The Pathology AI Library

The library

CleanSlide

Leakage-audited pan-cancer whole-slide multiple-choice benchmark.

Worth knowing about because it takes seriously a problem most benchmarks ignore: whether a model can score well without actually looking at the slide.

Cost
Free
Hardware
Needs a workstation GPU
Offline
Works without a connection
Your data
Stays on your machine
Licence
unknown
Maturity
research
Source codeDataset

What it is

TCGA-derived benchmark of about 149,000 four-option questions over 9,985 diagnostic slides, with train/validation/test splits disjoint by patient AND by tissue source site, and a blind image-free baseline reported for every task.

Licence notes

Check the repository and model card before any reuse.

Activity: 1 stars ยท last commit 2026-07-23 ยท 30 downloads โ€” refreshed nightly

Related

Catalogued 2026-08-02, last checked 2026-08-02. View the source record ยท Report an error