REPRODUCIBILITY AND RELIABILITY OF CEPHALOMETRIC LANDMARK DETECTION USING THE CEFBOT ARTIFICIAL INTELLIGENCE SOFTWARE

Abstract

<h3>Objectives</h3> Evaluate the reproducibility and reliability of a Cefbot-AI software in the identification of cephalometric points in lateral cephalometric radiographs. <h3>Study Design</h3> Thirty lateral cephalometric radiographs obtained with Orthophos XG5/Ceph (Dentsply Sirona, Pennsylvania) were selected, and 19 cephalometric points were marked on each radiograph by a control (ECont) and a calibrated (ECal) examiner and by the Cefbot-AI software (RadioMemory, Belo Horizonte, Brazil). Reliability was assessed by a new analysis of the radiographs 15 days after the first record. Statistical significance was set at <i>P</i> < .05. <h3>Results</h3> Excellent reliability was observed in the landmarks of all examiners, as well as in those made by Cefbot-AI (mean intraclass correlation coefficient >0.93). As for reproducibility, comparisons between the performance of ECont and ECal revealed some cephalometric points with a statistically significant difference on the <i>x</i>-axis, namely, S (<i>P</i> = .048), Po (<i>P</i> < .001), and ENA (<i>P</i> = .013). ECont and Cefbot revealed only 1 point, Or (<i>P</i> = .010), with a statistically significant difference on the <i>x</i>-axis. No statistically significant differences were found on the <i>y</i>-axis. <h3>Conclusions</h3> Excellent reliability and good reproducibility of Cefbot-AI was observed, suggesting that Cefbot is an excellent auxiliary tool for cephalometric analysis but still dependent on professional assessment for clinical use.