E2-MIL: An explainable and evidential multiple instance learning framework for whole slide image classification

Jiangbo Shi, Chen Li, Tieliang Gong, Huazhu Fu

Abstract: Highlights•Propose an explainable and evidential multiple instance learning (MIL) framework for whole slide image (WSI) classification.•A detail-aware attention distillation module and a structure-aware attention refined module to improve the model’s interpretability.•An uncertainty-aware classifier to improve the model’s reliability by providing robust instance-level predictive uncertainty estimation.•Extensive studies demonstrate our framework outperforms state-of-the-art methods for WSI classification.