Hierarchical Adaptation for Pathology Slide-Level Domain-Shift, MICCAI2025 Provisional Accept.
Jingsong Liu*, Han Li*, Chen Yang, Michael Deutges, Ario Sadafi, Xin You, Katharina Breininger, Nassir Navab, Peter J. Schüffler
Abstract: Domain shift is a critical problem for artificial intelligence (AI) in pathology as it is heavily influenced by center-specific conditions. Current pathology domain adaptation methods focus on image patches rather than whole-slide images (WSI), thus failing to capture global WSI features required in typical clinical scenarios. In this work, we address the challenges of slide-level domain shift by proposing a Hi erarchical Adaptation framework for Slide-level Domain-shift (HASD). HASD achieves multi-scale feature consistency and computationally ef ficient slide-level domain adaptation through two key components: (1) a hierarchical adaptation framework that integrates a Domain-level Align ment Solver for feature alignment, a Slide-level Geometric Invariance Regularization to preserve the morphological structure, and a Patch-level Attention Consistency Regularization to maintain local critical diagnos tic cues; and (2) a prototype selection mechanism that reduces com putational overhead. We validate our method on two slide-level tasks across five datasets, achieving a 4.1% AUROC improvement in a Breast Cancer HER2 Grading cohort and a 3.9% C-index gain in a UCEC sur vival prediction cohort. Our method provides a practical and reliable slide-level domain adaption solution for pathology institutions, mini mizing both computational and annotation costs.
- 18/06/2025: The HASD paper gets provisional acceptance by MICCAI2025.
- Submit to arxiv
- Clean the codes
- Update to git repo