Predicting postrestorative facial appearance in edentulous patients using deep learning: A prospective cohort study.

STATEMENT OF PROBLEM: Postrestorative facial appearance in edentulous patients remains unpredictable because of complex soft tissue dynamics, posing challenges for prosthodontic treatment planning and patient counseling. Traditional methods lack precision in simulating esthetic outcomes critical for psychological well-being and clinical success. PURPOSE: The purpose of this study was to develop FacePointNet, a bidirectional deep learning model, to predict postrestorative facial changes in edentulous patients after dental restoration, enhancing prerestorative planning through quantitative metrics and expert-rated visual similarity scores (VSSs). MATERIAL AND METHODS: Sixteen edentulous patients underwent pre- and postrestorative 3-dimensional (3D) facial scans. FacePointNet, a point-set neural network with dual registration subnetworks, learned bidirectional geometric transformations using composite loss functions (geometric and consistency losses). Performance was evaluated via 4-fold cross-validation, with metrics including chamfer distance (CD), Euclidean distance (ED), and VSS by 3 oral implant surgeons. The Wilcoxon signed-rank test was used to compare Pre-Post with Pre-Predicted distributions and was selected because of nonnormal data residuals. RESULTS: Artificial intelligence (AI)-predicted 3D morphology matched restorative outcomes with mean landmark errors of 3.80 to 5.98 mm (nasal tip: 2.17 mm; left commissure: 6.88 mm). No significant differences were found between the predicted and actual results (P>.05). Expert reviewers reported strong visual concordance (VSS: 4.2/5), particularly in nasolabial folds. CONCLUSIONS: FacePointNet demonstrated clinical utility for the prerestorative visualization of localized facial changes. Future work should expand datasets, integrate biomechanical data, and refine dynamic modeling to enhance global prediction robustness, advancing AI-driven precision prosthodontics.