Updated
Updated · Detroit News · Aug 2
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%.

Insights

With 75% accuracy, will this AI triage tool prevent unnecessary heart tests, or dangerously deny critical PET scans to patients in need?
How exactly does an AI trained on standard heart scans uncover a deadly microvascular disease completely invisible to traditional cardiology?