Impact of cone-beam computed tomography image quality on artificial intelligence-driven three-dimensional tooth segmentation and evaluation of external apical root resorption.

Objectives: To assess the reliability of the diagnostic method using Diagnocat artificial intelligence (AI)-generated standard tessellation language (STL) files and Geomagic Wrap software and to evaluate whether cone-beam computed tomography (CBCT) quality influences the consistency of AI-based segmentation and related measurements. Materials and Methods: Forty orthodontic patients (20 from Spain and 20 from Sweden) were analyzed at two treatment time points (T0: beginning; T1: before using stainless-steel wires). STL files for each upper incisor were analyzed in Geomagic Wrap considering volume loss and tooth length. Reproducibility was assessed through repeated tooth reconstruction in an AI model, and the impact of the mesh correction tool and CBCT type was evaluated. Results: Intraobserver and interobserver reliability was high (intraclass correlation coefficient = 0.93). AI reconstructions were highly consistent. No significant differences were found between STL generations, except in one case. Mesh correction significantly affected volume measurements in the teeth from higher-dose CBCT scans. No significant differences were found between CBCT types in either root volume (P = .861) or length loss (P = .082). Analysis of variance showed no significant differences between CBCT types. Conclusions: The method is reproducible and reliable for linear and volumetric external apical root resorption (EARR) measurements using AI-generated STL models. CBCT image quality does not appear to influence volume measurements. Mild EARR was observed between T0 and T1, with no significant differences.