Comparative analysis of YOLO variants for dental implant brand identification

This research primarily focuses on identifying implant brands, specifically Bicon, Bego and ITI, and detecting their locations on the periapical and panoramic X-ray images using YOLO (You Only Look Once) object detection algorithm. A dataset of 2290 annotated images sourced from Roboflow.com was utilized. The dataset was augmented by applying 50% probability of horizontal flip, 50% probability of vertical flip, equal probability of one of the following 90-degree rotations: none, clockwise, counter-clockwise and salt and pepper noise by the Roboflow. Prior to training, the dataset was splitted into three subsets: train of 1893 samples, test of 218 samples and validation of 179 samples. The detection algorithm was trained using multiple YOLOv8, YOLOv9, YOLOv10 and YOLOv11 models and their sub-versions medium and large variants. In order to evaluate the effectiveness of the models, the performance metrics including precision, recall, F1 score, and mAP@50 (mean Average Precision at Intersection over Union (IoU) threshold 0.50) were analyzed. The highest precision values are 98% for ITI with YOLOv8m, 91.9% for Bicon with YOLOv10m and YOLOv11l, 91.9% for Bego with YOLOv10m and 93.1% for all classes with YOLOv10m. The top-performing recall values were achieved at 94.1% for Bego using YOLOv11l and YOLOv9m, 94.9% for Bicon using YOLOv9m and YOLOv8m, 94.7% for ITI using YOLOv10m and 92.9% for all classes using YOLOv8l. The optimal F1 scores were acquired through 88% for Bego by YOLOv11l, 92.1% for Bicon by YOLOv10m, 96% for ITI by YOLOv11m and %91.4 for all classes by YOLOv10m. YOLO models are effective tools for dental imaging, providing a reliable and efficient approach to automating implant identification. Future studies could improve detection accuracy by utilizing larger and more diverse datasets and exploring other potential applications in dental imaging.