Researchers from New York City-based organizations Columbia University and NewYork-Presbyterian have developed an AI tool capable of identifying patients at risk of having undiagnosed structural heart disease.
The tool, called EchoNext, was trained on more than 1.2 million electrocardiogram and echocardiogram data pairs from 230,000 patients, according to a July 16 news release from NewYork-Presbyterian. Upon validation, the tool accurately identified 77% of structural heart problems in 3,200 electrocardiograms compared to 64% accuracy of 13 cardiologists analyzing the same data.
One of the developers of EchoNext is Pierre Elias, MD, medical director for artificial intelligence at NewYork-Presbyterian and assistant professor of medicine and biomedical informatics at Columbia University.
Dr. Elias and his team published their EchoNext research July 16 in Nature. They are currently testing EchoNext in a clinical trial across eight emergency departments. He shared more about the tool with Becker’s, and expounded on how foundational science can break through the “smoke and mirrors” around AI.
Editor’s note: Responses have been lightly edited for clarity and length.
Question: What inspired you and your team to focus on using AI to enhance the diagnostic capabilities of electrocardiograms, specifically for detecting structural heart disease?
Dr. Pierre Elias: This started for me years ago after a patient was transferred to our care from another emergency department. He turned out to have severe valvular heart disease and unfortunately passed away. I’m convinced that if we just had a system where we could have known about this patient sooner, that patient could have gotten an outpatient procedure and be alive today. Since then, our goal has been to develop, validate and deploy AI technologies to take better care of patients, with a focus on how to take the best care of patients who we don’t know about today. That has blossomed into a screening program for cardiovascular disease that exists across our eight hospitals and 190 clinical sites.
Q: How do you envision the integration of AI tools like EchoNext into day-to-day clinical workflows?
PE: We hope that there will be a new generation of AI-augmented biomarkers that can help clinicians make decisions, much in the same way we use current laboratory results when someone comes in with symptoms like shortness of breath to determine the most appropriate next steps.
Q: What excites you most about the clinical trial underway across emergency departments?
PE: The opportunity to measure the impact of patients helped with these technologies. There’s a lot of smoke and mirrors with AI these days, so we believe foundational science that measures the positive impact of new technologies is going to be important in separating fact from fiction.
Q: What would success look like for that next phase of EchoNext’s development?
PE: The technology receiving FDA approval and becoming a widely available technology for clinicians to use.
Leave a Reply