BaldGraphFormer: An Explainable Deep Learning Framework for Early Baldness Prediction by Integrating Swin Transformer, Graph Attention Networks, and Clinical Features

BaldGraphFormer is a deep learning framework designed to improve early detection of androgenetic alopecia (AGA) by integrating scalp-image features and clinical variables. It combines a Swin Transformer for image analysis and a Graph Attention Network for clinical data, with a fusion layer for classification. The model provides interpretability through visual and clinical data mapping, aiding clinicians in tracing predictions to specific factors. In tests, BaldGraphFormer outperformed existing models, achieving a 97.62% F1-score and a 0.992 macro-average AUC, showing significant improvements over unimodal baselines. This framework could enhance dermatological decision-making and patient care for hair loss disorders.