Abstract
Purpose: This study aimed to develop an artificial intelligence (AI) system using a convolutional neural network (CNN) to design major connectors for removable partial denture (RPD) frameworks. Unlike rule-based or case-matching expert systems, the proposed model automatically generated RPD designs from comprehensive oral data using gradient-weighted class activation mapping (Grad-CAM) enhancing explainability.
Methods: Data were obtained from 1000 RPDs (457 maxillary, 543 mandibular) designed by prosthodontic specialists, including 255 dental variables from oral examinations. Ten CNN models were constructed sequentially to predict denture type (resin/metal), unilateral/bilateral design, absence/presence of a major connector, and four and two connector types for the maxilla and mandible, respectively. This sequential prediction followed clinical logic. Model performance was assessed by accuracy, sensitivity, specificity, positive and negative predictive values, and F1 score. Grad-CAM was employed to visualize input factors influencing predictions.
Results: The CNN models predicted major connector types with high accuracy (maxilla: palatal bar/strap, 93.0%; palatal plate, 87.1%; horseshoe bar, 89.5%; and mandible: lingual bar or plate, 87.3%). Heatmaps highlighted clinically relevant features including periodontal condition, dentition defects, and occlusal relationships as key determinants.
Conclusions: This study demonstrated the feasibility of an AI-based system for predicting RPD major connector types with acceptable performance. The system is expected to provide valuable support to dentists by minimizing the variability owing to differences in clinical skills and experience. Grad-CAM visualization enhances interpretability, enabling both clinicians and patients to better understand the reasoning behind the RPD design process. Ultimately, this approach could improve consistency, efficiency, and training in prosthodontics.
