Assessment of virtual bracket removal by artificial intelligence and thermoplastic retainer fit.

Introduction: Digital orthodontics is here to make our specialty more efficient, and the integration of artificial intelligence (AI) is no exception. This study aimed to compare the accuracy of a workflow involving virtual bracket removal (VBR) by AI to traditional bracket removal. A secondary objective was to compare the clinical fit of thermoplastic orthodontic retainers fabricated from 3-dimensional (3D) printed models created by each method.

Methods: Thirty dental arches were scanned intraorally with the fixed labial orthodontic brackets in place, creating 30 standard tessellation language (STL) files which were labeled "pre-debond." For each patient, all brackets and residual adhesive were then clinically removed, and an additional intraoral scan was taken and labeled "post-debond." Each pre-debond scan was uploaded to the uDesign by uLab software (version 7.0; uLab Systems, Inc, Memphis, Tenn), which uses AI to virtually remove orthodontic brackets using 1-touch bracket identification. New STL files were created for each of the 30 arches and labeled VBR. Best fit superimpositions were performed using Geomagic Control X (3D Systems, Inc, Rock Hill, SC), and data were collected as both numerical values and color-coded deviation maps. Superimposition data were segmented for each arch (anterior, posterior right, and posterior left) to determine if the arch location was a relevant factor. All post-debond and VBR STL files were 3D printed. Thermoplastic retainers were fabricated on each 3D-printed model. Retainers were delivered to patients and clinically graded by 2 practitioners.

Results: There was a statistically significant difference in standard deviation and out-of-tolerance percentage values when comparing traditional debonded arches and virtually debonded arches when using sextant locations as a factor (anterior, posterior right, and posterior left). Grading scores for both evaluators show a statistically significant difference for the evaluation of facial surface adaptation and snap-fit acceptability between retainer fabrication methods. However, there was no correlation between out-of-tolerance values and clinical retainer fit evaluation among any of the graded variables in either the maxillary or mandibular arches.

Conclusions: VBR by AI is considered accurate enough to be used for the fabrication of clinically acceptable thermoplastic orthodontic retainers.