Michigan Team Builds AI That Detects Microvascular Dysfunction From ECGs at 75% Accuracy
Updated
Updated · Detroit News · Aug 2
Michigan Team Builds AI That Detects Microvascular Dysfunction From ECGs at 75% Accuracy
1 articles · Updated · Detroit News · Aug 2
Summary
A University of Michigan study found an AI model could flag microvascular dysfunction from routine ECGs with about 75% accuracy, offering a cheaper screening option for patients with chest pain and related symptoms.
The tool targets a diagnosis now typically confirmed by cardiac PET scans, which Murthy called the gold standard but said are available at only a handful of Michigan centers and can require 3- to 4-month waits.
Researchers trained the system on roughly 4,000 paired PET scan-ECG cases after pretraining it on 800,000 ECGs without PET data, aiming to identify who likely needs advanced testing and who may not.
Murthy said the model is not yet FDA-cleared and will next be tested on patient data from across the U.S. and abroad through the Wellcome Leap VISIBLE program, with a goal of pushing accuracy above 90%.