AI Model Diagnoses Coronary Microvascular Dysfunction From a 10-Second EKG Strip

U of Michigan AI detects CMVD from standard EKGs, potentially transforming cardiac care.

"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 EKG strip."
Dr. Venkatesh L. Murthy, U of Michigan

Coronary microvascular dysfunction (CMVD) is a condition affecting millions of patients with chest pain, yet it remains notoriously difficult to diagnose with conventional methods. Researchers at the University of Michigan have developed an AI model that can identify CMVD from standard electrocardiogram (EKG) strips in just 10 seconds.

The breakthrough model was trained on a massive dataset of unlabeled EKGs and learns to make 12 different prediction tasks simultaneously. This multi-task learning approach enables the AI to detect subtle patterns that human clinicians might miss, even in emergency department settings where rapid diagnosis is critical.

The model's ability to diagnose CMVD from standard EKGs addresses a significant clinical gap. Many patients with CMVD visit emergency departments with chest pain but receive a diagnosis of "normal" because conventional tests fail to detect the underlying dysfunction. This AI solution could help clinicians accurately identify the condition and provide appropriate treatment.

Key Data Points

14M
Annual chest pain visits
800K+
Unlabeled EKGs for pre-training
12
Prediction tasks
10s
Diagnosis time per EKG

Sources