AI Framework Reads Cardiac MRI Scans With Up to 98% Accuracy

AI Framework Reads Cardiac MRI Scans With Up to 98% Accuracy

Researchers from Carnegie Mellon University and Cleveland Clinic have developed an artificial intelligence system that can rapidly analyze cardiac magnetic resonance (CMR) imaging with accuracy levels approaching 99% for certain heart conditions.

The vision-language model, called CMR-CLIP (Cardiovascular Magnetic Resonance-Contrastive Language Image Pretraining), learns by pairing video sequences from CMR scans with their corresponding clinical text reports. In real-world clinical evaluations, the framework achieved 98.6% accuracy for hypertrophic cardiomyopathy, 96.2% for cardiac amyloidosis, 88.5% for nonischemic cardiomyopathy, and 88.0% for ischemic cardiomyopathy, according to findings published in Nature Communications.

Why it matters

CMR imaging is the gold standard for cardiac evaluation but requires 40 minutes or more per patient and specialized expertise available only at major medical centers. An AI assistant that can automate screening and standardize reporting could expand access to this diagnostic technology in settings where expert readers are scarce, while reducing interpretation time and variability in clinical reporting.

Training on real clinical data

Unlike typical AI systems that rely on manual labels, CMR-CLIP was trained on more than 13,000 CMR studies from Cleveland Clinic performed between 2008 and 2022, comprising over one million images and hundreds of thousands of motion sequences. The system learned by aligning these images with the impression sections of clinical reports—the summaries containing key findings, differential diagnoses, and care recommendations.

The framework focused on left ventricular diseases, among the most common conditions evaluated with CMR. After training, researchers tested it on two independent datasets: one from University Hospital of Dijon in France and another from Cleveland Clinic Florida sites, both using different scanners and readers than the training data.

Outperforming general AI tools

When compared against two other available AI CLIP systems, CMR-CLIP demonstrated superior performance in identifying common pathologies such as myocardial fibrosis and left ventricular hypertrophy—outperforming them by 32% or more. The system also showed what researchers call "zero shot" capabilities, meaning it could recognize conditions beyond those in its training data.

Notably, CMR-CLIP achieved with a single data instance what other programs required 32 instances to accomplish, according to co-principal investigator David Chen, PhD, of Cleveland Clinic's Cardiovascular Innovation Research Center.

Path to clinical use

The framework remains in the research phase and requires several developments before clinical implementation. Researchers note it will need continuous learning capabilities to stay current, exposure to a more varied mix of diagnoses and patient populations, and additional testing across different locations to ensure generalizability.

Beyond diagnostics, the system could serve as a training tool for residents and fellows, providing a searchable library of images with diagnoses to supplement clinical education. Cardiologist Deborah Kwon, MD, Director of Cardiac MRI at Cleveland Clinic and the study's clinical lead, emphasized the potential for standardizing CMR reporting, which currently shows significant variability in how imaging findings translate to actionable clinical information.

These details were first reported by Cleveland Clinic's AI Watch.