Domain Generalization of Pathological Image Segmentation by Patch-Level and WSI-Level Contrastive Learning

TL;DR

This paper proposes a novel domain generalization method for pathological image segmentation that leverages intra-hospital domain shifts through clustering and contrastive learning at WSI and patch levels, improving robustness without multi-hospital data.

Contribution

It introduces a two-stage contrastive learning framework that captures intra-hospital domain shifts by clustering WSI features and applying contrastive learning to reduce feature gaps.

Findings

Effective reduction of domain gaps between WSIs from different clusters

Improved segmentation robustness across intra-hospital domain shifts

No need for multi-hospital data collection

Abstract

In this paper, we address domain shifts in pathological images by focusing on shifts within whole slide images~(WSIs), such as patient characteristics and tissue thickness, rather than shifts between hospitals. Traditional approaches rely on multi-hospital data, but data collection challenges often make this impractical. Therefore, the proposed domain generalization method captures and leverages intra-hospital domain shifts by clustering WSI-level features from non-tumor regions and treating these clusters as domains. To mitigate domain shift, we apply contrastive learning to reduce feature gaps between WSI pairs from different clusters. The proposed method introduces a two-stage contrastive learning approach WSI-level and patch-level contrastive learning to minimize these gaps effectively.