ContriMix: Scalable stain color augmentation for domain generalization without domain labels in digital pathology

Lisa Hensens, Sergio Sabroso-Lasa, Caroline Verbeke, Nuria Malats, ThePanGenEU consortium, Geert Litjens, Pierpaolo Vendittelli
Proceedings of the MICCAI Workshop on Computational Pathology, PMLR 316:97-105, 2026.

Abstract

Pancreatic ductal adenocarcinoma (PDAC) remains one of the deadliest cancers due to late detection and limited treatment response. This study investigates the prognostic value of combining multiple AI-based image biomarkers—tumor-stroma ratio (TSR), mitosis density, stromal cell density (via HoVer-Net), tumor-to-tissue ratio from histopathological whole-slide images (WSIs) for survival prediction in resected PDAC patients. A multi-tissue segmentation model was developed to generate tissue masks for downstream biomarker extraction. Using logistic and Cox regression models, both univariate and multivariate survival analyses were performed across four datasets. Results show that while combining biomarkers did not outperform single-biomarker models (notably TSR), mitosis density showed consistent statistical significance and may serve as a valuable prognostic feature.

Cite this Paper

BibTeX
@InProceedings{pmlr-v316-hensens26a, title = {ContriMix: Scalable stain color augmentation for domain generalization without domain labels in digital pathology}, author = {Hensens, Lisa and Sabroso-Lasa, Sergio and Verbeke, Caroline and Malats, Nuria and consortium, ThePanGenEU and Litjens, Geert and Vendittelli, Pierpaolo}, booktitle = {Proceedings of the MICCAI Workshop on Computational Pathology}, pages = {97--105}, year = {2026}, editor = {Studer, Linda and Ciompi, Francesco and Khalili, Nadieh and Faryna, Khrystyna and Faryna, Khrystyna and Yeong, Joe and Lau, Mai Chan and Chen, Hao and Liu, Ziyi and Brattoli, Biagio}, volume = {316}, series = {Proceedings of Machine Learning Research}, month = {27 Sep}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v316/main/assets/hensens26a/hensens26a.pdf}, url = {https://proceedings.mlr.press/v316/hensens26a.html}, abstract = {Pancreatic ductal adenocarcinoma (PDAC) remains one of the deadliest cancers due to late detection and limited treatment response. This study investigates the prognostic value of combining multiple AI-based image biomarkers—tumor-stroma ratio (TSR), mitosis density, stromal cell density (via HoVer-Net), tumor-to-tissue ratio from histopathological whole-slide images (WSIs) for survival prediction in resected PDAC patients. A multi-tissue segmentation model was developed to generate tissue masks for downstream biomarker extraction. Using logistic and Cox regression models, both univariate and multivariate survival analyses were performed across four datasets. Results show that while combining biomarkers did not outperform single-biomarker models (notably TSR), mitosis density showed consistent statistical significance and may serve as a valuable prognostic feature.} }
Endnote
%0 Conference Paper %T ContriMix: Scalable stain color augmentation for domain generalization without domain labels in digital pathology %A Lisa Hensens %A Sergio Sabroso-Lasa %A Caroline Verbeke %A Nuria Malats %A ThePanGenEU consortium %A Geert Litjens %A Pierpaolo Vendittelli %B Proceedings of the MICCAI Workshop on Computational Pathology %C Proceedings of Machine Learning Research %D 2026 %E Linda Studer %E Francesco Ciompi %E Nadieh Khalili %E Khrystyna Faryna %E Khrystyna Faryna %E Joe Yeong %E Mai Chan Lau %E Hao Chen %E Ziyi Liu %E Biagio Brattoli %F pmlr-v316-hensens26a %I PMLR %P 97--105 %U https://proceedings.mlr.press/v316/hensens26a.html %V 316 %X Pancreatic ductal adenocarcinoma (PDAC) remains one of the deadliest cancers due to late detection and limited treatment response. This study investigates the prognostic value of combining multiple AI-based image biomarkers—tumor-stroma ratio (TSR), mitosis density, stromal cell density (via HoVer-Net), tumor-to-tissue ratio from histopathological whole-slide images (WSIs) for survival prediction in resected PDAC patients. A multi-tissue segmentation model was developed to generate tissue masks for downstream biomarker extraction. Using logistic and Cox regression models, both univariate and multivariate survival analyses were performed across four datasets. Results show that while combining biomarkers did not outperform single-biomarker models (notably TSR), mitosis density showed consistent statistical significance and may serve as a valuable prognostic feature.
APA
Hensens, L., Sabroso-Lasa, S., Verbeke, C., Malats, N., consortium, T., Litjens, G. & Vendittelli, P.. (2026). ContriMix: Scalable stain color augmentation for domain generalization without domain labels in digital pathology. Proceedings of the MICCAI Workshop on Computational Pathology, in Proceedings of Machine Learning Research 316:97-105 Available from https://proceedings.mlr.press/v316/hensens26a.html.

Related Material