Comparison of cephalometric analysis using artificial intelligence algorithm and computer aided digital cephalometric analysis

The aim of this study is observing whether fully automatic cephalometric analysis programs with artificial intelligence technology are as reliable as computer aided digital cephalometric analysis programs. For this purpose, radiographs that were manually marked and analyzed by 3 researchers with different levels of experience were compared with those that were automatically marked and analyzed by AudaxCeph, OrthoDx and WebCeph, which are fully automatic digital cephalometric analysis programs with artificial intelligence algorithm. In the study, in which the measurements of the fully automatic cephalometric analysis programs after the intervention of the investigator were also evaluated, the X and Y axis coordinates of 33 anatomical points (20 hard and 12 soft tissues) and 19 cephalometric analyzes, 13 angular, 5 linear and 1 proportional, on 210 digital cephalometric radiographs formed from 7 patient groups. value measurements were examined. In the measurements made by the AudaxCeph program as a result of comparing the obtained measurement values with the reference values between the patient groups; In the coordinates of Ba, Sigmiod Notch, Gl', N,' Sn', A' points and in the values of U1-NA° U1-NA (mm), Facial angle, Y axis angle, Jarabak ratio, in the measurements made by the OrthoDx program; Pt, N' points coordinates and Facial angle, Y axis angle, U1-NA (mm) values, Prn, A' points coordinates in measurements made by WebCeph program and Facial angle, Y axis angle, Gonial angle, U1-NA°, U1- Statistically significant difference was observed in NA (mm) values. After investigator interventions, some of the differences disappeared. WebCeph was the most successful program in hard tissue coordinates measurements and OrthoDx program in soft tissue coordinate and analysis values measurement. Hard tissue coordinate measurements of the AudaxCeph program were found to be more reliable than soft tissue. Although all programs with artificial intelligence algorithms are considered successful, it is thought that researcher intervention is needed to obtain more accurate results