Artificial intelligence in dentistry: Training and testing a caries detection model with intraoral scans

Abstract Aim: The aim of this study was to develop and evaluate a deep learning-based model for the detection of occlusal caries in intraoral scans using YOLOv5. Materials and Methods: Intraoral scans were obtained from 117 patients, yielding 330 carious teeth (moderate and extensive lesions). Teeth were annotated using the International Caries Detection and Assessment System framework. YOLOv5 variants were trained with different preprocessing modules (single-channel, mixed channel, Ohta transformation). Performance was assessed using precision, recall, F1-score, mean average precision, and validated against expert annotations using Cohen’s kappa and McNemar’s test. Results: The best-performing YOLOv5 model achieved an accuracy of 90% in detecting moderate and extensive lesions. There was a significant amount of agreement with expert examiners ( κ =0.77). No discernible difference ( P = 1.0) between artificial intelligence (AI) predictions and human annotations according to McNemar’s test was found. Conclusion: YOLOv5-based detection of occlusal caries in intraoral scans demonstrated strong diagnostic performance, comparable to expert annotations. Targeted preprocessing further enhanced model reliability, underscoring the potential of intraoral scan-based AI diagnostics in conservative dentistry.