CRISP: Unsupervised Framework for Multi-WSI Pathology Case Analysis

Researchers have introduced CRISP (Clustering-Based Redundancy-Reduced Instance Sampling for Pathology), an unsupervised framework for case-level analysis in digital pathology that integrates information from all available whole-slide images (WSIs) within a case. Current methods typically rely on a single pathologist-selected slide, discarding potentially informative evidence from other slides that capture spatially distinct tumour regions and morphological heterogeneity. CRISP operates in two stages: first reducing redundancy within individual WSIs, then constructing case-level representations by selectively distilling informative patches across WSIs. The framework addresses the lack of autonomous multi-WSI case processing, enabling comprehensive representation without requiring pathologist input. The work is detailed in a preprint on arXiv (2605.24253).