Dual-stream multi-dependency graph neural network enables precise cancer survival analysis

Area of Science:

• Computational pathology
• Artificial intelligence in oncology
• Biomedical image analysis

Background:

• Histopathology image analysis is crucial for cancer prognosis and personalized treatment.
• Current methods struggle to capture complex correlations between diverse image patches in whole slide images (WSIs).
• This limitation hinders accurate patient status inference and survival prediction.

Purpose of the Study:

• To develop a novel deep learning framework for precise cancer patient survival analysis using histopathology images.
• To address the limitations of existing methods in modeling inter-patch correlations within WSIs.
• To improve the accuracy and interpretability of cancer prognosis.

Main Methods:

• Proposed a dual-stream multi-dependency graph neural network (DM-GNN) framework.
• Modeled WSIs as two graphs based on morphological affinity and global co-activating dependencies.
• Introduced an affinity-guided attention recalibration module for robust dependency utilization.

Main Results:

• DM-GNN demonstrated superior performance compared to state-of-the-art methods on five TCGA datasets.
• The framework effectively models complex correlations between image patches.
• Achieved interpretable prediction insights linked to high-attention patch morphology.

Conclusions:

• DM-GNN offers a powerful tool for personalized cancer prognosis from histopathology images.
• The framework has the potential to assist clinicians in treatment decision-making.
• Improved prognostic accuracy can lead to better patient outcomes.