. Hierarchical Cell-to-Patch Graphs for Context-Aware Cell Classification in Digital Pathology. In 21st European Congress on Digital Pathology. Barcelona: The European Society of Digital and Integrative Pathology; 2025. 
(600.71 KB)
(600.71 KB) Abstract
Cell classification in digital pathology requires spatial context beyond individual cell morphology, particularly for distinguishing tumor vs. normal cells in H&E-stained images. While traditional segmentation and models like CellNuc-DETR achieve high detection accuracy, their reliance on small image patches limits classification performance by excluding tissue-level context. We propose a hierarchical cell-to-patch graph framework that integrates fine-grained cell features with global spatial information using graph neural networks (GNNs) to enhance classification accuracy while maintaining computational efficiency.
