AI Model Detects CMVD with Impressive Speed and Accuracy

A newly developed artificial intelligence model is showing early promise as a fast and cost-effective way to detect coronary microvascular dysfunction (CMVD) using just a 10-second electrocardiogram.

In findings published in NEJM AI, researchers from the University of Michigan describe a deep learning model capable of identifying signs of CMVD—an often-overlooked heart condition that affects the small vessels of the heart but doesn’t show up on traditional angiograms. The condition is notoriously difficult to diagnose, particularly in emergency settings, where patients may present with chest pain but normal angiogram results.

“Our model creates a way for clinicians to accurately identify a condition that is notoriously hard to diagnose—and often missed in emergency department visits—using a 10-second ECG strip,” said Venkatesh L. Murthy, MD, PhD, senior author and associate chief of cardiology for translational research and innovation at the U-M Health Frankel Cardiovascular Center.

To train the model, Murthy and colleagues used data from over 800,000 unlabeled ECGs. The system was then fine-tuned using PET imaging results and clinical reports, allowing it to learn more nuanced patterns associated with cardiac function and dysfunction. According to the authors, the model showed strong performance across several key diagnostic areas. Its area under the ROC curve ranged from 0.763 for detecting impaired myocardial flow reserve to 0.955 for impaired left ventricular ejection fraction.

“Essentially, we taught the model to ‘understand’ the electrical language of the heart without human supervision,” Murthy explained.

The team reported improved diagnostic performance for 11 out of the 12 prediction tasks evaluated in the study, suggesting the AI model may help identify patients with CMVD who would benefit from further testing—especially in hospitals that may lack access to advanced imaging or cardiovascular specialists.

“People who come to the ER for chest pain might have CMVD, but their angiogram will show up as ‘clear,’” said co-author Sascha N. Goonewardena, MD, associate professor of cardiology at U-M Medical School. “In hospitals with limited resources or non-specialty centers, using our ECG-AI model to predict myocardial flow reserve and CMVD will be an easy, cost-effective and noninvasive way to identify when a patient would benefit from advanced testing for a serious condition.”

The work was supported by the Department of Veterans Affairs, the National Heart, Lung and Blood Institute, the National Institute of Diabetes and Digestive and Kidney Diseases, and the National Institute on Aging.