Yale investigators from the section of cardiovascular medicine have identified a new artificial intelligence (AI)-based video biomarker that is able to identify those who might develop and have rapidly worsening aortic stenosis. This research was highlighted in a new paper, “A Multimodality Video-Based AI Biomarker For Aortic Stenosis Development And Progression,” which was published in JAMA Cardiology on April 6, 2024.
More than 1.5 million people in the United States have aortic stenosis, which is a narrowing of the valve that connects the main chamber of the heart to the rest of the body, resulting in stress on the heart to push the blood to the body. As aortic stenosis progresses, it can lead to increased pressure in the heart and reduced oxygen supply to the body, causing chest pain, difficulty breathing, lightheadedness, and reduced ability to do everyday activities. Ultimately, severe aortic stenosis can lead to heart failure and premature death.
While relatively new procedures like surgical or transcatheter aortic valve replacements can effectively modify the prognosis for aortic stenosis, there is an unmet need to develop more precise algorithms to better define how aortic stenosis progresses in each person.
“So far, we have not had a way to know who develops aortic stenosis or who gets worse. There are no accepted clinical biomarkers for the progression of aortic stenosis, and most research to date has been focused on fixing valves once they are diseased,” said Evangelos K. Oikonomou, MD, DPhil, clinical fellow (cardiovascular medicine), and first author of the study. “This is foundational research that we believe will facilitate further study of new treatments to address the progression of aortic stenosis and, eventually, help prevent bad outcomes.”
The researchers built a deep learning model to detect valve features that suggested stenosis of the aortic valve on videos of the heart captured using cardiac ultrasound. Researchers identified a signature, the Digital AS Severity index (DASSi), that predicts which patients will progress to a severe form of aortic stenosis.
“One of the most exciting aspects of this research is that DASSi can convert cardiac MRIs, point of care ECGs, and cardiac ultrasound all using the same model,” said senior author of the paper, Rohan Khera, MD, MS, assistant professor of medicine (cardiovascular medicine) and the director of the Cardiovascular Data Science (CarDS) Lab. “This is crucial because it shows that DASSi flags a distinct myocardial and valvular phenotypic signature and is not restricted to modality-specific features or limited to some selected populations.”
The features identified by their model are not aspects of the image that even expert readers can visualize looking at the videos, highlighting the role of AI in detecting hidden signatures of disease. Unlike prior methods, views needed to get this information do not require much expertise to obtain.
“I am incredibly proud of all of the work we do at Yale to not only provide the best and least-invasive care for patients with aortic stenosis and also to continue to learn more about how the disease progresses,” said Eric J. Velazquez, MD, Robert W. Berliner Professor of Medicine, chief of Yale Cardiovascular Medicine, and co-author of the study. “This research shows how creatively using AI can help us better understand a common heart disease and ultimately help us identify new approaches to stem the progression of the disease so that fewer patients develop severe aortic stenosis.”
The study’s findings also support the use of DASSi for opportunistic screening of aortic stenosis on ECGs.
“I believe that every time a clinician gets a view of the heart, it’s an opportunity to screen patients and diagnose structural heart disorders,” said Khera. “This research shows that it’s possible to diagnose aortic stenosis and prognosticate risk of aortic stenosis using cardiac ultrasounds and cardiac MRI. That’s potentially practice-changing.”
Khera, Oikonomou, and other colleagues in CarDS lab are planning to launch a multi-randomized clinical trial with two goals: confirming the role of AI for video-based diagnosis of aortic stenosis and identifying people with mild or moderate aortic stenosis who may progress to severe aortic stenosis.
“That’s our next frontier,” Khera said. “We don’t want to just stop at building tools. We want to see whether these tools and discoveries can change practice and help cardiologists better identify each patient’s potential risk factors.”
Oikonomou recently presented the study results at the 2024 American College of Cardiology’s Scientific Sessions.
Other Yale researchers on the study include Gregory Holste; Andreas Coppi, PhD; Robert L. McNamara MD, MHS; Norrisa Haynes, MD; Amit N. Vora, MD, MPH; Fan Li, PhD; Thomas Gill, MD; and Harlan M. Krumholz, MD, SM.