Jonathan Moody, Alexis Poitrasson-Rivière, Jennifer M. Renaud, Michael Vanderver, Edward P. Ficaro, Venkatesh Locharla Murthy. Self-supervised deep representation learning of a foundation transformer model enabling comprehensive ECG-based assessment of cardiovascular health with limited labeled data. Journal of the American College of Cardiology. 2025;85(12):2758.
Description:
This poster, presented at the 2025 American College of Cardiology Annual Scientific Session and published as an abstract in the Journal of the American College of Cardiology (2025), presents the development of an artificial intelligence (AI) self-supervised learning (SSL) model for interpreting ECGs using a transformer architecture. The model was trained on over 1.6 million ECGs and fine-tuned on multiple classification tasks with limited labeled data.
Key Findings:
“We observed high performance across all tasks despite limited labeled data, demonstrating the potential of self-supervised learning for scalable cardiovascular assessment,” the authors wrote.
Tasks included assessment of stress and rest myocardial blood flow, myocardial flow reserve, total perfusion deficit and left ventricular ejection fraction.
Clinical Relevance:
This early-stage investigation highlights the potential for AI-driven, self-supervised transformer models to support scalable and comprehensive ECG-based cardiovascular assessment.
Partners in Research:
INVIA Medical Imaging Solutions and the Division of Cardiovascular Medicine in the Department of Internal Medicine at the University of Michigan collaborated on this research.
