Evaluation of artificial intelligence-based cephalometric tracing versus semi-automatic and manual tracing

Artificial intelligence (AI)-based cephalometric tracing has emerged as a promising tool that reduces operator variability and offers standardized, rapid, and reproducible assessments. This study aimed to evaluate the reliability and accuracy of three cephalometric tracing methods: AI-based automatic digital tracing, semi-automatic digital tracing, and manual digital tracing.

This study analyzed pre-treatment lateral cephalograms from 120 patients, comparing AI-based automatic digital tracing, semi-automatic digital tracing against manual digital tracing, which served as the control group. Angular and linear measurements were used as primary parameters, with 34 cephalometric measurements derived from 20 skeletal (SK) and 10 soft tissue (ST) landmarks. For parametric data, comparisons among the three methods were conducted using a repeated measures ANOVA test, while non-parametric data were analyzed using the Friedman test. A significance level of P ≤ 0.05 was applied to all statistical analyses.

In maxillary skeletal measurements, SNA (°) showed significant variation, with the Automatic method reporting slightly higher mean values (P < 0.001). Similarly, mandibular skeletal measurements revealed a small but significant difference in SNB (°) values (P = 0.002), while MP-SN (°) showed no notable differences (P = 0.118). For inter-maxillary measurements, ANB (°) displayed significant differences (P < 0.001), with the Automatic method reporting higher mean values. Vertical measurements, such as GO-GN to SN (°) and the Gonial Angle (°), also demonstrated significant variations (P < 0.001), with manual methods generally reporting higher values. In dental measurements, L1 to NB (°) differed significantly across methods (P = 0.022), and U1 to NA measurements showed both angular (P = 0.006) and linear (P < 0.001) significant differences, with Automatic methods tending to report lower medians. Lastly, soft tissue measurements, particularly the nasolabial angle (°), exhibited significant differences (P < 0.001), with the Manual method recording the highest mean values.

AI-based automatic tracing tended to overestimate certain skeletal values, while manual methods showed greater consistency. Dental measurements were largely comparable across methods. The semi-automatic approach provided a practical balance between accuracy and efficiency, indicating potential for clinical application with further refinement.