Ischemic heart disease remains the top cause of death in the world. Acute myocardial infarction (MI) causes regional dysfunction which places remote areas of the heart at a mechanical disadvantage resulting in long term adverse left ventricular (LV) remodeling and complicating congestive heart failure (CHF). The course of MI and post-MI remodeling is complex and includes vascular and myocellular injury, acute and chronic inflammation, alterations of the extracellular matrix (ECM) and angiogenesis. Stress echocardiography is a clinically established, cost-effective technique for detecting and characterizing coronary artery disease and myocardial injury by imaging the LV at rest and after either exercise or pharmacologically-induced stress to reveal ischemia and/or scar. In our previous effort on this project, we developed quantitative 3D differential deformation measures for stress echocardiography from 4DE-derived LV strain maps taken at rest and after dobutamine stress. These measures can localize and quantify the extent and severity of LV myocardial injury and reveal ischemic regions. We now propose that improved versions of these same measures can be used for both targeting of therapy and outcomes assessment in the treatment of adverse local myocardial remodeling following MI. We choose a particular up and coming therapeutic strategy as an exemplar: the local delivery of injectable hydrogels within the MI region that are intended to alter the biomechanical properties of the LV myocardium, as well as inflammation, and thereby help to minimize adverse remodeling. Our new, robust approach for estimating improved dense displacement and differential deformation measures is based on an innovative data-driven, deep feed-forward, neural network architecture that employs domain adaptation between data from labeled, carefully-constructed synthetic models of physiology and echocardiographic image formation (i.e. with ground truth), and data from unlabeled noisy in vivo porcine or human echocardiography (missing or very limited ground truth). Training is based on tens of thousands of four-dimensional (4D) image-derived patches from these two domains, initially based on displacements derived separately from shape-based processing of conventional B-mode data and block-mode, speckle-tracked processing of raw radio-frequency (RF) data; and later based on learning directly from B-mode and RF image intensity information. After non-rigid registration of rest and stress 4DE image sequences, quantitative 4D differential deformation parameters will be derived from porcine and human echocardiographic test data. These parameters will be derived at baseline, and at several timepoints after delivery of injectable hydrogels into the MI region. The ability of the differential deformation parameters derived from 4D stress echocardiography to guide local delivery of injectable hydrogels in a MI region and assess/predict outcomes will then be determined in a hybrid acute/chronic porcine model of MI and post-MI remodeling. The technique will be translated to humans and evaluated by measuring the reproducibility and the relationship to remodeling of our new robust, deep learning-based differential deformation parameters in a small cohort of subjects.
At the core of the proposed effort is the development and evaluation of novel 4D (three spatial dimensions over time) echocardiographic imaging, image analysis, and machine learning methods that will enable the accurate and robust quantification of changes in myocardial deformation due to stress. Our methods will use this information to guide delivery and assess outcome of a promising new therapy to improve the biomechanical properties of the heart after myocardial injury based on injectable hydrogels.
