Ai-Powered Dental Defect Recognition For Removable Partial Denture Design

This study proposes a novel algorithm that integrates computer vision-based image understanding technology with Kennedy classification rules to enhance the scientific validity and reliability of the model, thereby establishing a foundation for a removable partial denture (RPD) computer-aided design system. A dataset of 1300 images of patients' oral dentition images was collected from the Stomatological Hospital of China Medical University and expanded to 3470 images through image enhancement techniques for model training. An innovative comparison strategy was implemented to evaluate the performance of various segmentation models, with results quantitatively analyzed. The YOLOv11-seg model demonstrated robust performance in the dental image segmentation task, achieving a mean average precision (seg_mAP) of 98.2%. It effectively identified most dental defects and accurately classified them according to Kennedy classification rules, validating its clinical validity and strong generalization capability. YOLOv11-seg is a highly effective model that provides prosthodontists with a reliable basis for formulating subsequent RPD design plans. Its integration of computer vision and Kennedy classification principles advances precision in prosthodontic diagnostics and treatment planning.