11/06/2025
🧬 We are pleased to share new research: CPGNet - Multimodal Graph Learning with Hierarchical Category Guidance for Multi-Label Whole Slide Image Classification.
This work addresses a key challenge in computational pathology: whole slide images (WSIs) often involve multiple labels (rather than just one), and class imbalance complicates the task.
Our proposed CPGNet model:
• Uses super-pixel segmentation (via MaskSLIC) to capture fine spatial structures in WSIs.
• Models these segments as graph nodes and constructs a graph representation of the slide.
• Employs a Graph Neural Network (GNN) + multi-head self-attention to capture both local and global dependencies in the tissue.
• Incorporates a module that uses hierarchical category relationships (“category-prompted”) to guide visual feature classification - improving accuracy on multi-label tasks.
This opens up improved possibilities for pathology image classification, better diagnostics, and ultimately better patient outcomes.
Read the full paper here: https://bit.ly/47jeUcn