AI Algorithm Could Improve Heart Disease Detection and Prediction

Doctors use 12-lead ECGs to assess a heart’s electrical activity, and artificial intelligence is often used to interpret the results. However, current AI tools require a large amount of manually labeled data to learn to recognize signs of heart disease. As a result, it is difficult for AI tools to learn new tasks.

Scientists at Scripps Research realized this bottleneck and built a new model called ECG-CLIP to fix it. The tool trained on over 1.7 million ECGs from more than 540,000 people. This new AI algorithm analyzes the test results and reads the clinicians’ notes attached to them.

AI Tool that Learns Like a Human

By learning from both charts and notes, ECG-CLIP matched older models while using about 91% less hand-labeled data.

“Our new algorithm only needs to see on the order of a dozen confirmed ECGs of a specific disease to detect that disease in the future,” Giorgio Quer, senior author and assistant professor at Scripps Research, explained. “This is similar to how a clinician would learn: not from a million examples, but from understanding the general physiology behind an ECG first and then seeing a few specific cases.”

The team tested the tool on several tasks. It easily spotted heart diseases like acute myocardial infarction. Additionally, it beat other tools at predicting future atrial fibrillation from a normal-looking ECG. It even did the best job predicting other outcomes, like a patient’s chances of developing chronic kidney disease or type II diabetes within three years.

Building Trust With Doctors

One of the most interesting aspects, doctors say, is that ECG-CLIP works when data or resources are highly limited. For example, it performed well using single-lead ECG data.

“We found that ECG-CLIP was better at detecting and predicting cardiovascular diseases, particularly in cases where there was much less data,” Quer added. “This may be particularly useful in cases like rare diseases, where there are only a dozen or so positive examples of well-labeled ECGs that can be used for training the model.”

The tool uses maps that highlight exactly which part of the heart signal the model used to make its choice, providing more trust to doctors. Eventually, the team hopes to test the model with wearable devices for remote monitoring.

“While these findings demonstrate substantial potential, rigorous validation in prospective clinical trials will be required to establish ECG-CLIP’s applicability in real-world clinical settings,” said study co-first author Michael Ko.