Real-Time Detection Of Dental Treatments In Panoramic Radiographs

This study compares three deep learning models—YOLOv11, YOLOv12, and ResNet-50—for detecting seven key dental conditions: fillings, root canal-treated teeth, crowns, bridges, caries, posts, and implants. A curated dataset of orthopantomograms (OPGs) was annotated using CVAT and preprocessed through resizing, normalization, flipping, rotation, and color jitter. The models were trained using transfer learning, regularization techniques, and class-weighted loss functions. Model performance was evaluated using metrics such as precision, recall, F1-score, mean Average Precision (mAP), and inference speed. YOLOv12 outperformed all other models with an mAP of 87.2%, precision of 90.5%, and recall of 88.3%, while maintaining real-time inference capability. YOLOv11 showed slightly lower performance but still demonstrated strong potential for clinical use. Although ResNet-50 achieved high classification accuracy, its lack of object localization limited its utility for spatially dependent tasks. Overall, YOLOv12 demonstrated the best balance between accuracy and speed, making it highly suitable for real-time clinical integration. This model can support automated diagnosis of crowns, fillings, implants, and caries, thereby improving clinical outcomes and reducing practitioner workload. The study highlights the promise of AI in dental diagnostics and encourages future research involving expanded datasets and broader detection of multiple conditions.