Reproduce “CAMEL: A Weakly Supervised Learning Framework for Histopathology Image Segmentation” 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
Gang Xu, Zhigang Song, Zhuo Sun, Calvin Ku, Zhe Yang, Cancheng Liu, Shuhao Wang, Jianpeng Ma and 1 more
A weakly supervised learning framework, CAMEL, generates high-quality instance and approximate pixel-level labels for histopathology image segmentation from only image-level annotations. It achieved segmentation performance comparable to fully supervised methods on CAMELYON16 and a colorectal adenoma dataset.
Key insights
- Combining complementary Max-Max and Max-Min Multiple Instance Learning criteria effectively generates balanced and high-quality instance-level pseudo-labels from image-level supervision.
- Automatically enriching coarse image-level labels into fine-grained instance and approximate pixel-level labels enables weakly supervised methods to achieve segmentation performance on par with fully supervised models.
- Finer instance granularity (e.g., 160x160 vs. 320x320) improves segmentation performance and mitigates potential over-labeling artifacts in histopathology images.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.