An interpretable cell-centric representation of pancreatic ductal adenocarcinoma matches multiple-instance learning for clinical outcomes and leads to novel pathologist-driven histological biomarkers
- 1Centre Hospitalier de l'Université de Montréal
Abstract
Current state of the art computational pathology foundation models attain satisfying accuracy most general pathology tasks including detection of pancreatic ductal adenocarcinoma (PDAC) but fail to provide sufficient explainability, operating as "black boxes". Some interpretability effort has been made, with models providing attention maps that project back on the slide the models high attended areas that pushed for the given prediction result, without further naming or reasoning explanation which can limit pathologists trust toward the prediction. We developed C3PRO, a cell-centric framework attempting to provide deeper understanding and explainability, by using embeddings extracted using a pathology foundation model, from cell centered tiles containing both the cell and its close neighborhood, which were then clustered into recurrent morphological phenotypes. A spatial graph was additionally constructed on top of the back-projected vocabulary to account for recurrent motifs. The evaluation of the model performance involved a cohort of 809 patients (825 diagnostic biopsies) using nested cross validation. C3PRO was benchmarked against four different multiple instance learning frameworks and achieved on-par performance for diagnostic (0.958 AUC), survival (0.617) and resectability (0.696) prediction, while providing additional insights compared to the strongest baseline models. Each clustered phenotype was probed and extracted as a representative tile mosaic and annotated by two expert pathologists, constituting a phenotype vocabulary available for statistical analysis. These phenotypes were scored by pathologists with physical brightfield microscopy and replicated the independent statistical significance impact on overall survival prediction from diagnostic biopsies.
STATEMENT OF SIGNIFICANCEPathology foundation model multiple-instance-learning frameworks cannot be audited by the pathologists. We demonstrate that a representation constructed from identifiable cell phenotypes is on par with state-of-the-art models for diagnosis, prognosis, and resectability in pancreatic cancer. This approach also uncovers prognostic stromal states that expert observers cannot differentiate using light microscopy.
Related
- Borderline resectable pancreatic ductal adenocarcinoma: a narrative review
- Combined halp and GRIm scores for risk stratification in metastatic pancreatic ductal adenocarcinoma
- Exceptionally Durable Response to First-Line S-1 Monotherapy in Metastatic Pancreatic Adenocarcinoma: A Case Report
- Prognostic and Spatial Organization of Tertiary Lymphoid Structures in Pancreatic Ductal Adenocarcinoma
- Identification of miR-615-5p/ID1 axis crucial in the pathogenesis of pancreatic ductal adenocarcinoma (PDAC)
BibTeX
@misc{pelloux2026interpretable,
title = {{An interpretable cell-centric representation of pancreatic ductal adenocarcinoma matches multiple-instance learning for clinical outcomes and leads to novel pathologist-driven histological biomarkers}},
author = {M. Pelloux and C. T. Tran and J. Baig and A. Khellaf and S. L. Huang and O. T. Nguyen and M. Batardiere and S.-S. Diwan and Z. Yang and J.-L. Li and V. Vu and S. F. Roy and G. Soucy and B. N. Nguyen and B. Tessier-Cloutier and M. S. Hosseini and D. J. Knapp and S. Valmary-Degano and V. Q. Trinh},
howpublished = {medRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.03.26364596},
url = {https://www.medrxiv.org/content/10.64898/2026.10.03.26364596v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - An interpretable cell-centric representation of pancreatic ductal adenocarcinoma matches multiple-instance learning for clinical outcomes and leads to novel pathologist-driven histological biomarkers
AU - M. Pelloux
AU - C. T. Tran
AU - J. Baig
AU - A. Khellaf
AU - S. L. Huang
AU - O. T. Nguyen
AU - M. Batardiere
AU - S.-S. Diwan
AU - Z. Yang
AU - J.-L. Li
AU - V. Vu
AU - S. F. Roy
AU - G. Soucy
AU - B. N. Nguyen
AU - B. Tessier-Cloutier
AU - M. S. Hosseini
AU - D. J. Knapp
AU - S. Valmary-Degano
AU - V. Q. Trinh
PY - 2026
DA - 2026/10/03
PB - medRxiv
DO - 10.64898/2026.10.03.26364596
UR - https://www.medrxiv.org/content/10.64898/2026.10.03.26364596v1
AB - Current state of the art computational pathology foundation models attain satisfying accuracy most general pathology tasks including detection of pancreatic ductal adenocarcinoma (PDAC) but fail to provide sufficient explainability, operating as "black boxes". Some interpretability effort has been made, with models providing attention maps that project back on the slide the models high attended areas that pushed for the given prediction result, without further naming or reasoning explanation which can limit pathologists trust toward the prediction. We developed C3PRO, a cell-centric framework attempting to provide deeper understanding and explainability, by using embeddings extracted using a pathology foundation model, from cell centered tiles containing both the cell and its close neighborhood, which were then clustered into recurrent morphological phenotypes. A spatial graph was additionally constructed on top of the back-projected vocabulary to account for recurrent motifs. The evaluation of the model performance involved a cohort of 809 patients (825 diagnostic biopsies) using nested cross validation. C3PRO was benchmarked against four different multiple instance learning frameworks and achieved on-par performance for diagnostic (0.958 AUC), survival (0.617) and resectability (0.696) prediction, while providing additional insights compared to the strongest baseline models. Each clustered phenotype was probed and extracted as a representative tile mosaic and annotated by two expert pathologists, constituting a phenotype vocabulary available for statistical analysis. These phenotypes were scored by pathologists with physical brightfield microscopy and replicated the independent statistical significance impact on overall survival prediction from diagnostic biopsies. STATEMENT OF SIGNIFICANCEPathology foundation model multiple-instance-learning frameworks cannot be audited by the pathologists. We demonstrate that a representation constructed from identifiable cell phenotypes is on par with state-of-the-art models for diagnosis, prognosis, and resectability in pancreatic cancer. This approach also uncovers prognostic stromal states that expert observers cannot differentiate using light microscopy.
ER -