Foundation Model-Driven Classification of Atypical Mitotic Figures with Domain-Aware Training Strategies
Piotr Giedziun, Jan Sołtysik, Mateusz Górczany, Norbert Ropiak, Marcin Przymus, Piotr Krajewski, Jarosław Kwiecień, Artur Bartczak, Izabela Wasiak, Mateusz Maniewski · 2025
Read Foundation Model-Driven Classification of Atypical Mitotic Figures with Domain-Aware Training Strategies by Piotr Giedziun, Jan Sołtysik, Mateusz Górczany, Norbert Ropiak, Marcin Przymus, Piotr Krajewski, Jarosław Kwiecień, Artur Bartczak, Izabela Wasiak, Mateusz Maniewski (2025) 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/arxiv-2509-02601.txt.
Description
Piotr Giedziun and nine co-authors present a 2025 computational pathology paper on telling normal mitotic figures from atypical ones in tissue images, the second track of the MIDOG 2025 challenge. Shape cues are subtle, and models must work across hospitals and scanners.
Their entry leans on H-optimus-0, a pathology foundation model chosen after cross-domain benchmarks and local tests. Cheap low-rank fine-tuning adapts it, while MixUp augmentation blends training images. Soft labels encode how strongly human experts agreed on each case. Hard negative mining, focal loss, metric learning, and domain adaptation all support robustness. The authors report reasonable preliminary performance, and they frame the work as evidence of both promise and difficulty for foundation models in this task.
Catalog record
| CREATOR | Piotr Giedziun, Jan Sołtysik, Mateusz Górczany, Norbert Ropiak, Marcin Przymus, Piotr Krajewski, Jarosław Kwiecień, Artur Bartczak, Izabela Wasiak, Mateusz Maniewski |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 13 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-02601