标题：Avaliação da performance de dois softwares com inteligência artificial por meio das medidas geradas pela análise de Mcnamara em telerradiografia cefalométrica lateral
出版社：Grupo de Pesquisa Metodologias em Ensino e Aprendizagem em Ciências
摘要：The aim of this study was to compare the performance of two software programs with AI in lateral cephalometric teleradiography by assessing the reproducibility and reliability of the linear and angular measurements of McNamara's analysis. Thirty cephalometric teleradiographs were marked using the digital method by the examiner in Radiocef (RadioMemory). Subsequently, the sample was marked using the CEFBOT (RadioMemory) and WebCephTM (AssembleCircle) software AI to evaluate the reproducibility and reliability of the examiner and the software. To calibrate the examiner and evaluate the reliability of the examiner, CEFBOT, and WebCephTM markings, the Intraclass Correlation Coefficient (ICC) was used, as well as the ANOVA test and Tukey's post-test evaluated the reproducibility of the software, using the cephalometric landmarks that comprise McNamara's analysis. The mean ICC of the examiner, CEFBOT and WebCeph were 0.960, 0.940 and 0.954, respectively, indicating almost perfect agreement. When comparing CEFBOT with examiner, statistical difference (p<0.01) was observed only in the perpendicular A-N measurement. As for WebCephTM, when comparing with the examiner there was a significant difference between factors two to six and ten. And compared to CEFBOT, there was divergence in the same factors plus factor eleven. In addition, WebCephTM did not identify the measurements Nfa-Nfp and Bfa-Bfp. CEFBOT showed reproducibility and reliability in identifying the cephalometric landmarks determined by McNamara's analysis but required human supervision. WebCeph showed almost perfect agreement in the markings, but six measurements were different from the examiner and two were not performed by the application.
