AI-based Pipeline for Automated 3D Prosthetic Design Using Virtual Tooth Extraction and Alveolar Bone Remodeling Prediction

This study proposes an end-to-end digital dentistry pipeline that performs virtual tooth extraction on three-dimensional intraoral scans and cone-beam computed tomography data, predicts patient-specific post-extraction alveolar morphology over time, and generates automated prosthetic design candidates. Unlike existing workflows focused on tooth segmentation or healed-ridge assumptions before prosthetic planning, the framework unifies pre-extraction scanning, socket- preserving virtual extraction, temporal ridge prediction, uncertainty visualization, and prosthesis placement in one process. The method includes tooth instance segmentation, extraction socket boundary preservation, multimodal conditional shape prediction, pontic or implant design optimization, and clinician-in-the-loop review. Validation showed DSC/IoU of 0.962/0.928 for virtual extraction, and average DSC/MAE of 0.889/0.46 mm for temporal ridge prediction. Automated prosthetic designs achieved a mean geometric error of 0.31 mm and significantly reduced revision time (p<0.01). Blinded evaluation by five clinicians yielded scores of 4.42±0.51 for esthetics, 4.36±0.49 for functionality, and 4.48±0.43 for clinical applicability.