Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Survival Modeling from Whole Slide Images via Patch-Level Graph Clustering and Mixture Density Experts

Ardhendu Sekhar, Vasu Soni, Keshav Aske, Garima Jain, Pranav Jeevan, Amit Sethi

TL;DR

This work tackles cancer survival prediction from WSIs by introducing a four-module framework that jointly addresses prognostic tissue heterogeneity and spatial coherence. It combines Quantile-Based Patch Filtering, Graph-Regularized Patch Clustering, Hierarchical Feature Aggregation, and Expert-Guided Mixture Density Modeling to estimate complex, multi-modal survival distributions with interpretable components. Across TCGA LUAD, KIRC, and BRCA cohorts, the method achieves state-of-the-art time-dependent concordance indices ($\text{TDC} = 0.653$, $0.719$, $0.733$ respectively), demonstrating strong discrimination and calibration from histopathology alone. The approach emphasizes interpretability through cluster-aware representations and a multi-expert density model, offering practical potential for prognosis and risk stratification in personalized oncology.

Abstract

We propose a modular framework for predicting cancer specific survival directly from whole slide pathology images (WSIs). The framework consists of four key stages designed to capture prognostic and morphological heterogeneity. First, a Quantile Based Patch Filtering module selects prognostically informative tissue regions through quantile thresholding. Second, Graph Regularized Patch Clustering models phenotype level variations using a k nearest neighbor graph that enforces spatial and morphological coherence. Third, Hierarchical Feature Aggregation learns both intra and inter cluster dependencies to represent multiscale tumor organization. Finally, an Expert Guided Mixture Density Model estimates complex survival distributions via Gaussian mixtures, enabling fine grained risk prediction. Evaluated on TCGA LUAD, TCGA KIRC, and TCGA BRCA cohorts, our model achieves concordance indices of 0.653 ,0.719 ,and 0.733 respectively, surpassing existing state of the art approaches in survival prediction from WSIs.

Survival Modeling from Whole Slide Images via Patch-Level Graph Clustering and Mixture Density Experts

TL;DR

This work tackles cancer survival prediction from WSIs by introducing a four-module framework that jointly addresses prognostic tissue heterogeneity and spatial coherence. It combines Quantile-Based Patch Filtering, Graph-Regularized Patch Clustering, Hierarchical Feature Aggregation, and Expert-Guided Mixture Density Modeling to estimate complex, multi-modal survival distributions with interpretable components. Across TCGA LUAD, KIRC, and BRCA cohorts, the method achieves state-of-the-art time-dependent concordance indices (, , respectively), demonstrating strong discrimination and calibration from histopathology alone. The approach emphasizes interpretability through cluster-aware representations and a multi-expert density model, offering practical potential for prognosis and risk stratification in personalized oncology.

Abstract

We propose a modular framework for predicting cancer specific survival directly from whole slide pathology images (WSIs). The framework consists of four key stages designed to capture prognostic and morphological heterogeneity. First, a Quantile Based Patch Filtering module selects prognostically informative tissue regions through quantile thresholding. Second, Graph Regularized Patch Clustering models phenotype level variations using a k nearest neighbor graph that enforces spatial and morphological coherence. Third, Hierarchical Feature Aggregation learns both intra and inter cluster dependencies to represent multiscale tumor organization. Finally, an Expert Guided Mixture Density Model estimates complex survival distributions via Gaussian mixtures, enabling fine grained risk prediction. Evaluated on TCGA LUAD, TCGA KIRC, and TCGA BRCA cohorts, our model achieves concordance indices of 0.653 ,0.719 ,and 0.733 respectively, surpassing existing state of the art approaches in survival prediction from WSIs.

Paper Structure

This paper contains 17 sections, 16 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Quantile-Based Patch Filtering (QPF)
  5. Graph-Regularized Patch Clustering (GRPC)
  6. Hierarchical Feature Aggregation (HFA)
  7. Expert-Guided Mixture Density Modeling (EGMDM)
  8. Training Objective and Regularization
  9. Experimental Setup
  10. Datasets
  11. Training Configuration
  12. Results and Analysis
  13. Quantitative Results
  14. Ablation Studies and Interpretability
  15. Conclusion
  16. ...and 2 more sections

Figures (2)

  • Figure 1: (a) Proposed survival model. (b) Expert-guided Mixture Density Modeling block.
  • Figure 2: Kaplan–Meier survival plots illustrating distinct high and low-risk group separation across TCGA cohorts.