680-SMMILe: Spatial Quantification in Digital Pathology

Paper Talk · Mar 22, 2026 · 20m

This report introduces SMMILe, a novel computational pathology method designed to improve both whole-slide image classification and spatial quantification across various cancers. While traditional multiple-instance learning models often sacrifice spatial detail for global accuracy, SMMILe utilizes a superpatch-based approach to maintain high-resolution diagnostic maps without losing predictive power. The researchers mathematically demonstrate that instance-level aggregation overcomes the limitations of previous models, which often produce skewed or incomplete attention maps. By benchmarking…