Doctors may soon be able to diagnose a tricky heart condition within seconds using artificial intelligence and a standard heart test, according to new research from the University of Michigan.
The condition is called coronary microvascular dysfunction, or CMVD, and it affects the small blood vessels in the heart. It is difficult to detect using common tests and is often missed, even in emergency rooms.
Researchers trained an advanced AI model to identify signs of CMVD using only a 10-second electrocardiogram (EKG), a test that records the heart’s electrical signals.
Usually, CMVD requires complex and expensive imaging tests to diagnose, such as PET scans, which are not widely available. This new AI model could make diagnosis much easier, faster, and cheaper.
The study, published in NEJM AI, showed that this new AI tool worked better than earlier models in almost every task, including predicting something called myocardial flow reserve.
This measurement is considered the best way to find CMVD. Dr. Venkatesh Murthy, senior author of the study and a cardiology expert at the University of Michigan, explained that this tool could help doctors spot a condition that is usually very hard to catch.
Every year, about 14 million people in the U.S. visit a doctor or emergency room with chest pain. Some of them have CMVD, but because their large heart arteries are clear, traditional tests can miss the diagnosis. That’s where this AI tool could help.
To build the model, researchers used a method called self-supervised learning. They first trained the AI to “read” over 800,000 EKGs without telling it what the patterns meant.
Then they gave it a smaller group of EKGs linked to PET scan data, which showed more clearly what CMVD looks like. In this way, the model learned to connect electrical patterns in the heart to signs of disease.
Once trained, the model was tested on 12 prediction tasks, including those used to measure heart flow and detect different kinds of heart disease.
It was also able to make predictions about heart conditions that current AI tools using EKGs cannot. The model performed well across different databases and consistently improved the ability to detect not only CMVD but also other heart problems.
Some of the EKGs used for testing were taken during exercise, which is common in stress tests. However, the model worked nearly as well using resting EKGs, which are simpler and easier to collect.
While AI tools have already been used to help read EKGs, most current models only detect common heart issues, like irregular heartbeats or weak heart pumping. This new model goes further by helping to find CMVD, a condition that affects tiny blood vessels and does not show up in regular scans.
Dr. Sascha Goonewardena, one of the study’s co-authors, noted that many patients with chest pain are told their tests are normal, even when they might have CMVD. The AI model could be used in clinics or smaller hospitals to decide if someone needs more advanced testing, using just an EKG.
In summary, this new AI tool shows promise as a fast, affordable, and noninvasive way to catch a hidden form of heart disease that is often missed. If successful in real-world settings, it could help millions of patients get the right care sooner.
If you care about heart disease, please read studies that herbal supplements could harm your heart rhythm, and how eating eggs can help reduce heart disease risk.
For more health information, please see recent studies that apple juice could benefit your heart health, and results showing yogurt may help lower the death risks in heart disease.
Copyright © 2025 Knowridge Science Report. All rights reserved.
