A Deep Learning-Based Automated Scoring System for Predicting Eyelid Rejuvenation Outcomes After Monopolar...

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Upper eyelid rejuvenation using monopolar radiofrequency (MRF) is a minimally invasive treatment for eyelid laxity.
• Current outcome evaluation relies on subjective methods, leading to variability and bias.
• Objective measures are needed for effective treatment planning and patient counseling.

Purpose of the Study:

• To develop and validate a deep learning (DL) system for automated scoring of clinical outcomes after eyelid MRF treatment.
• To provide objective, reproducible assessments of treatment efficacy.

Main Methods:

• A retrospective study involving 50 patients treated with eyelid MRF.
• A hybrid DL model combining Convolutional Neural Network (CNN) and U-Net for image segmentation and outcome scoring.
• Model performance validated against evaluations by board-certified dermatologists using RMSE and MAPE.

Main Results:

• The CNN-U-Net model demonstrated high accuracy, with a RMSE of 0.4 and MAPE of 0.08.
• Predicted scores closely correlated with expert dermatologist assessments.
• No significant differences in accuracy were found across patient age or sex subgroups.

Conclusions:

• The study proves the feasibility of a DL-based automated scoring system for eyelid MRF outcomes.
• This system offers objective, consistent, and reproducible evaluations, enhancing patient counseling and treatment planning.
• Further validation with larger datasets and longer follow-up is recommended.