Automated Trichoscopic Analysis for Hair Disorder Classification Using Semantic Segmentation

This study addresses the challenges of diagnosing hair disorders like Alopecia, Alopecia Areata, and Telogen Effluvium by proposing an automated trichoscopic image analysis framework. Utilizing a U-Net architecture for semantic segmentation, the framework analyzes 110 trichoscopic images to extract morphological and texture features. Significant intergroup differences were found using one-way ANOVA ($p<0.05$), supporting the clinical relevance of these features. Among various classifiers, Random Forest showed the best performance with 86.67% accuracy and a ROC-AUC of 93.38%. This framework offers an objective and reproducible method for scalp disorder classification, paving the way for advanced diagnostic systems.