AI Can Identify Cardiomyopathies Early, Says New Study

An experimental new screening test that pairs AI with portable cardiac ultrasounds often performed during emergency room visits and in the community could ensure more people receive care for two types of cardiomyopathy, a deadly heart condition, according to a study published Jan. 29 in The Lancet Digital Health.

Cardiomyopathies occur when genetic or other health conditions weaken or damage heart muscles. Testing for these conditions relies on advanced diagnostic imaging tests, which can be expensive, hard to access, and require experts to interpret these tests, so they often go underdiagnosed.

Now, researchers at the Cardiovascular Data Science (CarDS) Lab at Yale School of Medicine (YSM) have developed an algorithm to recognize signs of two common types of cardiomyopathies in ultrasound imaging captured during emergency room visits over the last decade across two health systems—with AI picking up signs of disease two years, on average, before the patients were formerly diagnosed.

AI-based tests like this one could be deployed in emergency rooms to flag people at high risk of cardiomyopathies, says Evangelos Oikonomou, MD, DPhil, a clinical fellow in cardiovascular medicine and postdoctoral fellow in the CarDS Lab at YSM. “We can actually minimize how many cases fall through the cracks,” he says.

Cardiomyopathies are far more common than we used to think. The Centers for Disease Control and Prevention (CDC) now estimates that as many as 1 in 500 people might have a form of cardiomyopathy. But even today, many people who live with the condition —or die of it—go undiagnosed.

Part of the problem is that getting tested for cardiomyopathies takes time and money. Because there is no routine screening or telltale symptoms, patients need to be in a situation where a medical professional recognizes that they have symptoms of the illness, such as shortness of breath or chest pains, before undergoing expensive testing.

But there might be an easier way for doctors to flag that someone may be at risk for cardiomyopathy. Ultrasounds for heart conditions usually last around 30 minutes. But people who visit emergency rooms with symptoms of heart distress sometimes receive cheaper, fast ultrasounds of their chests. These tests aren’t meant to pick up cardiomyopathy. But they might present an opportunity to screen for cardiomyopathy risk—not by asking overworked emergency room physicians to search for symptoms, but to have an algorithm scan information that is already acquired as part of clinical care.

AI helps detect cardiomyopathies early

To test this hypothesis, the team decided to focus on two specific types of cardiomyopathy. The first is hypertrophic cardiomyopathy, where heart muscles thicken to the point that the heart has trouble beating. The researchers also screened for transthyretin amyloid cardiomyopathy, where a protein buildup causes the heart muscles to stiffen.

Oikonomou and the rest of the CarDS Lab developed their algorithm by using over 90,000 brief ultrasounds collected over a decade. Around 550 people in their data set were later diagnosed with one of the two cardiomyopathies.

After training, the algorithm correctly flagged almost all the positive cases. But that wasn’t all. Screening also picked up early signs of illness in several ultrasounds captured between six months and four and a half years before the patients received an official diagnosis—suggesting that the algorithm could spot things that experts wouldn’t have noticed.

This matters because early intervention is crucial for treating cardiomyopathy. Transthyretin amyloid cardiomyopathy is particularly time-sensitive—early intervention can increase a patient’s odds of survival by 30%.

The algorithm also picked up what may have been a few false positives. However, this might be related to the underdiagnosis of cardiomyopathies, says Oikonomou. People who were flagged “at risk” of cardiomyopathy by the algorithm but never diagnosed were still 17-32% more likely to die within two years of receiving their ultrasound than people who were never flagged. This suggests to Oikonomou that some of them may have had a heart condition that was simply never identified.

Ultimately, the goal of the algorithm is to develop a cheap and easy way for doctors to screen for cardiomyopathies while performing other care. Once flagged, people will be sent for further testing to confirm their condition. AI won’t replace diagnosis or medical treatment from a medical professional, says Oikonomou.

That being said, AI can provide invaluable new tools for spotting signs of disease, says Rohan Khera, MD, an assistant professor of cardiovascular medicine and health informatics and the director of the CarDS Lab.

“Why do AI?” he asks. “Because it can pick up things from these images that human experts—even us cardiologists—cannot.”

The research reported in this news article received support from the National Heart, Lung, and Blood Institute (R01HL167858 and K23HL153775 to RK, and F32HL170592 to EKO), the National Institute on Aging (R01AG089981 to RK), the Doris Duke Charitable Foundation

(2022060 to RK), and BridgeBio through an investigator-initiated study (RK), with approximately 20% nongovernmental funding.