Evaluation of an AI-based Remote Monitoring System for Detection of Post-orthodontic Anterior Tooth Movement

Abstract

Objective: Monitoring anterior alignment during retention is challenging because in‑office visits are infrequent. Dental Monitoring (DM), an AI–based remote system, may allow earlier detection of post-orthodontic misalignment, but its agreement with clinician assessments is uncertain. Materials and Methods: This retrospective study included 80 patients (160 arches) previously treated with fixed orthodontic appliances and monitored with DM during retention. Scans at debond and at the first DM tooth-movement alert were evaluated for 12 anterior teeth (maxillary and mandibular canine to canine). Three raters (DM, Clinician 1, Clinician 2) scored each tooth as aligned or misaligned. Agreement was quantified with Fleiss kappa. A generalized estimating equation logistic regression model with rater, tooth, and their interaction as predictors accounted for clustering of teeth within patients and yielded predicted misalignment probabilities by rater and tooth. Results: Overall interrater agreement among DM, Clinician 1, and Clinician 2 was substantial (Fleiss κ = 0.77). The GEE model showed significant effects of rater, tooth, and their interaction (all P 0.001). Predicted probabilities of misalignment were 0.21 for Clinician 1, 0.24 for Clinician 2, and 0.20 for DM. Collapsing across raters, probabilities ranged from 0.32 to 0.47 at the maxillary central and lateral incisors and from 0.08 to 0.12 at UL3, LL3, and LR3, with a maximal difference of about 0.16 between raters at any single tooth. Conclusions: DM showed good agreement with calibrated orthodontists in classifying anterior misalignment during retention. Rater-related differences were modest compared with tooth-to-tooth variation in misalignment probability, supporting DM as an adjunctive tool for monitoring post-orthodontic alignment.