Evaluating Foundation Models with Pathological Concept Learning for Kidney Cancer

2025-09-29

AI Summary

This study addresses survival risk stratification for renal cell carcinoma using histopathological whole-slide images (WSIs). Methodologically, it leverages foundation models to extract deep features from WSIs and constructs a pathology graph—where tissue regions serve as nodes and spatial adjacency relations as edges—then applies graph neural networks to model local–global pathological semantic relationships, enabling interpretable concept learning and survival prediction. Its key contribution lies in explicitly integrating structured clinical knowledge—including TNM staging and pathology reports—into the model architecture, thereby enhancing predictive transparency and decision fairness. Experimental validation across multiple independent renal cancer cohorts demonstrates robust discrimination between high- and low-risk patients (C-index: 0.72–0.78). The implementation is publicly available, underscoring its potential for clinical translation.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Computer Vision: Other Foundations of Computer Vision
• Reasoning under Uncertainty: Graphical Models

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

To evaluate the translational capabilities of foundation models, we develop a pathological concept learning approach focused on kidney cancer. By leveraging TNM staging guidelines and pathology reports, we build comprehensive pathological concepts for kidney cancer. Then, we extract deep features from whole slide images using foundation models, construct pathological graphs to capture spatial correlations, and trained graph neural networks to identify these concepts. Finally, we demonstrate the effectiveness of this approach in kidney cancer survival analysis, highlighting its explainability and fairness in identifying low- and high-risk patients. The source code has been released by https://github.com/shangqigao/RadioPath.

Problem

• Evaluating foundation models' translational capabilities for kidney cancer diagnosis
• Developing pathological concept learning using TNM staging and pathology reports
• Applying graph neural networks to improve cancer survival analysis and risk stratification

Innovation

• Leveraging TNM staging for pathological concept learning
• Constructing spatial graphs from whole slide images
• Training graph neural networks for survival analysis