Learning Emergent Modular Representations in Multi-modality Medical Vision Foundation Models
About
Multi-modality medical vision (MV) foundation models (FM) are fundamentally challenged by pronounced Non-IID feature statistics across heterogeneous imaging modalities. Monolithic self-supervised optimization on such data induces conflicting gradients, driving representations to collapse toward modality-dominant shortcuts. This work reframes this failure as an imbalance between specialization and coordination in emergent modularity, and proposes Director-Experts (DEX), a modular network that explicitly regulates these dynamics in stacked modules. Each DEX module comprises a pool of experts, dynamically adapted by our image-wise activation strategy, autonomously specializing in modality-dominant statistics, together with a director, updated via our group exponential moving average, which distills multi-expert knowledge into a shared space for semantic integration across modalities, thus driving the emergence of modular representations. We curate a new benchmark, Medical Vision Universe, over 4 million images across 10 modalities, which provides a FM-level pre-training with the broadest coverage of distinct imaging modalities to our DEX. Extensive evaluations on 26 downstream tasks demonstrate improved optimization behavior and transferability, indicating DEX as a principled step toward general-purpose multi-modality medical AI. Our code and dataset will be opened at https://github.com/YutingHe-list/DEX.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Medical Image Analysis | Fundus 2 tasks | Average Performance (%)75.7 | 13 | |
| Medical Image Analysis | Path 4 tasks | Average Performance69.9 | 13 | |
| Medical Image Analysis | X-ray 6 tasks | Average Performance85.9 | 13 | |
| Medical Image Analysis | US 3 tasks | Average Performance (%)82.5 | 13 | |
| Medical Image Analysis | 26 Medical Downstream Tasks | Average Performance78.4 | 13 | |
| Medical Image Analysis | CT 2 tasks | Average Performance83.9 | 13 | |
| Medical Image Analysis | Endo | Average Performance65 | 13 | |
| Medical Image Analysis | MR 2 tasks | Average Performance72.9 | 13 | |
| Medical Image Analysis | OCT 2 tasks | Average Performance87.5 | 13 | |
| Medical Image Analysis | Photo 2 tasks | Average Performance87.1 | 13 |