Artificial Intelligence Imaging Biomarkers of Longitudinal Cardiovascular Stress
ABSTRACT Across every type of prediction model, applied to virtually any population, the single most substantial contributor to incident cardiovascular disease (CVD) is older age. However, our understanding of why cardiovascular risk dramatically rises with advancing age remains limited. Evidence to date suggests that CVD culminates from age- related changes in cardiac structure and function that begin very early in adulthood and continue to develop on a background of lifelong exposure to risk factors – contributing to an ‘age gap’ between biological age and chronological age that grows over time. It logically follows that a person’s increasing burden of cardiovascular risk should manifest as progressive cardiac abnormalities that can be tracked, analyzed, and identified over the lifespan. Until recently, methods for assessing the cardiovascular ‘age gap’ have been limited. Recent advances in artificial intelligence (AI), when applied to biomedical images, indicate that deep learning algorithms can both offer precise measurements beyond human fidelity and also identify subtle traits in imaging that are unrecognized by the human eye. In fact, our prior work applying AI to echocardiography as well as others applying AI to other forms of medical imaging, have shown deep learning can not only reproduce standard measures of cardiac structural and function but is also identify CVD risk features including chronologic age, biological sex, diabetes, hypertension, and smoking. In effect, AI applied to echocardiography now offers the potential to capture an aggregate measure of cardiac aging and, in turn, identify intervenable target for mitigating age-related CVD risk. Thus, we hypothesize that AI methods applied to echocardiography can be used to not only predict biological cardiovascular age but also (i) identify trajectories of accelerated versus delayed cardiovascular aging, (ii) discern key contributors to accelerated cardiovascular aging phenotypes, and (iii) elucidate the excess age- related risk for specific CVD outcomes that may be independent from any co-existing burden of risk factors or subclinical cardiac disease. To test these hypotheses, we will leverage echocardiography images that are acquired as a part of routine clinical care in addition to serial imaging frequently obtained in large cohort studies of aging adults. Because healthcare cohorts are enriched for patients with accelerated aging trajectories, while epidemiologic cohorts are enriched for individuals with delayed aging trajectories, we plan to analyze imaging and outcomes data collected from both types of cohorts. Accordingly, our aims are to train and validate AI models on predicting biological age from cardiac imaging and identify its component contributors to excess cardiovascular risk in both the healthcare setting and in the community setting.
Investigator: Ouyang, David
Funder: National Heart Lung and Blood Institute
Deep Learning Assessment of the Right Ventricle: Function, Etiology, and Prognosis
ABSTRACT Heart failure imposes a tremendous burden of morbidity and mortality, costing the United States in excess of $31 billion annually. An increasingly recognized major determinant of outcomes in heart failure is right ventricular (RV) dysfunction. However, the nature and character of RV contribution to cardiovascular outcomes remains poorly understood, largely due to the imprecision of imaging and interpretation of RV morphology and function. Echocardiography, with its high temporal resolution and low cost of acquisition, serves as frontline cardiovascular imaging and a mainstay in approaches to assessing RV morphology and function. However, echocardiographic imaging of the RV is limited by factors that include technical variation in image acquisition and heterogeneity in image assessment as well as overall interpretation. We postulate that deep learning based phenotyping can offer the ability to not only more precisely characterize RV function but also classify RV imaging phenotypes according to etiologic disease states and, even further, refine prognostic evaluations of future cardiovascular risk. Therefore, in Aim 1, we will use video-based deep learning segmentation models to assess RV function, evaluate its cross-sectional relation with a range of expert-measured parameters, and examine its variation in the context of patient characteristics derived from large hospital-based cohorts. In Aim 2, we will use video-based deep learning models to produce imaging-based classification of RV disease and assess the ability of unsupervised approaches to classify RV dysfunction into various categories of disease etiology. In Aim 3, we will use models developed in part from training in Aims 1 and 2 to predict major cardiovascular outcomes including heart failure in addition to coronary artery disease, stroke, and cardiovascular death in both hospital- based and community-based cohorts. The overarching goal of this proposal is to improve the precision and standardization of RV phenotyping and determine the extent to which deep learning models can augment human assessment of the RV. This research will be accomplished in the setting of a comprehensive career development program designed to provide the candidate with the skills needed to become an independent physician-scientist in cardiovascular medicine and translational imaging science. An advisory committee of established scientists/mentors in the fields of cardiac imaging, deep learning, data science, and translational science will guide the candidate in his transition to scientific independence over the course of the award period.
Investigator: Ouyang, David
Funder: National Heart Lung and Blood Institute