Framework for enhanced respiratory disease identification with clinical handcrafted features

Md. Ibrahim Patwary khokan, Tasnim Jahan Tonni, Md. Awlad Hossen Rony, Kaniz Fatema, Md. Zahid Hasan

Abstract: Highlights•Combined three chest X-ray datasets of six respiratory diseases, applying image preprocessing techniques to enhance quality.•Applied Albumentations augmentation and bitwise segmentation to address class imbalance and isolate ROIs within the images.•Extracted five clinical feature sets (GLCM, shape, LBP, lung, and pattern features) from images for detailed analysis.•Used KNN-based graph construction to convert tabular data to graphs and proposed CHXGNN, an optimized GNN with feature pruning.•Conducted active feature analysis with GNNExplainer to validate key nodes, edges, and attributes, clarifying the CHXGNN model’s predictions.

External IDs:doi:10.1016/j.compbiomed.2025.110588