Prototype-driven fusion of pathology and spatial transcriptomics for interpretable survival prediction

Researchers introduced PathoSpatial, an AI framework that fuses whole-slide pathology images with spatial transcriptomics data to predict cancer patient survival. The model combines visual tissue patterns with precise molecular gene expression maps. Tested on triple-negative breast cancer, it matched or outperformed existing methods across five survival metrics. Crucially, it provides interpretable results, highlighting specific biological factors behind its predictions, moving beyond a 'black box' approach.