Automated 3D facial smile attractiveness assessment before and after orthognathic surgery using transfer learning: A preliminary study

Abstract

The aesthetic appearance of the mouth during smiling significantly influences facial attractiveness, thereby, making smile analysis crucial in orthodontics, craniofacial surgery, and cosmetic dentistry. Accurate and quantitative evaluation of facial smile attractiveness is crucial for surgical planning and outcome assessment in orthognathic surgery (OGS). In this study, a transfer learning (TL) model using a convolutional neural network (CNN) based on three-dimensional (3D) contour line features was employed to assess facial smile attractiveness before and after OGS. A retrospective cohort study involving 135 patients was conducted between 2021 and 2024 to compare facial smile attractiveness before and after OGS. Using the 3dMD™ face system, 3D facial photos were captured in a natural head position with forward-facing eyes, relaxed facial muscles, and habitual dental occlusion, before and at least 6 months after surgery. Subsequently, 3D contour images were extracted from these photos for web-based automatic facial smile attractiveness assessment using TL with CNN model. Postoperatively, facial smile attractiveness significantly improved, with scores increasing from 2.62 to 3.27, representing a 25% enhancement as determined by the constructed machine learning model. The web-based system offered clinicians a user-friendly interface, providing rapid assessment of results and serving as an effective tool for doctor-patient communication. This study marks the first attempt to automatically evaluate facial smile attractiveness before and after surgery in an objective and quantitative manner, using a machine learning model based on the 3D contours feature map.

Keywords: Convolutional neural network (CNN); Facial smile attractiveness; Orthognathic surgery; Quantitative evaluation; Three-dimensional (3D) contour line features; Transfer learning (TL).