Semantic-Anchored Evidential Fusion for Domain-Robust Whole-Slide Survival Analysis

2026-06-18 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

The Semantic-Anchored Evidential Fusion Survival (SAEFS) framework addresses the challenge of domain generalization in computational cancer prognosis using whole-slide images (WSIs). Existing methods struggle across clinical centers due to their reliance on pixel-derived representations, which are highly susceptible to domain-specific artifacts from staining protocols and scanner hardware. SAEFS hypothesizes that high-level pathology semantics, such as tumor grade and micro-environmental architecture, provide domain-invariant representations mirroring human pathologist logic. The framework derives semantic anchors via Visual Question Answering (VQA), employs a dual-stream WSI evidence extraction architecture, uses Dirichlet-based Subjective Logic for uncertainty modeling, and fuses semantic and visual evidence through a cautious conjunction rule. Trained exclusively on one source domain and evaluated zero-shot across four unseen domains, SAEFS consistently outperformed state-of-the-art models, improving the average C-index by 10.2%. Quantitative analyses confirmed VQA-derived semantic features exhibit significantly lower cross-center divergence.

Key takeaway

For AI Scientists developing WSI-based cancer prognosis models, addressing cross-center generalization is critical. Your models can achieve significantly higher prediction accuracy and reliability by adopting the Semantic-Anchored Evidential Fusion Survival (SAEFS) framework. Consider integrating VQA-derived semantic features and evidential fusion techniques to mirror human pathologist diagnostic logic, enhancing robustness across diverse clinical domains.

Key insights

High-level pathology semantics provide domain-invariant representations crucial for robust whole-slide image analysis in cancer prognosis.

Principles

• Pathology semantics offer domain-invariant representations.
• Pixel-derived features are susceptible to domain artifacts.

Method

SAEFS uses VQA for semantic anchors, dual-stream WSI evidence extraction, Dirichlet-based Subjective Logic for uncertainty, and cautious conjunction for fusion.

In practice

• Improve WSI cancer prognosis across clinical centers.
• Integrate VQA for robust feature extraction.

Topics

• Whole-Slide Imaging
• Cancer Prognosis
• Domain Generalization
• Visual Question Answering
• Evidential Fusion
• Subjective Logic

Best for: AI Scientist, Research Scientist, Computer Vision Engineer

Related on AIssential

• Prototype-driven fusion of pathology and spatial transcriptomics for interpretable survival prediction — PathoSpatial fuses WSI and ST data via prototype learning for interpretable, spatially-informed survival prediction.
• Evidential learning driven Breast Tumor Segmentation with Stage-divided Vision-Language Interaction — TextBCS enhances breast tumor segmentation in MRI by combining text prompts with evidential learning to address low contrast and blurred boundaries.
• Multi-modal AI for comprehensive breast cancer prognostication — A multi-modal AI test significantly improves breast cancer prognostication by integrating pathology and clinical data.
• Geometry-Aware Superpixel Graph Transformer with Metadata for Skin Lesion Classification — Explicitly modeling skin lesions as geometry-aware superpixel graphs with integrated metadata improves classification accuracy.
• I Gave a Model Eyes and a Genome. Here’s What It Learned About Brain Cancer. — Cross-attention fusion of pathology and genomics significantly improves glioblastoma survival prediction over unimodal approaches.
• Learning Directional Semantic Transitions for Longitudinal Chest X-ray Analysis — ProTrans models disease progression in CXRs as directional semantic transitions using vision-language pretraining and radiology reports.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.