SeLa-MIL: Developing an instance-level classifier via weakly-supervised self-training for whole slide image classification

Yingfan Ma, Mingzhi Yuan, Ao Shen, Xiaoyuan Luo, Bohan An, Xinrong Chen, Manning Wang

Abstract: Highlights•We propose a weakly supervised self-training method that reformulates MIL as a semi-supervised instance classification problem, allowing for efficient instance-level and bag-level classification in WSIs.•To tackle the challenge of hard positive instance recognition, we introduce global and local constraints that guide self-labeling and prevent pseudo-label degeneration during the training process.•Our method is extensively validated on synthetic, classic MIL benchmark, and public WSI benchmark datasets, demonstrating superior performance and generalization in pathological image analysis.