Medical Image Segmentation: Foundation Model Adaptation versus Task-Specific Training
arXiv cs.AI · 11 Nov 2026, 04:59 AEST
Foundation models pretrained on large natural-image collections are being adapted for medical image segmentation. The adapted foundation model generalised better on external datasets compared to task-specific models.
Key facts
- Foundation models pretrained on large natural-image collections are increasingly adapted for medical image segmentation, in the expectation that they will replace models trained for each task.
- We adapted a foundation model, DINOv2, with strategies ranging from tuning only its bias terms to full fine-tuning, and compared it with U-Nets and ResU-Nets trained from scratch.
- The adapted foundation model generalised better on every external dataset.
- Only a small fraction of the foundation model required adaptation, and the capacity needed was determined by the imaging domain rather than by the adaptation method.
Read aloud
Foundation models pretrained on large natural-image collections are being adapted for medical image segmentation. The adapted foundation model generalised better on external datasets compared to task-specific models. Foundation models pretrained on large natural-image collections are increasingly adapted for medical image segmentation, in the expectation that they will replace models trained for each task. We adapted a foundation model, DINOv2, with strategies ranging from tuning only its bias terms to full fine-tuning, and compared it with U-Nets and ResU-Nets trained from scratch. The adapted foundation model generalised better on every external dataset. Only a small fraction of the foundation model required adaptation, and the capacity needed was determined by the imaging domain rather than by the adaptation method.
Sources in this cluster
- Medical Image Segmentation: Foundation Model Adaptation versus Task-Specific Training
arXiv cs.AI · 11 Nov 2026, 04:59 AEST