Foundation models for computational pathology are expected to facilitate the development of high-performing, generalisable deep learning systems. However, in addition to biologically relevant features, current foundation models also capture pre-analytic and scanner-specific variation that bias the predictions made by downstream task-specific models trained on these features. Here we show that introducing novel robustness losses during downstream model training reduces sensitivity to technical variability. A purpose-designed comprehensive experimentation setup with 27,042 whole-slide images from 6,155 patients is used to train thousands of models from the features of eight well-known foundation models for computational pathology. In addition to a substantial improvement in robustness, our approach improves classification accuracy by focusing on biologically relevant features. It mitigates robustness limitations of foundation models for computational pathology without retraining the foundation models themselves, enabling development of models that are more suitable in real-world clinical use.
@article{arxiv.2602.22347,
title = {Enabling clinical use of foundation models for computational pathology},
author = {Audun L Henriksen and Ole-Johan Skrede and Lisa van der Schee and Enric Domingo and Karolina Cyll and Sepp de Raedt and Ilyá Kostolomov and Jennifer Hay and Wanja Kildal and Joakim Kalsnes and Robert W Williams and Manohar Pradhan and John Arne Nesheim and Hanne Askautrud and Maria Isaksen and Karmele Saez de Gordoa and Miriam Cuatrecasas and Joanne Edwards and TransSCOT group and Arild Nesbakken and Neil A Shepherd and Ian Tomlinson and Daniel-Christoph Wagner and Rachel Kerr and Tarjei Sveinsgjerd Hveem and Knut Liestøl and Yoshiaki Nakamura and Marco Novelli and Masaaki Miyo and Sebastian Försch and David N Church and Miangela M Lacle and David J Kerr and Andreas Kleppe},
journal= {arXiv preprint arXiv:2602.22347},
year = {2026}
}