Researchers at Ann Arbor, Mich.-based University of Michigan have developed an AI model that can detect coronary microvascular dysfunction — a frequently missed and difficult-to-diagnose heart condition — using a standard 10-second electrocardiogram.
The condition affects the heart’s small vessels and often goes undiagnosed because it requires specialized imaging, such as PET myocardial perfusion scans, according to a Dec. 15 news release. The AI model, described in a study published Nov. 26 in NEJM AI, was trained on more than 800,000 unlabeled EKG waveforms and fine-tuned using PET data to predict myocardial flow reserve, considered the gold standard for diagnosing the condition.
The model consistently outperformed earlier AI tools and was effective even when using resting EKGs, with minimal performance gain from stress tests, according to a Dec. 15 news release from the health system. Researchers said it could be particularly valuable in hospitals without access to advanced cardiac imaging, where it may help identify patients who would benefit from further testing.
About 14 million people visit emergency departments or clinics annually for chest pain, but many cases go undiagnosed when traditional imaging appears normal. The tool could offer a noninvasive, low-cost solution to close that gap, the release said.
