Doctors at the University of Michigan have developed an artificial intelligence model that can rapidly detect coronary microvascular dysfunction (CMVD) using standard electrocardiograms (EKGs), according to a new study published in NEJM AI, part of the New England Journal of Medicine family.
The research team trained the AI tool to identify CMVD, a condition affecting small blood vessels in the heart that is often difficult to diagnose and typically requires advanced imaging such as PET myocardial perfusion scans. These scans are costly and not widely available outside specialized centers, limiting access for many patients who experience chest pain each year.
To overcome data limitations, the researchers used self-supervised learning techniques. They first pre-trained a deep learning model known as a vision transformer on over 800,000 unlabeled EKG waveforms before fine-tuning it with a smaller dataset labeled with PET scan results. "Essentially, we taught the model to 'understand' the electrical language of the heart without human supervision," said lead researcher Murthy.
The AI model was evaluated across 12 demographic and clinical prediction tasks—including several beyond what current EKG-AI models can perform—and demonstrated improved diagnostic accuracy for both CMVD and more common cardiac conditions compared to earlier tools. Notably, using stress test EKGs provided only minimal additional benefit over resting EKGs in these predictions.
"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, M.D., associate professor of internal medicine-cardiology at U-M Medical School. "In hospitals with limited resources or non-specialty centers, using our EKG-AI model to predict myocardial flow reserve and CMVD will be an easy, cost-effective and non-invasive way to identify when a patient would benefit from advanced testing for a serious condition."
By training their system with gold-standard PET scan data instead of relying solely on large EKG databases—typically used for general rhythm analysis—the University of Michigan team aims to make detection of this challenging condition faster and more accessible.
