Determination of dental treatments in panoramic radiographs taken from children using an artificial intelligence system

Dental treatment may be required in pediatric patients due to caries, trauma, or developmental anomalies. If untreated, such problems may lead to functional loss and negatively affect jaw and facial development. Therefore, identifying the dental condition accurately and planning appropriate treatment is important. Panoramic radiographs are often used for overall dental assessment in children. Recently, AI-based systems have been developed to detect teeth and identify treatments from radiographs, and this study aimed to automatically classify different treatment needs—such as filling, pulpotomy, root canal therapy, extraction, and space maintainer—using the YOLOv8x deep learning model. In our study, 1,061 panoramic radiographs from patients aged 6 to 10 were used. An artificial intelligence algorithm was developed to identify five treatment needs: filling, pulpotomy, extraction, root canal treatment, and space maintainer. During deep learning, the YOLOv8x architecture was applied to detect the dental conditions. Data labeling was performed using CranioCatch, a web-based AI-supported software, and 8,552 labels were obtained. The model's performance was evaluated by calculating precision, recall, and F1-score, and mAP@50 was used to indicate overall accuracy. According to the results, the F1 score was 79.9% for extraction, 74.2% for space maintainer, 73.5% for restorative filling, and 70.1% for root canal treatment. The lowest F1 score, 51.4%, was observed in pulpotomy. Recall scores were 86.8%, 82.3%, 79.7%, and 73.8% for space maintainer, extraction, filling, and root canal, respectively. Pulpotomy recall was 52.9%. Precision was 77.6% for extraction, 68.2% for filling, 66.7% for root canal, and 64.7% for space maintainer. Precision for pulpotomy remained the lowest at 50.0%. The overall model performance, measured by mAP@50, was calculated as 73.2%. According to the results of our study, successful outcomes were achieved in determining the dental treatment needs from panoramic radiographs using the YOLOv8x-based deep learning model. The model showed high performance particularly in extraction, space maintainer, and restorative filling categories. The overall mAP@50 value indicated reliable performance in distinguishing and detecting treatment classes.