Accuracy of DentalMonitoring's artificial intelligence in detecting the loss of aligners’ auxiliaries: a retrospective multi-centric study

Abstract

Introduction: This study aimed to evaluate the performance of DentalMonitoring's (DM) (Dental Monitoring SAS, Paris, France) Artificial Intelligence (AI) in detecting the loss of aligner auxiliaries, specifically attachments and buttons, by assessing its sensitivity and specificity using a binary classification approach (positive vs. negative). Methods: This multi-center retrospective study analyzed data from 719 patients treated with aligners, randomly selected by an independent third party from DM's U.S.-based clinical data pool to ensure an unbiased, representative dataset. For each patient, sequential intraoral image sets (≥8 views) generated by DM scans were processed by DM's AI to detect attachment and button loss. Three blinded, trained US-based orthodontic residents independently reviewed each case to establish a reference outcome. In cases of disagreement between expert consensus and AI output, a fourth blinded external expert adjudicated the final ground truth. Attachment and button loss were assessed through evaluation of sequential scans. Sensitivity and specificity with 95% confidence intervals were calculated using a Generalized Estimating Equations (GEE) model. All analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC, USA). Results: The final dataset included 765 results for the attachment loss parameter and 659 results for the button loss parameter. For DM's AI detection of attachment loss and button loss, sensitivity was 98.2% (95% CI: 94.3%–99.4%) and 98.4% (95% CI: 94.0%–99.6%), respectively; specificity was 100% (95% CI: 98.7%–100%) and 99.0% (95% CI: 96.9%–99.7%), respectively. Conclusion: Current results indicate that DM's AI analysis system has extremely high accuracy in detecting debonding of auxiliaries. Consequently, DM may help significantly reduce the occurence of these undetected clinical incidents.