Abstract This study investigates the application of recent YOLO (You Only Look Once) algorithms for automated detection and classification of apical periodontitis using the Periapical Index (PAI) scoring system (1–5). A dataset of 699 digital periapical radiographs was collected from diverse sources, de-identified, and annotated by calibrated experts before splitting into training, validation, and testing sets. Three deep learning models (YOLOv8m, YOLOv11m, YOLOv12m) were trained and their performance was evaluated using Precision, Recall, F1 score, mean average precision (mAP50), Intersection over union (IoU), and confusion matrices. Two-sided McNemar’s exact test was further conducted to compare models on lesion-level outcomes. The results showed comparable mAP50 scores among the tested models. YOLOv11m and YOLOv12m demonstrated higher Precision (88.5% and 89.1%, respectively) compared to YOLOv8m (86.8%). YOLOv11m exhibited the highest Recall (86.2%) and maximum F1 score (87.1%). Confusion matrix analysis indicated superior prediction for PAI scores 3–5 across all models, with YOLOv11m excelling in scores 1 and 2, YOLOv8m performing best for score 4, and equal performance for score 5. The findings demonstrate the prospective of YOLO algorithms in automating the detection and classification of apical periodontitis using PAI. Accuracy and efficiency of tested models suggest their potential for integration in clinical workflows.
Saber et al. (Mon,) studied this question.