A generalizable deep learning system for cardiac MRI - PubMed
6 hours ago
- #cardiac MRI
- #deep learning
- #clinical diagnostics
- A deep learning system for cardiac MRI uses self-supervised contrastive learning, trained on cine-sequence scans and radiology reports.
- The model is trained on data from four U.S. institutions and tested on UK BioBank and public datasets, showing generalizability.
- It achieves clinical-grade accuracy in tasks like left-ventricular ejection fraction regression and diagnosing 39 conditions, including cardiac amyloidosis.
- The system requires less training data than typical for such tasks and can contextualize complex cardiovascular diseases.