Related Experiment Video
Updated: Jul 17, 2026
Introduction of an Integrated Pathology Image Management, Artificial Intelligence, and Reporting System
Published on: July 11, 2025
Adapting pathology foundation models for continual cross-center WSI retrieval
Xinyu Zhu1, Zhiguo Jiang1, Kun Wu1
1Beijing Advanced Innovation Center for Biomedical Engineering, School of Engineering Medicine, Beihang University, Beijing, 100191, China; Image Processing Center, School of Astronautics, Beihang University, Beijing, 100191, China; Tianmushan Laboratory, Hangzhou, 311115, China.
This study introduces a new continual learning framework to adapt foundation models for cross-center whole slide image retrieval. The method effectively addresses domain shifts, improving retrieval accuracy in multi-center medical imaging.
Area of Science:
- Computational pathology
- Digital pathology
- Machine learning in healthcare
Background:
- Medical centers generate vast whole slide images (WSIs) for histopathological analysis.
- Content-based histopathological image retrieval (CBHIR) leverages digital WSIs for morphologic content analysis.
- Foundation models show promise for CBHIR but face challenges like cross-center domain shifts and feature drift.
Purpose of the Study:
- To develop a continual learning framework for adapting pathology foundation models for continual cross-center WSI retrieval (CCBHIR).
- To mitigate domain discrepancies and enable stable, efficient feature rehearsal for long-term model adaptation.
Main Methods:
- A novel continual learning framework aligning foundation model outputs into a unified latent domain using instance-wise prompts.
- An embedding consistency replay mechanism for stable feature rehearsal without index rebuilding.
- Evaluation on a large-scale dataset of 10,837 WSIs from TCGA projects.
Main Results:
- The proposed framework significantly improves both intra-center and cross-center retrieval performance.
- Demonstrated superior performance compared to state-of-the-art continual learning methods.
- Preserved forward and backward retrieval compatibility across different medical centers.
Conclusions:
- The framework offers an effective strategy for deploying pathology foundation models in real-world, multi-center scenarios.
- Bridges the gap between foundation model research and practical computational pathology applications.
- Facilitates robust and adaptable WSI retrieval in diverse clinical settings.
