Automated periodontitis diagnosis and staging using an end-to-end deep learning model on panoramic dental radiographs.

OBJECTIVES: This study aimed to develop and validate a comprehensive deep learning model utilizing YOLOv11 for the automatic segmentation of teeth, detection of anatomical landmarks, and staging of periodontitis through the analysis of panoramic radiographs. METHODS: A total of 607 panoramic radiographs were annotated by qualified dentists The YOLOv11-based system executed tooth segmentation and classified teeth into four anatomical categories, subsequently identifying six critical landmarks to facilitate the calculation of radiographic bone loss. The performance of the model was assessed using metrics such as precision, recall, F1-score, and mean average precision (mAP). Agreement with the assessments provided by dentists was evaluated using Pearson correlation coefficients (PCC) and intraclass correlation coefficients (ICC). RESULTS: The segmentation model demonstrated exceptional performance, achieving a precision of 0.981, recall of 0.984, F1-score of 0.983, and an mAP50 of 0.994. The keypoint detection models also exhibited robust performance, with precision and recall exceeding 0.951 and mAP50 surpassing 0.958. Correlation and agreement with the assessments of dental professionals were highest for incisors (PCC = 0.877; ICC = 0.871) and lowest for canines (PCC = 0.712; ICC = 0.705) (p < 0.001). CONCLUSIONS: The proposed YOLOv11-based framework facilitates automated, high-precision tooth segmentation, landmark detection, and periodontitis staging, thereby providing a reliable clinical decision support tool for the diagnosis of periodontal disease.