AI Model Helps Diagnose Often Undetected Heart Disease from Simple EKG

Unlike blockages in large coronary arteries, this condition affects tiny blood vessels and requires advanced imaging to diagnose, which is often unavailable outside specialist centers. New research now shows that artificial intelligence (AI) can identify this condition within seconds using a routine electrocardiogram (ECG).

Researchers at the University of Michigan (Ann Arbor, MI, USA) trained an AI model to detect coronary microvascular dysfunction using standard ECG waveforms, avoiding reliance on expensive imaging. The approach leverages a deep learning architecture known as a vision transformer combined with advanced training strategies.

The team used self-supervised learning to overcome the limited availability of gold-standard imaging data. The model was first pre-trained on more than 800,000 unlabeled ECG waveforms to learn fundamental cardiac electrical patterns, then fine-tuned using a smaller dataset linked to PET myocardial perfusion imaging. It was evaluated across 12 demographic and clinical prediction tasks, including myocardial flow reserve, the gold standard marker for coronary microvascular dysfunction.

The AI model significantly outperformed existing ECG-based AI tools across nearly all diagnostic tasks. It accurately predicted myocardial flow reserve and coronary microvascular dysfunction using resting ECGs, with only minimal performance gains when stress ECGs were added. The findings, published in NEJM AI, show that the model also improved prediction accuracy for several common cardiac conditions, demonstrating robust performance across multiple datasets.

The approach could allow rapid, low-cost screening for coronary microvascular dysfunction in emergency departments and community hospitals. Patients with chest pain and normal angiograms could be identified earlier and referred for advanced testing when needed. Researchers believe this strategy extends the diagnostic power of ECGs beyond rhythm analysis to complex microvascular disease.

“People who come to the ER for chest pain might have CMVD, but their angiogram will show up as ‘clear,’” said study co-author Sascha N. Goonewardena, MD. “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.”

Related Links:
University of Michigan