Deep learning enhanced detection and personalized monitoring of aortic stenosis - The DETECT-AS Study

PROJECT SUMMARY This is a new application by an early-stage investigator with a long-term career objective of transforming cardiovascular care using artificial intelligence and data science. The proposal focuses on aortic stenosis (AS), a progressive narrowing of the aortic valve, which manifests in older adults and causes significant disability and

premature mortality despite minimally invasive treatment strategies. AS is either diagnosed following symptom- driven diagnostic testing or incidentally discovered, which has simultaneously led to a vast underdiagnosis of advanced stages of AS while identifying many with early-stage aortic valve disease without clarity on appropriate

follow-up. There is a critical need for novel screening and prognostication strategies for AS. We show that artificial intelligence (AI) models applied to 1-lead electrocardiograms (AI-ECGs) can be a sensitive and convenient screen for advanced (moderate/severe) AS. AI-ECG can be paired with a second, more specific, AI-enhanced

handheld cardiac point-of-care ultrasound (POCUS). This AI-POCUS automates the diagnosis of advanced AS without specialized imaging or expert evaluation. In Aim 1, we propose a multicenter pragmatic RCT evaluating this 2-stage, AI-driven screening strategy for advanced AS. This innovative, technology-driven screening strategy

will define a new paradigm for the efficient identification of advanced AS. In addition, we evaluate a novel strategy to bridge the critical gap in precision follow-up, especially for early-stage aortic valve disease. Early aortic valve disease – aortic sclerosis or mild AS – affects nearly a fourth of older adults over 65-years. However, there are

no guideline recommendations on follow-up for aortic sclerosis, and recommendations for mild AS do not account for the substantial heterogeneity in disease progression. In our preliminary investigations from a multicenter observational cohort study, we show that a deep learning tool for echocardiographic videos – deep learning-

based aortic stenosis severity index (DASSi) – can define those at substantially elevated risk of progression to advanced AS and adverse clinical outcomes. In Aim 2, we will conduct a multicenter, prospective evaluation of an individualized AS progression score among older adults with aortic sclerosis or mild AS through a protocolized

Doppler echocardiogram to distinguish those with high and low rates of progression. The investigations in Aim 2 will establish the reliability of a digital biomarker for AS progression that can enable precision care and follow- up. The work is supported by the team’s broad expertise in (a) clinical medicine, including cardiology, geriatrics,

and imaging; (b) technology, spanning informatics, data science, and AI; and (c) clinical trials, with experience in designing and executing studies. The evidence generated from a multicenter evaluation of low-cost AI-driven interventions can be immediately adopted and scaled to have a major public health impact. Moreover, an

objective approach to the diagnosis and follow-up of AS will reduce healthcare disparities for vulnerable patients. Future work will build on these results and engage directly with communities using low-cost portable devices to improve disease detection and outcomes among those without adequate healthcare access.