Cross-Domain Validation of a Resection-Trained Self-Supervised Model on Multicentre Mesothelioma Biopsies

Farzaneh Seyedshahi, Francesca Damiola, Sylvie Lantuejoul, Ke Yuan, John Le Quesne

Keywords: Mesothelioma, Biopsy, Self-Supervised Learning, Histopathology

TL;DR: We evaluate whether a self-supervised model trained on mesothelioma resections can generalise robustly to multicentre biopsy data under significant domain shift.

Abstract: Accurate subtype classification and outcome prediction in mesothelioma are essential for guiding therapy and predicting patient outcomes. However, most computational pathology models are trained exclusively on large tissue images from resection specimens, which limits their relevance in real-world diagnostic settings where small biopsies are the primary tissue source. Here, we assess the biopsy-level generalisability of a self-supervised encoder using a large, multicentre French cohort. We identify 53 biopsy-specific histomorphological clusters, quantify each patient’s proportional representation across these clusters, and use these profiles as inputs to two downstream tasks: (i) survival prediction using a Cox proportional hazards model and (ii) subtype classification (epithelioid vs. non-epithelioid) using logistic regression. The survival model achieved a test C-index of 0.6 and robustly separated cohort patients into high- and low-risk groups ($ p = 3.96 \times 10^{-29} $). For subtype classification, the logistic model reached an average AUC of 0.92. These results demonstrate that a self-supervised encoder trained on resection tissue can be reliably transferred to biopsy material despite significant domain shifts. The resulting biopsy-level morphological atlas enables clinically meaningful survival stratification and subtype prediction, supporting the translational integration of AI-driven decision tools in mesothelioma diagnostics.

Primary Subject Area: Transfer Learning and Domain Adaptation

Secondary Subject Area: Application: Histopathology

Registration Requirement: Yes

Reproducibility: https://github.com/FarzanehSeyedshahi/Histomorphological-Phenotype-Learning

Visa & Travel: Yes

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Submission Number: 21