Comparative Performance of YOLO Models in the Automated Segmentation of Dental Traumas on Panoramic Radiographs

This study aimed to compare the performance of artificial intelligence-based deep convolutional neural networks, YOLOv8, YOLOv11, and YOLOv12, in segmenting dental injuries using panoramic films of pediatric patients with traumatic dental injuries.

Panoramic radiographs of pediatric patients aged 6-13 years with traumatic dental injuries presented to the Gaziantep University Faculty of Dentistry were input into an artificial intelligence model (CranioCatch, Eskisehir-Turkey) using YOLOv8, YOLOv11, and YOLOv12 as models to automatically detect and classify dental injuries.

The AUC values of YOLOv8, YOLOv11, and YOLOv12 were 0.72, 0.69, and 0.73 for hard tissue injuries and 0.61, 0.67, and 0.69 for soft tissue injuries, respectively. Multiclass F1-scores are 0.592, 0.653, and 0.683, respectively. All models were able to better discriminate hard tissue injuries, and the most consistent results were obtained with YOLOv12.

The YOLOv12-based deep learning model demonstrated better performance in detecting dental trauma in pediatric panoramic radiographs compared to other models. Nevertheless, artificial intelligence has not yet achieved flawless performance in Dental Traumatology. Therefore, AI tools should be developed in collaboration with expert dentists to better support clinical decision-making in dental trauma cases.