Comparison Between the Accuracy of Different AI-Based Cephalometric Analysis and Conventional Manual Analysis

Objective: This study compared artificial intelligence-based cephalometric analysis platforms with conventional manual analysis using the Steiner Cephalometric analysis system. Methods: Three AI online software platforms: WEBCEPH, AssembleCircle Corp, South Korea, CEPHIO, Cephio Sp. Z o.o., Poland, and VOXEL Voxel3Di, United Kingdom, were evaluated using 25 cephalometric radiographs and compared with two manual tracings performed using (Lab Pronto, Blue Sky Plan) for the Steiner Cephalometric analysis system. Independent sample t-test was used to compare the two manual analysis. The Kruskal-Wallis H-test, followed by pairwise comparisons, was used to evaluate differences between AI-based and manual analyses, with significance at p≤0.05. Results: No significant discrepancies were found between the two manual analyses. However, the Kruskal-Wallis H-test and pairwise comparisons revealed significant differences between the AI platforms regarding the SNA angle and the U1 to NA angle. Specifically, VOXEL AI analysis differed from WEBCEPH (at p=0.003) and CEPHIO (at p=0.011) for these measurements. Conclusions: The findings highlight that while AI platforms such as VOXEL show promising alignment with manual analysis across most measurements, discrepancies, particularly in specific angles, underscore the need for cautious integration and ongoing validation of AI in orthodontic practice. The selection of appropriate AI-based analysis is crucial; however, manual analysis remains the gold standard for cephalometric analysis.