Toward the Development of an Oral-diagnosis Framework: A Case Study of Teeth Segmentation and Numbering in Bitewing Radiographs via YOLO Models

This study presents an oral-diagnosis framework integrating the YOLOv8 model for precise tooth localization in dental imaging. The dental segmentation and numbering in the right-side bitewing radiographic images were evaluated through comparison of the YOLOv5 and YOLOv8 models, employing confidence thresholds. The dataset comprised 800 training images and 152 testing images, with the YOLOv8 architecture deployed in three variants. Precision, recall, F1-score, and mean average precision (mAP) were evaluated for both models. YOLOv8 demonstrated superior performance over YOLOv5 in precision (0.913 vs. 0.897), F1-score (0.931 vs. 0.920), and mAP (0.96 vs. 0.954). Variations in model dimensions were observed among YOLOv8 S, M, and L variants, with marginal mAP improvements in specific classes. In conclusion, while YOLOv8 did not enhance dental segmentation and numbering tasks across varying architecture sizes, it consistently outperformed YOLOv5, exhibiting superior segmentation and detection abilities.