Discrepancy Minimization Improves Cross-Hospital Robustness in Digital Pathology

2026-05-24

AI Summary

This work addresses the significant performance degradation of pathology foundation models when deployed across hospitals due to distributional shifts in data. To mitigate this issue, the study introduces Local Maximum Mean Discrepancy (LMMD) into digital pathology for the first time, leveraging it during fine-tuning to align cross-domain representations. The proposed approach unifies domain adaptation and domain generalization within a single framework and enhances model generalization without requiring labeled data from target hospitals. Evaluated across multiple pathology foundation models and tasks, the method consistently yields performance gains at both patch and whole-slide levels, substantially improving robustness in multi-center deployment scenarios.

Abstract

Pathology foundation models (PFMs) have advanced rapidly in recent years and support training classifiers for a range of histopathology tasks. However, their robustness across hospitals remains limited: performance often degrades when training a classifier on data from one hospital and evaluating it on another target hospital. We address this challenge by fine-tuning PFMs with a local maximum mean discrepancy (LMMD) objective that applies to two settings: domain adaptation, where unlabeled target-hospital data is available, and domain generalization, where target-hospital data is unavailable at all. Experiments at both the patch- and slide-level show consistent improvements across multiple PFMs and tasks.

Problem

• cross-hospital robustness
• domain adaptation
• domain generalization
• digital pathology
• discrepancy minimization

Innovation

• Local Maximum Mean Discrepancy
• Domain Adaptation
• Domain Generalization
• Pathology Foundation Models
• Cross-Hospital Robustness