Can AI Help Cardiologists Diagnose Structural Heart Disease?

From developing business plans to writing wedding toasts, artificial intelligence is everywhere. Health care is increasingly embracing AI to aid research, support busy staff, and improve the lives of patients.

Cardiology is one of many areas turning to AI, and at Columbia, cardiologist Pierre Elias, MD, is exploring its potential to detect heart disease that may otherwise be missed.

Dr. Elias and his team have designed a tool called EchoNext that can use a patient’s electrocardiogram, which is a cheap and easy test, to detect structural heart disease.

Dr. Elias explains how this tool works and the impact it could have on health care.

Before we talk about your AI tool, what is an electrocardiogram?

The electrocardiogram (ECG) has been cardiologists’ go-to technology for decades, telling us quite a lot about the heart’s health and function. By attaching a few sticky electrodes to the chest, arms, and legs, cardiologists use ECGs to detect conditions such as abnormal heart rhythms or evidence of heart attacks.

That said, the ECG cannot tell cardiologists much about structural heart disease, which includes conditions that affect your heart’s valves, walls, chambers, or muscles and make it difficult for your heart to pump blood.

Was it the limitations of the ECG that motivated you to apply machine learning to the underlying data?

Yes. I came to medicine with a data science background, and my North Star for a long time has been that we all deserve better technology in health care.

How does EchoNext work?

In general, AI models take data and run mathematical operations to find unique patterns that a human may miss. With EchoNext, we input data from a patient’s ECG so it can be analyzed to identify patients who should be evaluated for structural heart disease. In this way, EchoNext ‘sees’ something a cardiologist can’t.

It turns out that when cardiologists use EchoNext, they are about twice as likely to identify structural heart disease as they would have been with routine clinical care. That’s a meaningful impact. The big picture: Our findings suggest that cardiology as a field may be missing up to half of all new structural heart disease diagnoses among patients.

Are there any individual cases from your study that you’d like to highlight?

Yes, I can share three EchoNext firsts, meaning these were the first instances of a structural heart disease diagnosis resulting from applying AI to ECGs.

In one patient, EchoNext detected severe aortic stenosis (a condition restricting blood flow out of the heart), which led to a TAVR procedure. In another patient, EchoNext detected severe mitral regurgitation (a condition where blood leaks backward between chambers of the heart), which led to mitral valve annuloplasty. And in a third patient, EchoNext detected heart failure, which led to a heart transplant. So, in a powerful way, these cases demonstrate the impact of EchoNext.

What are your team’s next steps?

We think the accuracy of EchoNext will continue to improve as we train and test the algorithm on additional patient data sets.

We’re also working on the issues of cost and accessibility. Access to digital technologies continues to be a challenge, but we hope to share some exciting developments about making this technology easily available in 2026.

Also in 2026, we’ll be running a randomized control trial. In medicine, this is the most rigorous way to test an intervention or treatment. Eventually, we want cardiologists everywhere to have as much confidence in EchoNext as we do.

Pierre Elias, MD, is an assistant professor of medicine and biomedical informatics at the Vagelos College of Physicians and Surgeons. He is also a principal investigator at the Center for Cardiovascular and Radiologic Deep Learning.