Accuracy of Eyes of AITM Artificial Intelligence-Driven Platform for Lateral Cephalometric Analysis

Abstract

Aim: The objective of this prospective study was to evaluate the accuracy of cephalometric analyses acquired through manual tracing and the Eyes of AI^TM AI-driven web-based program. Materials and Methods: This prospective study employed randomization conducted via computer software, with a determined sample size of 150 cases. Inclusion criteria encompassed good quality lateral cephalograms available in both digital and print formats, absence of artifacts that might hinder anatomical point location, and presence of a clear calibration ruler for magnification determination. Exclusion criteria included lateral cephalograms with identifiable motion artifacts, resolution disparity, or insufficient contrast, as well as those exhibiting positional errors indicated by ear rod markers. Each lateral cephalogram underwent tracing and analysis using the manual method, as well as Eyes of AI^TM software. Following landmark plotting, linear and angular measurements of Steiner, Downs, McNamara, and Jefferson analyses were calculated. Results: A comparison of thirty-six cephalometric measurements of Steiner, Downs, McNamara, and Jefferson analyses obtained from manual tracing and AI-driven Eyes of AI^TM revealed a Concordance Correlation Coefficient (CCC) value above 0.76 for all parameters, indicating strong agreement between manual and AI-driven cephalometric measurements. Furthermore, a CCC value exceeding 0.9 was observed for twenty-eight parameters, indicative of very strong agreement. Conclusion: Automated lateral cephalometric measurements obtained from Eyes of AI^TM are accurate when compared to manual measurements.