Developing an AI Algorithm to Automatically Detect Early Radiographic Changes of Tooth Mobility in Patients with Bruxism

Objective: To develop an artificial intelligence (AI) algorithm capable of automatically detecting early radiographic signs of tooth mobility in patients diagnosed with bruxism, thereby facilitating earlier intervention and personalised risk stratification. Methods: Digital periapical radiographs (n=3200) for upper and lower posterior teeth acquired from adult patients (≥ 18y, n=1200) who were with clinically diagnosed sleep and awake bruxism. Gold standard labels were generated by a panel of three board-certified dento-maxillofacial radiologists who reached consensus on the presence of key early mobility markers. A U Net variant with an EfficientNet B3 backbone was trained using focal Tversky loss. Results: On the test set (n = 3500 images), the algorithm achieved an overall AUC of 0.941 (95 % CI, 0.932–0.949), sensitivity of 0.884, and specificity of 0.903 for detecting ≥ 1 early mobility marker. Expert readers exhibited AUCs ranging from 0.769 to 0.916. The algorithm outperformed junior readers across all metrics (p < 0.001) and demonstrated non-inferiority to senior readers (ΔAUC = 0.008, p = 0.17). Visual saliency analysis confirmed that model attention co-localised with radiologist-defined regions of interest. Conclusion: The suggested algorithm shows reliable identification of subtle radiographic signs that often mark the earliest onset of tooth mobility in patients with bruxism. Specialist-level accuracy is reached. When placed within everyday chairside imaging workflows, it holds the potential to speed up diagnostic steps, support occlusal treatment planning more precisely, and ideally help prevent permanent damage to the periodontium.