DPsurv: Dual-Prototype Evidential Fusion for Uncertainty-Aware and Interpretable Whole-Slide Image Survival Prediction
Researchers have developed DPsurv, a novel network designed for survival prediction using pathology whole-slide images. This method addresses limitations in interpretability and predictive uncertainty common in existing approaches. DPsurv utilizes a dual-prototype evidential fusion technique to provide uncertainty-aware survival intervals and offers interpretability through patch prototype assignment maps and component prototypes. Experiments across five datasets demonstrate DPsurv's superior performance in concordance index and integrated Brier score, highlighting its reliability and transparency. AI
IMPACT Introduces a more interpretable and uncertainty-aware method for survival prediction in pathology, potentially improving diagnostic reliability.