Validation of an AI-aided 3D method for enhanced volumetric quantification of external root resorption in orthodontics

To compare and validate two tridimensional diagnostic methods for quantifying and categorizing external root resorption using an artificial intelligence (AI)-aided, automatic, or manual digital segmentation process.

40 teeth were segmented from 10 cone beam computed tomography (CBCT) records from five patients. Stereolithographic files were created, and automatic, manual, or AI-aided segmentation of each incisor was performed by two double-blinded operators. Two quantification methods were used and compared by analyzing final segmented regions of the tooth. This study followed QAREL (Quality Appraisal of Diagnostic Reliability) and COSMIN (COnsensus-based Standards for the selection of health Measurement Instruments) guidelines. Reproducibility was assessed using the Dahlberg formula, coefficient of variation, and intraclass correlation coefficient (ICC) (P value < .05).

Intra- and interobserver correlations were high (ICC: > 0.736; P < .01). Statistically significant differences were found between the two measurement methods for high-quality CBCT images of central incisors, mainly at the level of the apical third. Specific differences were found between methods when root resorption was evaluated in the middle and apical thirds using AI segmentation of the central incisor (P = .043). When referring to total volume loss of the lateral incisor, differences (P = .021) were observed between methods when segmented by manual or AI-aided procedures. Highest specificity (100%) was observed for AI-aided segmentation and Method 2 for evaluation of root resorption at the apical third volume.

Assessment of root resorption with CBCT is highly dependent on CBCT definition, type of segmentation, and measurement method. Three-dimensional (3D) measurement method described by three landmark points yielded satisfactory results using any tested segmentations.