CRISP -- Clustering-Based Redundancy-Reduced Instance Sampling for Pathology Case Representation and Retrieval
Researchers have developed CRISP, an unsupervised framework designed to process multiple whole-slide images (WSIs) for digital pathology cases. This method constructs comprehensive case-level representations by intelligently selecting informative patches across all available slides, thus avoiding the limitations of relying on a single pathologist-chosen slide. CRISP first reduces redundancy within individual WSIs and then uses clustering to select a compact, representative set of patches that capture case-level heterogeneity. This approach has demonstrated effectiveness in patient/case search and retrieval for diagnosis and treatment planning, potentially unlocking clinically relevant information previously overlooked. AI
IMPACT Automates the analysis of multiple pathology slides, potentially improving diagnostic accuracy and treatment planning by leveraging overlooked data.