An artificial intelligence (AI) model developed by researchers at the University of Michigan and Invia Medical Imaging Solutions can help diagnose a relatively common but underdiagnosed heart condition using a simple electrocardiogram (ECG).
In a recent study, the researchers showed that using a self-supervised artificial intelligence model they could accurately diagnose coronary microvascular dysfunction. This condition impacts the very small blood vessels in the heart that fail to dilate or regulate blood flow properly, causing ischemia even when the large coronary arteries appear to be normal.
“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 co-lead author Venkatesh Murthy, MD, PhD, a clinician and senior researcher at the University of Michigan, in a press statement.
The self-supervised AI model was initially trained using 800,000 unlabeled ECGs. It was then fine tuned using data from positron emission tomography (PET), single-photon emission computed tomography (SPECT), and magnetic resonance imaging (MRI) scans in addition to clinical ECG reports.
The results showed the model was able to detect coronary microvascular dysfunction correctly 70-80% of the time. It outperformed other AI models and was able to predict factors like myocardial flow reserve and left ventricular ejection fraction to a good degree of accuracy.
Notably, the model was able to match the accuracy of a fully supervised model for some tasks while using less than one‑fifth of the labeled data, and it still performed strongly when applied to different types of imaging and in different patient groups.
The gold-standard method for diagnosing coronary microvascular dysfunction, which is estimated to affect up to four million people in the U.S., is specialized invasive tests in the cath lab or advanced imaging such as PET or stress cardiac MRI. These can be time consuming and expensive.
“People who come to the ER for chest pain might have coronary microvascular dysfunction, but their angiogram will show up as ‘clear,’” said co-author Sascha Goonewardena, MD, an associate professor at University of Michigan Medical School.
“In hospitals with limited resources or non-specialty centers, using our ECG-AI model to predict myocardial flow reserve and coronary microvascular dysfunction will be an easy, cost-effective and non-invasive way to identify when a patient would benefit from advanced testing for a serious condition.”
