Reliability of Artificial Intelligence in Lateral Cephalometric Analysis

Abstract

Aim: The purpose of this study was to assess the reliability of lateral cephalometric analysis performed by an artificial intelligence-dependent software program. Materials and Methods: One Hundred and Eighty digital cephalometric radiographs acquired by Vatech PaX-i X-ray machine, were used in the study. The anatomical landmarks of both Steiner and McNamara analyses were manually traced using a third-party software AudaxCeph Empower, version 6.6.12.4731 (Audax d.o.o., Ljubljana, Slovenia), the tracing was performed by two radiologists with more than 5 years of experience in digital cephalometry to determine the inter-reliability, then it was repeated with an interval of two weeks to determine the intra-reliability. The landmarks were retraced automatically through the fully automatic option on the same software program using convolutional neural network. Results: Regarding McNamara analysis, the results of this study showed excellent reliability of the artificial intelligence measurements compared to the manual measurements, with an interclass correlation coefficient >0.9. Regarding Steiner analysis, our results showed excellent reliability of the artificial intelligence measurements compared to the manual measurements (0.75