ViTa: A Multi-Modal Foundation Model for Cardiac MRI
A new study published in Medical Image Analysis introduces ViTa, a multi-task, multi-modal foundation model for cardiac health assessment. ViTa combines 3D+T cardiac MRI with detailed tabular health data from 42,000 UK Biobank participants. This enables dynamic, patient-specific cardiac representations for phenotype prediction, disease classification, and segmentation.
The model is trained using self-supervised and contrastive learning methods and achieves state-of-the-art performance across several tasks, especially under data imbalance conditions. Suprosanna Shit of the Department of Quantitative Biomedicine (DQBM), University of Zurich, contributed to this interdisciplinary effort, which highlights DQBM's expertise in medical AI and scalable health analysis frameworks.
Read the full paper: https://doi.org/10.1016/j.media.2025.103756