Discrepancy Minimization Improves Cross-Hospital Robustness in Digital Pathology
For pathologists and AI researchers deploying models across hospitals, this method improves robustness without requiring target hospital labels, though improvements are incremental over existing domain adaptation techniques.
The paper addresses the cross-hospital robustness problem in digital pathology by fine-tuning pathology foundation models with a local maximum mean discrepancy (LMMD) objective, achieving consistent improvements in both domain adaptation and domain generalization settings across multiple models and tasks.
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.