Abstract
BACKGROUND: Upper eyelid rejuvenation with monopolar radiofrequency (MRF) is a minimally invasive option for patients with eyelid laxity. However, outcomes vary widely, and conventional evaluation methods rely on subjective photographic assessment and physician judgment, which are prone to observer bias and limited reproducibility. This lack of standardized, objective outcome measures complicates treatment planning and patient counseling. OBJECTIVE: To develop and validate a deep learning-based automated scoring system for predicting and assessing clinical outcomes following eyelid MRF treatment. METHODS: A retrospective, multicenter study of 50 patients (47 women, 3 men) treated with eyelid MRF was conducted. Pre- and post-treatment images were used to train a hybrid model combining a convolutional neural network (CNN) and U-Net architecture. The U-Net performed periorbital segmentation, while the CNN generated quantitative improvement scores. Ground truth ratings were provided by five board-certified dermatologists. Model performance was evaluated using root mean square error (RMSE) and mean absolute percentage error (MAPE). RESULTS: The CNN-U-Net model achieved a RMSE of 0.4 and a MAPE of 0.08, with predicted scores closely aligning with dermatologist evaluations. No significant differences in predictive accuracy were observed across patient age or sex subgroups. CONCLUSION: This proof-of-concept study demonstrates the feasibility of an automated deep learning-based scoring system for eyelid MRF outcomes. By providing objective, consistent, and reproducible evaluations, the system has the potential to enhance patient counseling, guide individualized treatment planning, and enable standardized research comparisons across clinics. Larger and more diverse datasets with longer follow-up are needed for further validation.