The Geometry of Transfer: Unlocking Medical Vision Manifolds for Training-Free Model Ranking

Researchers developed a new framework for selecting optimal medical AI foundation models without costly fine-tuning, achieving 31% better performance than existing methods. The topology-driven approach evaluates manifold tractability rather than statistical overlap to better assess model transferability for medical image segmentation tasks.

• →New topology-driven framework eliminates need for expensive fine-tuning when selecting medical AI models for segmentation tasks.
• →The approach uses three components: Global Representation Topology Divergence, Local Boundary-Aware Topological Consistency, and Task-Adaptive Fusion.
• →Method achieves 31% relative improvement over state-of-the-art baselines in weighted Kendall metric.
• →Framework addresses computational bottleneck in selecting from the growing number of medical foundation models.
• →Validation conducted on large-scale OpenMind benchmark across diverse anatomical targets and SSL foundation models.