Non-invasive prediction of dupilumab response in facial atopic dermatitis using early post-treatment reflectance confocal microscopy features.

BACKGROUND: Non-invasive prediction of the efficacy of dupilumab on facial lesions in patients with atopic dermatitis (AD) was unresolved. OBJECTIVE: To explore reflectance confocal microscopy (RCM), a non-invasive method, for predicting dupilumab efficacy in treating facial AD lesions and to achieve effective prediction via deep learning. METHODS: 52 AD patients received dupilumab treatment between May 2023 and June 2025 were enrolled. Patients were categorized into 'responder' and 'non-responder' groups based on whether achieved 75% improvement from baseline in Eczema Area and Severity Index at week 16. RCM was utilized to evaluate facial lesions. Deep learning was employed to construct a prediction model for identifying intergroup differences. RESULTS: Dermal papillary dilation (AUC = 0.83, P < 0.001) and tortuous papillary capillary dilation (AUC = 0.81, P < 0.001) at week 4 correlated significantly with therapeutic non-response. The ResNet101-based deep learning model, trained on patients' RCM images of week 4 post-treatment scans, had a final test AUC of 0.796, with heatmaps illustrating its decision-making basis. LIMITATIONS: The cohort size is relatively small. CONCLUSION: Early post-treatment RCM features can effectively predict the efficacy of dupilumab in facial AD lesions combined with deep learning.