AI Breakthrough Enables Expert-Level Heart Scan Analysis

A research team led by Penn Medicine has developed an advanced artificial intelligence system capable of interpreting cardiac MRI scans with accuracy comparable to specialist clinicians. The study, published in Nature Biomedical Engineering, highlights a major step forward in medical imaging and diagnostics.

Trained on over 300,000 MRI video clips from around 20,000 patients, the AI model can evaluate heart function and detect a wide range of cardiac conditions using only non-contrast imaging. This is particularly significant for hospitals that lack specialized expertise in cardiac MRI interpretation, especially in rural or resource-limited settings.

The system functions as a “foundation model,” learning by linking MRI videos with corresponding radiology reports. This approach allows it to recognize patterns and diagnose conditions without relying heavily on manually labeled data. In testing, the AI demonstrated expert-level accuracy in measuring ejection fraction—a key indicator of heart performance—and identified severe dysfunction more effectively than traditional AI tools.

Notably, the model successfully diagnosed 39 heart conditions, including hypertrophic and dilated cardiomyopathies, achieving accuracy scores as high as 0.97. In a large-scale real-world screening of over 40,000 scans, it flagged 112 previously undiagnosed cases of hypertrophic cardiomyopathy, underlining its potential for early detection.

Researchers believe this innovation could significantly improve access to quality cardiac care by supporting clinicians in settings where expertise is limited. Further clinical trials are planned, and the model has been made freely available for academic use to encourage continued research and development.