Reproduce “PRISM: Pathology Reliability In Scarce-label Medicine. A Tile-level Benchmark of Pathology Foundation Models” 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
Alicem Koyun
PRISM evaluates eight pathology foundation models across six tile-level datasets and five-hospital transfers, varying label budgets while measuring discrimination and calibration. It finds that target calibration worsens with more labels mainly when transfer AUROC is low: below 0.85, worsening occurred in 86–88% of combinations, versus 1–8% above it. This is an observed association under a fixed probe protocol, not a causal rule.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.