Mandibular angle bone appositions in bruxism: A deep learning-based detection and staging study

Abstract

This study aims to automatically detect, classify, and stage bone apposition changes in the mandibular angle region associated with bruxism using panoramic radiographs. The deep learning-based YOLO11x architecture was implemented to identify and categorize these structural stages. A total of 800 panoramic radiographs were annotated as stage 0–3 by a specialist dentomaxillofacial radiologist. Bruxism status was determined from clinical records, while bone appositions were staged radiographically. Each radiograph was divided along the midline and the left side mirrored, yielding 1600 half-jaw images used to train the YOLO11x model. Performance was evaluated using confusion matrices, precision, recall, F1-score, and mean average precision (mAP). The deep learning model achieved an overall mAP@50 of 0.864 on the validation set evalu ated during training and 0.834 on the independent test set, with the highest discriminative performance observed during training in stage 3 (0.901 mAP@50) and during testing in stage 0 (0.909 mAP@50). Confusion matrices confirmed high proficiency in anatomical localization and bounding-box precision. Classification errors were limited to adjacent stages, primarily due to morphological similarities in transition zones. Deep learning approaches have demonstrated acceptable performance in detecting bone changes associated with bruxism in panoramic radiographs, thereby establishing a basis for automatic staging. The findings reveal that the model has the potential to serve as an auxiliary tool to support radiographic assessment and staging of mandibular gonial bone appositions in clinically diagnosed/probable bruxism cases, despite the challenges encountered, particularly in transitional stages.