Towards cardiac MRI foundation models: Comprehensive visual-tabular representations for whole-heart assessment and...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Cardiac magnetic resonance (CMR) imaging is crucial for non-invasive cardiac assessment.
• Patient-level health factors significantly impact cardiovascular health but are often not integrated with CMR.
• Existing multi-modal approaches have limitations in data scope and task specificity.

Purpose of the Study:

• To introduce ViTa, a foundation model for comprehensive cardiac health representation and disease risk interpretation.
• To overcome limitations of current multi-modal approaches by integrating diverse data types.
• To develop a unified framework for a patient-specific understanding of cardiac health.

Main Methods:

• Leveraged data from 42,000 UK Biobank participants.
• Integrated 3D+T cine CMR imaging (short- and long-axis views) with tabular patient-level factors.
• Developed a multi-modal paradigm to learn a shared latent representation.

Main Results:

• ViTa captures the complete cardiac cycle and fuses imaging with patient context for holistic insights.
• The model supports diverse downstream tasks including phenotype prediction, segmentation, and disease classification.
• Achieved a unified framework for cardiac/metabolic disease analysis.

Conclusions:

• ViTa represents a significant step toward foundation models for cardiac health.
• The integrated approach enhances the precision of individual disease risk interpretation.
• ViTa demonstrates potential for advancing clinical utility and scalability in cardiac analysis.