AI-driven, real-time occlusion analysis for restorative dentistry

Thesis Abstract

This study addresses the critical need for objective, real-time assessment of occlusal contacts in restorative dentistry to reduce failure rates associated with improper tooth contacts and secondary wear. The aim is to develop and validate an AI-driven system that analyzes occlusion in real time during chairside procedures to guide restorative planning and execution. Specific objectives include (1) identifying quantifiable occlusal features from intraoral scans and real-time pressure sensors, (2) training a deep learning model to classify normal versus excessive or uneven occlusion during tooth-contact events, (3) integrating the model into a clinical workflow with a validated decision-support interface, and (4) evaluating clinical outcomes in terms of restoration longevity, chair time, and patient-reported comfort over a 12-month follow-up. The study adopts a mixed-methods design, combining quantitative performance evaluation with qualitative user feedback to ensure clinical relevance and usability. The population comprises adult patients undergoing adhesive or indirect restorations at a tertiary dental center, with a target sample size of 180 participants and a minimum of 320 occlusal assessments collected across pre-treatment, intraoperative, and post-operative stages. Data collection instruments include a multimodal sensing platform combining pressure-mressure sensors embedded in a custom occlusal splint, high-definition intraoral scanners for static geospatial data, and synchronized multi-angle video capture. The occlusal data are paired with clinical outcomes such as restoration success rates, marginal integrity, and post-operative sensitivity. The AI component employs a convolutional neural network (CNN) for feature extraction from sensor and imaging data, a recurrent neural network (RNN) to model temporal occlusal dynamics, and a fusion layer to integrate multimodal inputs. Model training uses 5-fold cross-validation on 80% of the dataset with hold-out testing on the remaining 20%, while performance metrics include accuracy, sensitivity, specificity, area under the receiver operating characteristic curve (AUC-ROC), and F1 scores. Statistical analyses include logistic regression to identify predictors of restoration failure and Cox proportional hazards models to assess time-to-failure outcomes, complemented by ANOVA to compare chair-time reductions across groups. The study integrates theoretical frameworks from Proxemics Theory in occlusal interaction and the Technology Acceptance Model (TAM) to interpret clinician adoption and workflow fit, with a conceptual model illustrating the relationships among real-time occlusal analytics, clinical decisions, and restoration outcomes. Anticipated findings indicate that the AI-driven system achieves >90% accuracy in detecting clinically significant occlusal discrepancies, reduces chair-time by an average of 12 minutes per procedure, and lowers restoration failure rates by 15–20% at 12 months compared with conventional methods. The research is expected to reveal that real-time feedback on occlusal balance enables more precise material deposition, enhanced marginal adaptation, and reduced postoperative sensitivity, thereby improving patient-reported outcomes. The study contributes to knowledge by (a) providing a validated, generalizable framework for real-time occlusion analytics in restorative dentistry, (b) demonstrating the feasibility and clinical value of multimodal AI integration in operative dentistry, and (c) offering evidence on how occlusal analytics influence decision-making and patient experience. The main conclusion posits that AI-driven real-time occlusion analysis can be seamlessly integrated into standard restorative workflows, yielding measurable improvements in both technical and patient-centered outcomes. Recommendations include expanding the sample to diverse populations across multiple centers to confirm generalizability, exploring integration with other digital dentistry modalities (e.g., CAD/CAM, augmented reality guidance), and developing standardized benchmarks for occlusal assessment to facilitate broader adoption and reimbursement considerations.