Quantitative evaluation of an artificial intelligence-driven remote monitoring system for occlusion assessment using...

To evaluate the accuracy of an artificial intelligence (AI) model developed by DentalMonitoring for assessing occlusal parameters from patient-acquired intraoral images, using intraoral scanner (IOS)-derived three-dimensional (3D)measurements as the reference standard.

This multicenter prospective study included 430 orthodontic patients from three clinics in the United States. Each participant completed a DentalMonitoring scan using the DM ScanBox and a clinician-acquired IOS scan. Midline deviation, overbite, overjet, and canine class were measured on IOS-generated 3D models using metrology-grade software (ZEISS Inspect). Three independent, blinded technicians performed measurements, with the median value used as the reference. Agreement between AI-generated and reference measurements was assessed using Passing-Bablok regression and relative bias analyses at predefined clinical thresholds.

All occlusal parameters demonstrated agreement within clinically acceptable limits. Midline deviation and overbite showed the highest concordance, with intercepts near 0.00 mm, relative biases below 3%, and mean biases of -0.01 ± 0.26 mm and -0.04 ± 0.39 mm, respectively. Overjet was modestly overestimated (mean bias = +0.29 ± 0.52 mm), while canine class showed increasing underestimation at higher values (mean bias = -0.31 ± 0.91 mm).

The evaluated AI model demonstrated high agreement with IOS-based 3D measurements for midline deviation and overbite, with greater variability for overjet and canine classification. These results support the use of AI-assisted monitoring for screening and follow-up, while highlighting the need for further validation prior to routine clinical implementation.