Deep Architectural Classification of Dental Pathologies Using Orthopantomogram Imaging

Artificial intelligence (AI), particularly deep learning algorithms utilizing convolutional neural networks, plays an increasingly pivotal role in enhancing medical image examination. It demonstrates the potential for improving diagnostic accuracy within dental care. Orthopantomograms (OPGs) are essential in dentistry; however, their manual interpretation is often inconsistent and tedious. To the best of our knowledge, this is the first comprehensive application of YOLOv5m for the simultaneous detection and classification of six distinct dental pathologies using panoramic OPG images. The model was trained and refined on a custom dataset that began with 232 panoramic radiographs and was later expanded to 604 samples. These included annotated subclasses representing Caries, Infection, Impacted Teeth, Fractured Teeth, Broken Crowns, and Healthy conditions. The training was performed using GPU resources alongside tuned hyperparameters of batch size, learning rate schedule, and early stopping tailored for generalization to prevent overfitting. Evaluation on a held-out test set showed strong performance in the detection and localization of various dental pathologies and robust overall accuracy. At an IoU of 0.5, the system obtained a mean precision of 94.22% and recall of 90.42%, with mAP being 93.71%. This research confirms the use of YOLOv5m as a robust, highly efficient AI technology for the analysis of dental pathologies using OPGs, providing a clinically useful solution to enhance workflow efficiency and aid in sustaining consistency in complex multi-dimensional case evaluations.