Reproduce “Mitigating Spatial Blindness in Multiple Instance Learning for Whole-Slide Image Analysis” 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
Xiangyu Li, Jincheng Li, Xing Xie, Bin Guan, Shaohua Kevin Zhou, Rui Yan, Ran Su
ResTopoMIL tests whether staged training helps a graph-based WSI predictor use neighborhoods. It first trains and freezes a composition-based predictor, then adds a graph correction trained using coordinate-reassigned views. On a spatial MNIST-bag task, it achieves AUC 0.987 versus 0.684 for joint training. On PANDA, mean AUC is 0.9426 versus 0.9309 for DS-MIL. Coordinate-free residuals also help, limiting attribution to neighborhoods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.