Comparative Analysis of Yolov5 and Yolov8 Deep Learning Models in the Detection and Anatomical Classification of Mandibular Fractures

Objective: This study aimed to compare the diagnostic performance of two deep learning-based object detection algorithms—YOLOv5 and YOLOv8—for the automatic detection and anatomical classification of mandibular fractures on panoramic radiographs. Methods: A total of 400 panoramic radiographs with confirmed mandibular fractures were collected from the clinical archives of Dicle University, Faculty of Dentistry. The dataset was expanded to 980 images using data augmentation techniques (rotation, contrast adjustment, and flipping). Images were annotated according to five anatomical regions (symphysis, body, angle, ramus, and condyle). YOLOv5 and YOLOv8 models were trained on 80% of the dataset, validated on 10%, and tested on 10%. Model performance was evaluated using precision, recall, F1-score, mean average precision (mAP), and intersection over union (IoU). Results: YOLOv8 achieved higher diagnostic accuracy compared to YOLOv5. The YOLOv8 model yielded 0.85 precision, 0.83 recall, 0.84 F1-score, 0.89 mAP, and 0.82 IoU, while YOLOv5 achieved 0.81, 0.78, 0.79, 0.84, and 0.77, respectively. Visual inspection of detection maps confirmed that YOLOv8 produced more stable bounding box predictions and better localization across all anatomical zones, particularly in the angle and condylar regions. Conclusion: Both models demonstrated high potential for assisting clinicians in detecting mandibular fractures on panoramic radiographs. YOLOv8 exhibited superior precision and generalization ability, suggesting that advanced deep learning architectures may improve diagnostic workflows in dental trauma assessment.