AI and the World's Hard Problems · Medicine & drug discovery · arXiv · 2026-10-20
ProtoGTX: Prototype-Aligned Graph Transformer for Explainable Whole-Slide Image Analysis
ProtoGTX is a graph-based framework for explaining whole-slide image predictions in computational pathology, linking patch-level evidence to morphological patterns. It uses a prototype vocabulary that is kept separate from the predictive model. On the CPTAC and BRACS datasets it outperformed the evaluated baselines, and on BRACS its prototype maps co-localized with pathologist annotations.