How it works

Self-supervised learning builds tissue representations from unlabeled slides. The model then fine-tunes on a small labeled set to reach diagnostic accuracy.