Reproduce “Multiple Instance Learning Framework with Masked Hard Instance Mining for Whole Slide Image Classification” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wenhao Tang, Sheng Huang, Xiaoxian Zhang, Fengtao Zhou, Yi Zhang, Bo Liu
An MIL framework for Whole Slide Image (WSI) classification, MHIM-MIL, uses masked hard instance mining to address the bias of prior methods toward easily classifiable instances. This approach leverages a momentum teacher to identify and prioritize training on subtle 'hard' tumor regions, enhancing diagnostic accuracy and generalization across histopathological datasets while improving computational efficiency.
Key insights
- Focusing training on "hard-to-classify" instances, rather than solely salient ones, is essential for improving model generalization and discriminability in MIL for WSIs.
- A momentum-updated teacher network can effectively and implicitly identify hard instances through attention score-based masking, circumventing the need for explicit instance-level labels.
- Combining a consistency loss between teacher and student bag representations with iterative EMA updates fosters stable training and progressively enhances both models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.