Reproduce “Hard Negative Sample 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
Wentao Huang, Xiaoling Hu, Shahira Abousamra, Prateek Prasanna, Chao Chen
This work introduces a hard negative sample mining approach combined with iterative encoder fine-tuning and a multiple instance ranking loss to enhance Whole Slide Image (WSI) classification. The method improves classification accuracy and efficiency by selectively focusing on the most informative negative examples, demonstrating superior performance on breast cancer metastasis detection and EGFR mutation prediction datasets.
Key insights
- Focusing on hard negative samples (those most challenging to distinguish) provides a more informative learning signal than using all negative patches, leading to better model discrimination.
- Selecting only a small fraction of negative patches significantly reduces computational costs and training time without sacrificing performance.
- Iteratively fine-tuning the feature encoder with selected hard negative samples and positive samples, guided by an initial MIL model, refines representations effectively for weakly supervised WSI tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.