Theses Doctoral

Myocardial Elastography for the Diagnosis of Coronary Artery Disease and Coronary Microvascular Disease

El Harake, Jad

Heart disease remains the leading cause of death globally, and prevalence has nearly doubled over the past three decades. It is estimated that up to 90% of cardiovascular events are preventable, but early detection and treatment is crucial. In this dissertation, we report on the optimization of the ultrasound-based cardiac strain imaging technique known as Myocardial Elastography (ME), a method for the detection of the most common and most lethal forms of heart disease: Coronary Artery Disease (CAD) which affects the major coronary arteries, and Coronary Microvascular Disease (CMD) which affects smaller coronary vessels.

CAD has historically been the primary focus of clinical cardiac imaging, whereas CMD has been under-diagnosed due to a lack of awareness and challenges associated with imaging at the microvascular level. Ultrasound-based cardiac strain imaging has been shown capable of detecting functional changes due to CAD and may be effective in CMD detection, although the latter has not yet been sufficiently investigated. However, the diagnostic accuracy of strain imaging is reduced by noise from transcostal imaging, known as clutter, and by the limited lateral resolution of high framerate ultrasound. These factors preclude accurate strain imaging in up to 30% of patients. Myocardial elastography is a precise high framerate strain imaging technique that analyzes radiofrequency (RF) signals to quantify myocardial deformation. We hypothesize that ME can effectively image and diagnose the functional effects of CMD and CAD, and that novel beamforming and clutter-filtering techniques can improve ME imaging and strain estimation quality, thereby increasing diagnostic accuracy.

To improve disease detection, Stress ME (S-ME) was proposed as a method to compare strain measurements at rest to strain during induced cardiac stress. A novel strain difference (Δ𝜺) metric was presented and investigated in a canine model of induced acute ischemia, as well as in a human CAD patient study with validation by myocardial perfusion imaging. In the canine model, flow-limiting stenosis was induced by partial ligation in N=2 canines, and stenosis was found to significantly reduce Δ𝜺 in the affected myocardial regions. In the clinical study, radial and circumferential ME strain and radial Δ𝜺 was measured in N=49 myocardial segments from 8 patients suspected to have ischemia or infarction due to CAD. The median Δ?, radial strain, and circumferential strain magnitudes were lowest in infarcted regions and highest in regions with normal perfusion, while measurements in ischemic regions fell in between. ROC analysis of radial strain metrics revealed that Δ𝜺 had the highest AUC for detecting ischemia (AUC=0.788 p<0.01) and infarction (AUC=0.792, p<0.05), followed by radial strain during stress (ischemia AUC=0.774 p<0.05, infarct AUC=0.758 p<0.05) while the AUC was lowest when considering only the radial strain at rest (ischemia AUC=0.52 p>0.05, infarct AUC=0.58 p>0.05). The results thus indicate that S-ME may improve detection of mild CAD cases that are functionally asymptomatic at rest.

Despite these promising findings, accurate strain imaging remains hindered by clutter noise and poor image quality. Two complementary techniques were thus developed to improve image quality and strain estimation in high frame rate cardiac strain imaging; a novel Sliding Window implementation of the Minimum Variance beamformer (SWMV) was proposed to enhance speckle quality, while a spatiotemporal singular value decomposition filter (SVD) was developed to increase tissue visibility and contrast by suppressing static clutter signals using automated cutoff selection. SWMV and SVD were shown to effectively improve image quality in simulation studies and phantom imaging experiments. In vivo performance evaluation consisted of applying SWMV beamforming and SVD filtering techniques to a dataset of N=70 strain images from 13 patients suspected to have CAD. CCTA imaging was used for validation of strain estimation. Tracking was improved in 92% of cases with a median improvement of 15% in displacement estimation accuracy as evaluated by an intersection-over-union (IoU) metric. The proposed techniques also improve agreement with CCTA results; ROC analysis shows improved AUC with SWMV+SVD compared to DAS when comparing healthy regions to those with any degree of stenosis (AUC 0.64 vs 0.56) as well as when comparing healthy to severely stenosed regions (AUC 0.69 vs 0.60). The observed results point to significant improvement in strain estimation reliability due to SWMV beamforming combined with SVD processing.

The final aim and the overarching goal of this work is a culmination of the previous sections for a clinical evaluation of ME as a diagnostic tool for CAD and CMD. In this clinical study, the enhanced ME technique utilizing SWMV and SVD filtering was applied to a cohort of N=201 patients with suspected coronary disease. All patients underwent invasive angiography or noninvasive cardiac imaging in the form of coronary computed tomography or nuclear stress testing. In addition, demographic information and patient clinical history were collected and accounted for in a multivariate statistical analysis. A K-nearest-neighbor (KNN) classifier was trained to distinguish between healthy and stenosed myocardial regions, and achieved an AUC of 0.91, with sensitivity of 86% and a specificity of 85% after training with 10-fold cross validation. CMD was also shown to significantly reduce regional strain measurements. This retrospective study identified the clinical factors which impact strain, and assessed the potential advantages of incorporating ME imaging to the existing clinical imaging pipeline for CAD and CMD diagnosis.

Files

  • thumnail for ElHarake_columbia_0054D_18345.pdf ElHarake_columbia_0054D_18345.pdf application/pdf 5.27 MB Download File

More About This Work

Academic Units
Biomedical Engineering
Thesis Advisors
Konofagou, Elisa E.
Degree
Ph.D., Columbia University
Published Here
April 17, 2024