Automated Early Detection of Androgenetic Alopecia Using Deep Learning on Trichoscopic Images From a Korean Cohort: A Retrospective Model Development and Validation Study

The study developed a deep learning model using a ResNet-18 convolutional neural network to detect early-stage androgenetic alopecia (AGA) from trichoscopic images in a Korean clinical cohort. Using 318 images labeled by dermatologists, the model was validated internally with stratified 5-fold cross-validation and externally with a separate test set of 20 images. The model showed high performance, achieving an accuracy above 0.90 and an AUC above 0.93 in internal validation, and an accuracy of 0.90 and an AUC of 0.97 in external validation. The model demonstrated perfect recall for early-stage hair loss, indicating its potential as a reliable screening tool in clinical and teledermatology settings.