Intelligent Anterior Teeth Design Using 3D Point Cloud Completion Technology

Aim or purpose: To establish an intelligent anterior teeth design method based on three-dimensional point cloud completion technology, providing clinicians with rapid and accurate anterior teeth restoration design solutions. Materials and methods: The study involved 600 cases of 10 different dental defect types, divided into training, validation, and testing sets (10:1:1 ratio). Data preprocessing included 3D standardized sampling and tensor transformation before input into a neural network based on the PyTorch framework, utilizing a Transformer architecture as its core structure and implementing the Adam optimizer for dynamic learning rate adjustment. Model validation was conducted using 50 models, with efficacy assessed through 3D-deviation analysis against complete reference models. Clinical evaluation compared AI with traditional digital designs for 10 patient cases, using manual wax patterns as the reference standard. Results: A Transformer-based neural network was successfully constructed using PyTorch. Artificial defect evaluations showed no statistically significant differences across most anatomical regions (P>0.05). However, in clinical cases, AI designs showed greater deviations from wax patterns particularly in cervical margins compared to traditional digital designs, indicating limitations in capturing complex morphologies. Conclusions: This research created a dental scanning database and used Transformer networks to directly complete 3D point cloud data of anterior teeth defect models. Unlike traditional AI approaches using 2D depth images, this method avoids depth information loss by directly training on 3D data. Clinical testing shows efficiency and morphological restoration qualities. Though precision needs optimization, this technology offers promising clinical applications for efficient, personalized dental restorations.