Assessing the Accuracy and Reliability of Fully AI-Driven Cephalometric Analysis in Comparison to Digital Manual Methods

Background: Artificial intelligence (AI) has invaded radiographic analysis in a massive way. Besides saving time and effort during decision-making and treatment planning, AI-assisted cephalometric analysis must be reliable, reproducible, accurate, and user-friendly. Objective: To assess fully AI-driven cephalometric analysis. Methods: 47 lateral cephalometric radiographs were used for comparing the accuracy of AI-operated and manually operated skeletal and dental cephalometric analysis. Both dental and skeletal analyses were digitally performed using web-based platforms (WebCephTM, Cephio, and Ceppro DDH Inc., Korea). SPSS was used for statistical analysis with paired t-test and intra-class correlation. Results: There were statistical differences between AI landmarking and manual landmarking regarding SNA, ANB, FH to AB, U1 to FH, U1 to SN, U1 to UOP, interincisal angle, U1 to NA (mm), and U1 to NA (deg). The intraclass correlation coefficient (ICC) data showed that the two sets of measurements were very consistent for most readings. Conclusions: Even though AI provides strong reliability and agreement between methods, the significant difference indicates that AI-operated and manually operated cephalometric analysis may not be interchangeable despite their consistency. AI-based analyses primarily function as assistant tools, and orthodontists need to make judgments or adjustments before making decisions.