This repository provides code and data for the analysis of synovial tissue architecture in knee osteoarthritis (OA) using a combination of deep learning–based semantic segmentation and graph neural network (GNN) modeling. We developed a computational pathology framework to analyze hematoxylin and eosin (H&E)-stained whole-slide images (WSIs) of human synovium. The pipeline integrates:
- Semantic segmentation of histological structures
- Spatial analysis of tissue organization
- Graph neural network–based modeling of microenvironmental interactions
Directory: HE_semantic_segmentation/
- Performs semantic segmentation of H&E-stained WSIs
- Extracts histological structures such as:
- Synovial lining
- Immune cell infiltration
- Microvessels
- Fibrotic regions
- Outputs spatial label maps used for downstream analysis
Directory: GNN/
- Constructs SLIC-based graphs from segmentation outputs
- Models spatial relationships between tissue components
Directory: Postprocess/
- Performs downstream analysis based on semantic segmentation and GNN outputs
- Computes quantitative metrics and spatial features
- Generates figures used in the manuscript