Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DPsurv: Dual-Prototype Evidential Fusion for Uncertainty-Aware and Interpretable Whole-Slide Image Survival Prediction

Yucheng Xing, Ling Huang, Jingying Ma, Ruping Hong, Jiangdong Qiu, Pei Liu, Kai He, Huazhu Fu, Mengling Feng

TL;DR

DPsurv presents a dual-prototype evidential fusion network for whole-slide image survival prediction that jointly models tissue heterogeneity and predictive uncertainty. By mapping patch embeddings to component prototypes via a patch-prototype GMM and then translating component embeddings into evidential GRFNs, it yields uncertainty-aware survival intervals through a mixture of GRFNs and provides interpretable reasoning paths from histology to risk. The approach achieves state-of-the-art discrimination and calibration across five TCGA cohorts and delivers multi-level explanations, including patch-level assignments, component-level risks, and quantified uncertainty, enhancing trustworthiness in clinical prognostication. This framework offers practical impact for precision prognosis and facilitates new pathology-driven insights by linking morphological phenotypes to survival risk.

Abstract

Pathology whole-slide images (WSIs) are widely used for cancer survival analysis because of their comprehensive histopathological information at both cellular and tissue levels, enabling quantitative, large-scale, and prognostically rich tumor feature analysis. However, most existing methods in WSI survival analysis struggle with limited interpretability and often overlook predictive uncertainty in heterogeneous slide images. In this paper, we propose DPsurv, a dual-prototype whole-slide image evidential fusion network that outputs uncertainty-aware survival intervals, while enabling interpretation of predictions through patch prototype assignment maps, component prototypes, and component-wise relative risk aggregation. Experiments on five publicly available datasets achieve the highest mean concordance index and the lowest mean integrated Brier score, validating the effectiveness and reliability of DPsurv. The interpretation of prediction results provides transparency at the feature, reasoning, and decision levels, thereby enhancing the trustworthiness and interpretability of DPsurv.

DPsurv: Dual-Prototype Evidential Fusion for Uncertainty-Aware and Interpretable Whole-Slide Image Survival Prediction

TL;DR

DPsurv presents a dual-prototype evidential fusion network for whole-slide image survival prediction that jointly models tissue heterogeneity and predictive uncertainty. By mapping patch embeddings to component prototypes via a patch-prototype GMM and then translating component embeddings into evidential GRFNs, it yields uncertainty-aware survival intervals through a mixture of GRFNs and provides interpretable reasoning paths from histology to risk. The approach achieves state-of-the-art discrimination and calibration across five TCGA cohorts and delivers multi-level explanations, including patch-level assignments, component-level risks, and quantified uncertainty, enhancing trustworthiness in clinical prognostication. This framework offers practical impact for precision prognosis and facilitates new pathology-driven insights by linking morphological phenotypes to survival risk.

Abstract

Pathology whole-slide images (WSIs) are widely used for cancer survival analysis because of their comprehensive histopathological information at both cellular and tissue levels, enabling quantitative, large-scale, and prognostically rich tumor feature analysis. However, most existing methods in WSI survival analysis struggle with limited interpretability and often overlook predictive uncertainty in heterogeneous slide images. In this paper, we propose DPsurv, a dual-prototype whole-slide image evidential fusion network that outputs uncertainty-aware survival intervals, while enabling interpretation of predictions through patch prototype assignment maps, component prototypes, and component-wise relative risk aggregation. Experiments on five publicly available datasets achieve the highest mean concordance index and the lowest mean integrated Brier score, validating the effectiveness and reliability of DPsurv. The interpretation of prediction results provides transparency at the feature, reasoning, and decision levels, thereby enhancing the trustworthiness and interpretability of DPsurv.

Paper Structure

This paper contains 36 sections, 40 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Method
  4. Preliminaries
  5. Deep Slide Component Embedding
  6. Component Evidence Modeling
  7. Component Evidence Mixture
  8. Mixture evidential loss
  9. Experiments
  10. Experimental setup
  11. Survival accuracy
  12. Survival Interpretability
  13. Survival uncertainty
  14. Conclusion
  15. Proof
  16. ...and 21 more sections

Figures (3)

  • Figure 1: Overview of the DPsurv framework Deep Slide Component Embedding encodes WSI into deep feature embeddings with patch prototypes; Component Evidence Modeling maps deep embeddings into component evidence through component prototypes, and Component Evidence Mixture aggregates component evidence into transformed survival functions, illustrated by the Plausibility (blue dashed line) and Belief (orange dashed line) curves.
  • Figure 2: Interpretation of the DPsurv in WSI survival prediction (A) Visualization of the assignment map, prototype distribution, and morphological annotations provided by a board-certified pathologist. (B) Visualization of component prototypes and the reasoning process for component evidence modeling. (C) Decision making with component-wise relative risk and its distribution over the WSI and region of interest (ROI).
  • Figure 3: Calibration plots of DPsurv across five TCGA datasets, showing the proportion of $\alpha$-level BPIs and PPIs that contain the uncensored survival times, for $\alpha \in \{0.1,\dots, 0.9\}$.