Study: Artificial Intelligence Model Reads Cardiac MRI Scans with Near Expert Accuracy
Study: Artificial Intelligence Model Reads Cardiac MRI Scans with Near Expert Accuracy
Publish Date: 2026-03-25 06:00:00
Source Domain: www.newswise.com
- Penn Medicine-led team has developed an innovative AI system to interpret cardiac MRI scans with performance near that of expert clinicians.
- Trained on over 300,000 MRI video clips from approximately 20,000 patients, the model assesses heart function and diagnoses numerous diseases using non-contrast imaging.
- The AI, referred to as a “foundation model,” accurately links MRI videos to radiology reports, allowing it to recognize a variety of conditions with minimal labeled data.
- In trials, the AI system achieved expert-level accuracy in estimating ejection fraction and effectively identified severe heart dysfunction compared to traditional AI methods.
- The model diagnosed 39 cardiac conditions, such as hypertrophic and dilated cardiomyopathies, with high diagnostic accuracy scores.
- In a real-world screening of over 40,000 scans, the AI identified 112 previously undiagnosed cases of hypertrophic cardiomyopathy.
- Researchers believe the system can assist hospitals lacking specialized cardiac imaging expertise in detecting rare but treatable diseases earlier.
- The team plans to conduct prospective clinical studies and expand the training data with additional scans, and a pretrained model is available for free use by academic institutions.
