Towards robust foundation models for digital pathology
Jonah Kömen, Edwin D. de Jong, Julius Hense, Hannah Marienwald, Jonas Dippel, Philip Naumann, Eric Marcus, Lukas Ruff, Maximilian Alber, Jonas Teuwen, Frederick Klauschen, Klaus-Robert Müller · 2026
Read Towards robust foundation models for digital pathology by Jonah Kömen, Edwin D. de Jong, Julius Hense, Hannah Marienwald, Jonas Dippel, Philip Naumann, Eric Marcus, Lukas Ruff, Maximilian Alber, Jonas Teuwen, Frederick Klauschen, Klaus-Robert Müller (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-13260997.txt.
Description
Jonah Kömen and co-authors introduce PathoROB, a public benchmark that tests how robust pathology foundation models are to non-biological signals. Such models can learn lab procedure and scanner traits instead of tissue biology, which risks harm in clinic. The benchmark scores both the inner representations and the outputs of models on tasks such as patch and slide calls, case search and clustering.
All twenty foundation models tested showed robustness gaps, with wide spread between them. Weak representations led to major diagnostic errors downstream. Using stronger models, vision-language alignment and after-the-fact fixes cut the risk but did not remove it. Robustness checks, the authors argue, must gate clinical use.
Catalog record
| CREATOR | Jonah Kömen, Edwin D. de Jong, Julius Hense, Hannah Marienwald, Jonas Dippel, Philip Naumann, Eric Marcus, Lukas Ruff, Maximilian Alber, Jonas Teuwen, Frederick Klauschen, Klaus-Robert Müller |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 111 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |