Enabling Ultra-Fast Cardiovascular Imaging Across Heterogeneous Clinical Environments with A Generalist Foundation Model and Multimodal Database
A new foundation model cuts cardiac MRI scan times from 45 minutes to under 2 minutes while preserving diagnostic quality.
A consortium of over 60 researchers has published a breakthrough in medical AI, introducing CardioMM—a generalist foundation model designed to solve the physics-constrained inverse problem of reconstructing cardiovascular MRI (CMR) scans from raw sensor (k-space) data. The model's power stems from its training on MMCMR-427K, the largest and most diverse multimodal CMR database ever assembled, containing 427,465 multi-coil k-space datasets. This unprecedented resource spans 13 international centers, 12 CMR modalities, 15 scanner models from four different field strengths, and 17 cardiovascular disease categories across three continents, providing the robust, heterogeneous data needed for a truly generalist model.
CardioMM unifies semantic contextual understanding with physics-informed data consistency, allowing it to dynamically adapt to varied scanners, protocols, and patient presentations. The key result is its ability to support acceleration factors up to 24x. In practice, this can reduce a standard 45-minute cardiac MRI scan to under two minutes. Comprehensive evaluations show CardioMM achieves state-of-the-art performance at internal centers and demonstrates strong zero-shot generalization to completely unseen external clinical environments. Critically, the research provides the first evidence that such extreme acceleration can preserve essential clinical information, including key cardiac phenotypes, quantitative myocardial biomarkers, and diagnostic image quality, without compromising clinical integrity.
- Trained on MMCMR-427K, a massive new database of 427,465 multi-coil k-space scans from 13 global clinical centers.
- Enables up to 24x acceleration, potentially reducing a 45-minute cardiac MRI scan to under 2 minutes.
- Demonstrates strong zero-shot generalization across unseen scanners and protocols while preserving diagnostic quality and biomarkers.
Why It Matters
This could dramatically increase patient throughput, reduce costs, and make critical cardiac diagnostics accessible in more diverse and resource-constrained clinical settings.