Pretraining Multiple Instance Learning Networks with Multi-Teacher Distillation from Pathology Slide Foundation Models

arXiv cs.AI · 6 hours ago

Researchers developed a pretraining framework for multiple instance learning networks used in pathology slide analysis by using knowledge distillation from two foundation models called TITAN and CARE. The framework was evaluated on 15 benchmark datasets, with results showing consistent improvements in linear-probing and few-shot scenarios compared to training from scratch. The pretrained weights enable better transfer learning and reduce overfitting for downstream pathology image analysis tasks.