Designing and developing a two-stage artificial intelligence model for available space analysis in the dental arch: A comparison with manual analysis
Abstract
Accurate measurement of available space is essential for orthodontic diagnosis and treatment planning. Conventional cast-based space analysis is time-consuming, operator-dependent, and prone to measurement variability. This study aimed to design and implement a two-stage artificial intelligence (AI) model, combining the You Only Look Once, version 8 (YOLOv8) algorithm and the Segment Anything Model (SAM), to enable rapid, automated measurement of available space in the dental arch using occlusal photographs and to compare its performance with manual cast analysis and specialist assessment. The dataset contained 1,428 occlusal photographs of the maxilla and mandible from 714 adult orthodontic patients with permanent dentition. In the first stage, YOLOv8 segmented and isolated teeth from soft tissues. In the second stage, SAM identified five anatomical landmarks and measured inter-landmark distances to calculate the total available space. Model performance was evaluated using the intraclass correlation coefficient (ICC), mean absolute error (MAE), and Bland–Altman plots. At the arch level, for the maxilla, ICC was 0.89, and MAE was 2.10 ± 0.60 mm; for the mandible, ICC was 0.93, and MAE was 1.82 ± 0.56 mm. Bland–Altman analysis showed no significant systematic bias. For overall space analysis, the orthodontist showed an ICC of 0.87 (95% confidence interval [CI]: 0.83–0.90) versus the standard castbased measurements, whereas the AI system achieved an ICC of 0.93 (95% CI: 0.90–0.95). The two-stage AI model accurately estimated available space from occlusal photographs, with agreement comparable to or better than manual cast measurements and lower error than specialist assessment in overall space analysis. This AI model can simplify orthodontic space analysis, offering a fast, accessible, and objective tool that also facilitates data storage, sharing, and potential integration with 3D printing workflows.