husmann L, Scheffel H, Duerst V, Eberli FR, Luscher TF, Alkadhi H, Kaufmann PA: Cardiac Image Fusion from Stand-Alone SPECT and CT: Clinical Experience. J Nucl Med 2007; 48:696–703.

69. Pichler BJ, Kolb A, Nägele T, Schlemmer H-P: PET/MRI: Paving the Way for the Next Generation of Clinical Multimodality Imaging Applications. J Nucl Med 2010; 51:333– 336.

70. Hendel RC, Patel MR, Kramer CM, et al.: ACCF/ACR/SCCT/SCMR/ASNC/NASCI/SCAI/SIR 2006 Appropriateness Criteria for Cardiac Computed Tomography and Cardiac Magnetic Resonance Imaging⁎A Report of the American College of Cardiology Foundation Quality Strategic Directions Committee Appropriateness Criteria Working Group, American College of Radiology, Society of Cardiovascular Computed Tomography, Society for Cardiovascular Magnetic Resonance, American Society of Nuclear Cardiology, North American Society for Cardiac Imaging, Society for Cardiovascular Angiography and Interventions, and Society of Interventional Radiology. J Am Coll Cardiol 2006; 48:1475–1497.

71. Hendel RC, Berman DS, Di Carli MF, Heidenreich PA, Henkin RE, Pellikka PA, Pohost GM, Williams KA: ACCF/ASNC/ACR/AHA/ASE/SCCT/SCMR/SNM 2009 Appropriate

116

Use Criteria for Cardiac Radionuclide ImagingA Report of the American College of Cardiology Foundation Appropriate Use Criteria Task Force, the American Society of Nuclear Cardiology, the American College of Radiology, the American Heart Association, the American Society of Echocardiography, the Society of Cardiovascular Computed Tomography, the Society for Cardiovascular Magnetic Resonance, and the Society of Nuclear Medicine Endorsed by the American College of Emergency Physicians. J Am Coll Cardiol 2009; 53:2201–2229.

72. Slavin GS, Bluemke DA: Spatial and Temporal Resolution in Cardiovascular MR Imaging: Review and Recommendations. Radiology 2005; 234:330–338.

73. Konofagou EE, D’hooge J, Ophir J: Myocardial elastography—a feasibility study in vivo. Ultrasound Med Biol 2002; 28:475–482.

74. Artis NJ, Oxborough DL, Williams G, Pepper CB, Tan LB: Two-dimensional strain imaging: A new echocardiographic advance with research and clinical applications. Int J Cardiol 2008; 123:240–248.

75. Perk G, Tunick PA, Kronzon I: Non-Doppler Two-dimensional Strain Imaging by Echocardiography–From Technical Considerations to Clinical Applications. J Am Soc Echocardiogr 2007; 20:234–243.

76. Garra BS, Cespedes EI, Ophir J, Spratt SR, Zuurbier RA, Magnant CM, Pennanen MF: Elastography of breast lesions: initial clinical results. Radiology 1997; 202:79–86.

77. Lee W-N, Ingrassia CM, Fung-Kee-Fung SD, Costa KD, Holmes JW, Konofagou EE: Theoretical Quality Assessment of Myocardial Elastography with In Vivo Validation. IEEE Trans Ultrason Ferroelectr Freq Control 2007; 54:2233–2245.

78. Lee W-N, Parker KM, Luo J, Holmes JW, Konofagou EE: Frame rate dependence of myocardial elastography estimates using a physiologic 3D biventricular finite-element model of the heart with preliminary in vivo validation. IEEE Int Ultrason Symp Rome, Italy, 2009,.

79. Luo J, Konofagou EE: High-frame rate, full-view myocardial elastography with automated contour tracking in murine left ventricles in vivo. IEEE Trans Ultrason Ferroelectr Freq Control 2008; 55:240–248.

80. Lee W-N, Konofagou EE: Angle-independent and multi-dimensional myocardial elastography--from theory to clinical validation. Ultrasonics 2008; 48:563–567.

81. Lee W-N, Qian Z, Tosti CL, Brown TR, Metaxas DN, Konofagou EE: Preliminary Validation of Angle-Independent Myocardial Elastography Using MR Tagging in a Clinical Setting. Ultrasound Med Biol 2008; 34:1980–1997.

82. Lee W-N, Provost J, Fujikura K, Wang J, Konofagou EE: In vivo study of myocardial elastography under graded ischemia conditions. Phys Med Biol 2011; 56:1155.

117

83. Grondin J, Wan E, Gambhir A, Garan H, Konofagou E: Intracardiac myocardial elastography in canines and humans in vivo. IEEE Trans Ultrason Ferroelectr Freq Control 2015; 62:337–349.

84. Mucci R: A comparison of efficient beamforming algorithms. IEEE Trans Acoust Speech Signal Process 1984; 32:548–558.

85. Luo J, Konofagou EE: A fast normalized cross-correlation calculation method for motion estimation. IEEE Trans Ultrason Ferroelectr Freq Control 2010; 57:1347–1357.

86. Konofagou E, Ophir J: A new elastographic method for estimation and imaging of lateral displacements, lateral strains, corrected axial strains and poisson’s ratios in tissues. Ultrasound Med Biol 1998; 24:1183–1199.

87. Chen H, Shi H, Varghese T: Improvement of elastographic displacement estimation using a two-step cross-correlation method. Ultrasound Med Biol 2007; 33:48–56.

88. Céspedes I, Huang Y, Ophir J, Spratt S: Methods for Estimation of Subsample Time Delays of Digitized Echo Signals. Ultrason Imaging 1995; 17:142–171.

89. Fung JWH, Chan JYS, Yip GWK, Chan HCK, Chan WWL, Zhang Q, Yu C-M: Effect of left ventricular endocardial activation pattern on echocardiographic and clinical response to cardiac resynchronization therapy. Heart 2007; 93:432–437.

90. Gurev V, Constantino J, Rice JJ, Trayanova NA: Distribution of Electromechanical Delay in the Heart: Insights from a Three-Dimensional Electromechanical Model. Biophys J 2010; 99:745–754.

91. Costet A, Wan E, Bunting E, Grondin J, Garan H, Konofagou E: Electromechanical wave imaging (EWI) validation in all four cardiac chambers with 3D electroanatomic mapping in canines in vivo. Phys Med Biol 2016; 61:8105.

92. Grondin J, Costet A, Bunting E, Gambhir A, Garan H, Wan E, Konofagou EE: Validation of electromechanical wave imaging in a canine model during pacing and sinus rhythm. Heart Rhythm 2016; 13:2221–2227.

93. Provost J, Gurev V, Trayanova N, Konofagou EE: Mapping of cardiac electrical activation with electromechanical wave imaging: An in silico–in vivo reciprocity study. Heart Rhythm 2011; 8:752–759.

94. Provost J, Lee W-N, Fujikura K, Konofagou EE: Imaging the electromechanical activity of the heart in vivo. Proc Natl Acad Sci 2011; 108:8565–8570.

95. Costet A, Provost J, Gambhir A, Bobkov Y, Danilo Jr. P, Boink GJJ, Rosen MR, Konofagou EE: Electromechanical Wave Imaging of Biologically and Electrically Paced Canine Hearts in Vivo. Ultrasound Med Biol 2014; 40:177–187.

118

96. Provost J, Gambhir A, Vest J, Garan H, Konofagou EE: A clinical feasibility study of atrial and ventricular electromechanical wave imaging. Heart Rhythm 2013; 10:856–862.

97. Provost J, Lee W-N, Fujikura K, Konofagou EE: Electromechanical Wave Imaging of Normal and Ischemic Hearts In Vivo. IEEE Trans Med Imaging 2010; 29:625–635.

98. Costet A, Bunting E, Grondin J, Gambhir A, Konofagou EE: Atrial electromechanical cycle length mapping in paced canine hearts in vivo. IEEE Trans Ultrason Ferroelectr Freq Control 2015; 62:1277–1287.

99. Provost J, Costet A, Wan E, Gambhir A, Whang W, Garan H, Konofagou EE: Assessing the atrial electromechanical coupling during atrial focal tachycardia, flutter, and fibrillation using electromechanical wave imaging in humans. Comput Biol Med 2015; 65:161–167.

100. Provost J: Electromechanical Wave Imaging [Internet]. Columbia University, 2012 [cited 2015 Nov 4],. Available from: http://academiccommons.columbia.edu/catalog/ac:175194

101. Provost J, Nguyen VT-H, Legrand D, Okrasinski S, Costet A, Gambhir A, Garan H, Konofagou EE: Electromechanical wave imaging for arrhythmias. Phys Med Biol 2011; 56:L1–L11.

102. Ophir J, Céspedes I, Ponnekanti H, Yazdi Y, Li X: Elastography: A quantitative method for imaging the elasticity of biological tissues. Ultrason Imaging 1991; 13:111–134.

103. Lopata RGP, Nillesen MM, Hansen HHG, Gerrits IH, Thijssen JM, de Korte CL: Performance of Two Dimensional Displacement and Strain Estimation Techniques Using a Phased Array Transducer. Ultrasound Med Biol 2009; 35:2031–2041.

104. Yeung F, Levinson SF, Fu D, Parker KJ: Feature-adaptive motion tracking of ultrasound image sequences using a deformable mesh. IEEE Trans Med Imaging 1998; 17:945–956.

105. Ma C, Varghese T: Lagrangian displacement tracking using a polar grid between endocardial and epicardial contours for cardiac strain imaging. Med Phys 2012; 39:1779– 1792.

106. Varghese T, Ophir J: Characterization of Elastographic Noise Using the Envelope of Echo Signals. Ultrasound Med Biol 1998; 24:543–555.

107. Ma C, Varghese T: Comparison of cardiac displacement and strain imaging using ultrasound radiofrequency and envelope signals. Ultrasonics 2013; 53:782–792.

108. Tanter M, Bercoff J, Sandrin L, Fink M: Ultrafast compound imaging for 2-D motion vector estimation: application to transient elastography. IEEE Trans Ultrason Ferroelectr Freq Control 2002; 49:1363–1374.

109. Pihl MJ, Marcher J, Jensen JA: Phased-array vector velocity estimation using transverse oscillations. IEEE Trans Ultrason Ferroelectr Freq Control 2012; 59.

119

110. Kallel F, Ophir J: A Least-Squares Strain Estimator for Elastography. Ultrason Imaging 1997; 19:195–208.

111. Varghese T, Ophir J: A theoretical framework for performance characterization of elastography: the strain filter. IEEE Trans Ultrason Ferroelectr Freq Control 1997; 44:164–172.

112. Walker WF, Trahey GE: A fundamental limit on delay estimation using partially correlated speckle signals. IEEE Trans Ultrason Ferroelectr Freq Control 1995; 42:301– 308.

113. Yu W, Yan P, Sinusas AJ, Thiele K, Duncan JS: Towards pointwise motion tracking in echocardiographic image sequences – Comparing the reliability of different features for speckle tracking. Med Image Anal 2006; 10:495–508.

114. Viola F, Walker WF: A comparison of the performance of time-delay estimators in medical ultrasound. IEEE Trans Ultrason Ferroelectr Freq Control 2003; 50:392–401.

115. Chen H, Varghese T, Rahko PS, Zagzebski JA: Ultrasound frame rate requirements for cardiac elastography: Experimental and in vivo results. Ultrasonics 2009; 49:98–111.

116. Hasegawa H, Kanai H: High-frame-rate echocardiography using diverging transmit beams and parallel receive beamforming. J Med Ultrason 2011; 38:129–140.

117. Wang S, Lee W-N, Provost J, Luo J, Konofagou EE: A composite high-frame-rate system for clinical cardiovascular imaging. IEEE Trans Ultrason Ferroelectr Freq Control 2008; 55:2221–2233.

118. D’hooge J, Konofagou E, Jamal F, Heimdal A, Barrios L, Bijnens B, Thoen J, Van de Werf F, Sutherland G, Suetens P: Two-dimensional ultrasonic strain rate measurement of the human heart in vivo. IEEE Trans Ultrason Ferroelectr Freq Control 2002; 49:281–286.

119. Kanai H, Koiwa Y: Myocardial rapid velocity distribution. Ultrasound Med Biol 2001; 27:481–498.

120. Provost J, Thiébaut S, Luo J, Konofagou EE: Single-heartbeat electromechanical wave imaging with optimal strain estimation using temporally unequispaced acquisition sequences. Phys Med Biol 2012; 57:1095.

121. Amundsen BH, Helle-Valle T, Edvardsen T, Torp H, Crosby J, Lyseggen E, Støylen A, Ihlen H, Lima JAC, Smiseth OA, Slørdahl SA: Noninvasive Myocardial Strain Measurement by Speckle Tracking EchocardiographyValidation Against Sonomicrometry and Tagged Magnetic Resonance Imaging. J Am Coll Cardiol 2006; 47:789–793.

122. Walker WF, Trahey GE: A fundamental limit on the performance of correlation based phase correction and flow estimation techniques. IEEE Trans Ultrason Ferroelectr Freq Control 1994; 41:644–654.

120

123. Lee W-N: Myocardial elastography: a strain imaging technique for the reliable detection and localization of myocardial ischemia in vivo. [New York, NY, USA]: Columbia University, 2010,.

124. Luo J, Konofagou EE: Effects of Various Parameters on Lateral Displacement Estimation in Ultrasound Elastography. Ultrasound Med Biol 2009; 35:1352–1366.

125. Konofagou EE, Varghese T, Ophir J: Theoretical Bounds on the Estimation of Transverse Displacement, Transverse Strain and Poisson’s Ratio in Elastography. Ultrason Imaging 2000; 22:153–177.

126. Korukonda S, Doyley MM: Estimating Axial and Lateral Strain Using a Synthetic Aperture Elastographic Imaging System. Ultrasound Med Biol 2011; 37:1893–1908.

127. Park S, Aglyamov SR, Scott WG, Emelianov SY: Strain imaging using conventional and ultrafast ultrasound imaging: numerical analysis. IEEE Trans Ultrason Ferroelectr Freq Control 2007; 54:987–995.

128. Tong L, Gao H, Choi HF, D’hooge J: Comparison of conventional parallel beamforming with plane wave and diverging wave imaging for cardiac applications: a simulation study. IEEE Trans Ultrason Ferroelectr Freq Control 2012; 59:1654–1663.

129. Papadacci C, Pernot M, Couade M, Fink M, Tanter M: High-contrast ultrafast imaging of the heart. IEEE Trans Ultrason Ferroelectr Freq Control 2014; 61:288–301.

130. Burckhardt CB: Speckle in ultrasound B-mode scans. IEEE Trans Sonics Ultrason 1978; 25:1–6.

131. Montaldo G, Tanter M, Bercoff J, Benech N, Fink M: Coherent plane-wave compounding for very high frame rate ultrasonography and transient elastography. IEEE Trans Ultrason Ferroelectr Freq Control 2009; 56:489–506.

132. Rao M, Chen Q, Shi H, Varghese T: Spatial-angular compounding for elastography using beam steering on linear array transducers. Med Phys 2006; 33:618–626.

133. Hansen HHG, Stuart MB, Villagomez Hoyos CA, Jensen JA, de Korte CL: Performance evaluation of compounding and directional beamforming techniques for carotid strain imaging using plane wave transmissions. Ultrason Symp IUS 2014 IEEE Int 2014, pp. 1814–1817.

134. Yingchoncharoen T, Agarwal S, Popović ZB, Marwick TH: Normal Ranges of Left Ventricular Strain: A Meta-Analysis. J Am Soc Echocardiogr 2013; 26:185–191.

135. Bohs LN, Trahey GE: A novel method for angle independent ultrasonic imaging of blood flow and tissue motion. IEEE Trans Biomed Eng 1991; 38:280–286.

136. Kolandaivelu A, Zviman MM, Castro V, Lardo AC, Berger RD, Halperin HR: Noninvasive Assessment of Tissue Heating During Cardiac Radiofrequency Ablation

121

Using MRI ThermographyClinical Perspective. Circ Arrhythm Electrophysiol 2010; 3:521–529.

137. Dickfeld T, Kato R, Zviman M, Lai S, Meininger G, Lardo AC, Roguin A, Blumke D, Berger R, Calkins H, Halperin H: Characterization of Radiofrequency Ablation Lesions With Gadolinium-Enhanced Cardiovascular Magnetic Resonance Imaging. J Am Coll Cardiol 2006; 47:370–378.

138. Wright M, Harks E, Deladi S, Suijver F, Barley M, van Dusschoten A, Fokkenrood S, Zuo F, Sacher F, Hocini M, Haïssaguerre M, Jaïs P: Real-time lesion assessment using a novel combined ultrasound and radiofrequency ablation catheter. Heart Rhythm 2011; 8:304– 312.

139. Granier M, Winum PF, Granier M, Liaud P, Cayla G, Messner P, Pasquie J-L, Schuster I: Real-time atrial wall imaging during radiofrequency ablation in a porcine model. Heart Rhythm 2015; 12:1827–1835.

140. Kwiecinski W, Provost J, Dubois R, Sacher F, Haïssaguerre M, Legros M, Nguyen-Dinh A, Dufait R, Tanter M, Pernot M: Validation of an intracardiac ultrasonic therapy– imaging dual mode transducer. IRBM 2015; 36:351–354.

141. Kwiecinski W, Provost J, Dubois R, Sacher F, Haïssaguerre M, Legros M, Nguyen-Dinh A, Dufait R, Tanter M, Pernot M: Quantitative evaluation of atrial radio frequency ablation using intracardiac shear-wave elastography. Med Phys 2014; 41:112901.

142. Kwiecinski W, Bessière F, Colas EC, N’Djin WA, Tanter M, Cyril Lafon, Pernot M: Cardiac shear-wave elastography using a transesophageal transducer: application to the mapping of thermal lesions in ultrasound transesophageal cardiac ablation. Phys Med Biol 2015; 60:7829.

143. Hollender PJ, Wolf PD, Goswami R, Trahey GE: Intracardiac Echocardiography Measurement of Dynamic Myocardial Stiffness with Shear Wave Velocimetry. Ultrasound Med Biol 2012; 38:1271–1283.

144. Eyerly SA, Bahnson TD, Koontz JI, Bradway DP, Dumont DM, Trahey GE, Wolf PD: Intracardiac acoustic radiation force impulse imaging: A novel imaging method for intraprocedural evaluation of radiofrequency ablation lesions. Heart Rhythm 2012; 9:1855–1862.

145. Hsu SJ, Fahey BJ, Dumont DM, Wolf PD, Trahey GE: Challenges and implementation of radiation-force imaging with an intracardiac ultrasound transducer. IEEE Trans Ultrason Ferroelectr Freq Control 2007; 54:996–1009.

146. Yue Y, Jr JWC, Khoury$^*$ DS: Speckle Tracking in Intracardiac Echocardiography for the Assessment of Myocardial Deformation. IEEE Trans Biomed Eng 2009; 56:416–425.

122

147. Stephens DN, Cannata J, Liu R, et al.: Multifunctional catheters combining intracardiac ultrasound imaging and electrophysiology sensing. IEEE Trans Ultrason Ferroelectr Freq Control 2008; 55:1570–1581.

148. Wang J, Urheim S, Korinek J, Abraham TP, McMahon EM, Belohlavek M: Analysis of postsystolic myocardial thickening work in selective myocardial layers during progressive myocardial ischemia. J Am Soc Echocardiogr Off Publ Am Soc Echocardiogr 2006; 19:1102–1111.

149. Bunting EA, Provost J, Konofagou EE: Stochastic precision analysis of 2D cardiac strain estimation in vivo. Phys Med Biol 2014; 59:6841.

150. Bunting E, Papadacci C, Wan E, Grondin J, Konofagou E: Intracardiac myocardial elastography for lesion quantification in cardiac radiofrequency ablation. 2016 IEEE Int Ultrason Symp IUS 2016, pp. 1–4.

151. Markelj P, Tomaževič D, Likar B, Pernuš F: A review of 3D/2D registration methods for image-guided interventions. Med Image Anal 2012; 16:642–661.

152. Fishbein MC, Meerbaum S, Rit J, Lando U, Kanmatsuse K, Mercier JC, Corday E, Ganz W: Early phase acute myocardial infarct size quantification: Validation of the triphenyl tetrazolium chloride tissue enzyme staining technique. Am Heart J 1981; 101:593–600.

153. Papadacci C, Bunting EA, Wan EY, Nauleau P, Konofagou EE: 3D myocardial elastography in vivo. IEEE Trans Med Imaging 2016; PP:1–1.

154. Dausch D, Castellucci J, Gilchrist K, Carlson J, Hall S, von Ramm O: REAL-TIME 3D INTRACARDIAC ECHO OF THE MITRAL VALVE USING A NOVEL LIVE VOLUMETRIC IMAGING CATHETER. J Am Coll Cardiol [Internet] 2013 [cited 2016 Nov 30]; 61. Available from: http://dx.doi.org/10.1016/S0735-1097(13)61098-0

155. Wildes D, Lee W, Haider B, Cogan S, Sundaresan K, Mills DM, Yetter C, Hart PH, Haun CR, Concepcion M, Kirkhorn J, Bitoun M: 4-D ICE: A 2-D Array Transducer With Integrated ASIC in a 10-Fr Catheter for Real-Time 3-D Intracardiac Echocardiography. IEEE Trans Ultrason Ferroelectr Freq Control 2016; 63:2159–2173.

156. Luke RA, Saffitz JE: Remodeling of ventricular conduction pathways in healed canine infarct border zones. J Clin Invest 1991; 87:1594–1602.

157. Akar FG, Nass RD, Hahn S, Cingolani E, Shah M, Hesketh GG, DiSilvestre D, Tunin RS, Kass DA, Tomaselli GF: Dynamic changes in conduction velocity and gap junction properties during development of pacing-induced heart failure. Am J Physiol - Heart Circ Physiol 2007; 293:H1223–H1230.

158. Costet A: Electromechanical wave imaging for the in vivo characterization and assessment of cardiac arrhythmias. 2016 [cited 2017 Mar 3]; . Available from: https://academiccommons.columbia.edu/catalog/ac:203116

123

159. Heidenreich PA, Albert NM, Allen LA, et al.: Forecasting the Impact of Heart Failure in the United States A Policy Statement From the American Heart Association. Circ Heart Fail 2013; 6:606–619.

160. León AR, Abraham WT, Brozena S, Daubert JP, Fisher WG, Gurley JC, Liang CS, Wong G: Cardiac Resynchronization With Sequential Biventricular Pacing for the Treatment of Moderate-to-Severe Heart Failure. J Am Coll Cardiol 2005; 46:2298–2304.

161. Chalil S, Stegemann B, Muhyaldeen S, Khadjooi K, Smith REA, Jordan PJ, Leyva F: Intraventricular Dyssynchrony Predicts Mortality and Morbidity After Cardiac Resynchronization TherapyA Study Using Cardiovascular Magnetic Resonance Tissue Synchronization Imaging. J Am Coll Cardiol 2007; 50:243–252.

162. Pernot M, Fujikura K, Fung-Kee-Fung SD, Konofagou EE: ECG-gated, Mechanical and Electromechanical Wave Imaging of Cardiovascular Tissues In Vivo. Ultrasound Med Biol 2007; 33:1075–1085.

163. Tracy CM, Epstein AE, Darbar D, et al.: 2012 ACCF/AHA/HRS Focused Update of the 2008 Guidelines for Device-Based Therapy of Cardiac Rhythm Abnormalities A Report of the American College of Cardiology Foundation/American Heart Association Task Force on Practice Guidelines and the Heart Rhythm Society. Circulation 2012; 126:1784–1800.

164. Jia P, Ramanathan C, Ghanem RN, Ryu K, Varma N, Rudy Y: Electrocardiographic imaging of cardiac resynchronization therapy in heart failure: Observation of variable electrophysiologic responses. Heart Rhythm 2006; 3:296–310.

165. Peters NS, Wit AL: Myocardial Architecture and Ventricular Arrhythmogenesis. Circulation 1998; 97:1746–1754.

166. Ploux S, Lumens J, Whinnett Z, et al.: Noninvasive Electrocardiographic Mapping to Improve Patient Selection for Cardiac Resynchronization Therapy: Beyond QRS Duration and Left Bundle Branch Block Morphology. J Am Coll Cardiol 2013; 61:2435–2443.

167. Mollema SA, Bleeker GB, van der Wall EE, Schalij MJ, Bax JJ: Usefulness of QRS Duration to Predict Response to Cardiac Resynchronization Therapy in Patients With End- Stage Heart Failure. Am J Cardiol 2007; 100:1665–1670.

124

List of Publications Related to this Thesis

1. Bunting E., Provost J., and Konofagou E. “Stochastic precision analysis of 2D cardiac strain estimation in vivo,” Physics in Medicine and Biology, vol. 59, no. 22, pp 6841- 6858, November 2014. 2. Hou Y., Provost J., Grondin J., Wang S., Marquet F., Bunting E., and Konofagou E. “Sparse matrix beamforming and image reconstruction for 2D HIFU monitoring using Harmonic Motion Imaging for Focused Ultrasound (HMIFU) with in vitro validation,” IEEE Transactions on Medical Imaging, vol. 33, no. 11, pp 2107-2117, November 2014. 3. Costet A., Bunting E., Grondin J., Gambhir A., and Konofagou E. “Atrial Electromechanical Cycle Length Mapping in Paced Canine Hearts In Vivo,” IEEE UFFC, vol. 62, issue 7, pp 1277-1288, July 2015. 4. Bunting E., Papadacci C., Wan E., Grondin J., and Konofagou E. “Intracardiac myocardial elastography for lesion quantification in cardiac radiofrequency ablation.” 2016 IEEE IUS Proceedings, September 2016. 5. Papadacci C., Bunting E., and Konofagou E.E. “3D quasi-static ultrasound elastography with plane wave in vivo, IEEE Transactions on Medical Imaging, vol. 35, issue 9, September 2016. 6. Grondin J., Costet A., Bunting E., Gambhir A., Garan H., Wan E., and Konofagou E. “Validation of Electromechanical Wave Imaging in a canine model during pacing and sinus rhythm,” Heart Rhythm Journal, vol. 13, issue 9, September 2016. 7. Costet A., Wan E., Bunting E., Grondin J., Garan H., and Konofagou E. “Electromechanical wave imaging (EWI) validations in all four cardiac chambers with 3D electroanatomic mapping in canines in vivo. Physics in Medicine and Biology, vol. 61, pp 8105-8119, October 2016 8. Bunting E., Lambrakos L., Kemper P., Whang W., and Konofagou E. “Imaging the propagation of the electromechanical wave in heart failure patients with cardiac resynchronization therapy.” Pacing and Clinical Electrophysiology, vol. 39, issue 11, November 2016. 9. Papadacci C., Bunting E., Wan E., Nauleau P., and Konofagou E. “3D myocardial elastography in vivo, IEEE Transactions on Medical Imaging, issue 99, November 2016. 10. Bunting E., Papadacci C., Grondin J., Sayseng V., and Konofagou E. “Improvements in Two-Dimensional Cardiac Strain Estimation using Coherent Spatial Compounding.” In Review 11. Bunting E., Papadacci C., Wan E., Grondin J., and Konofagou E. “Cardiac Lesion Mapping in vivo using Intracardiac Myocardial Elastography.” In Review

125