Is automatic cephalometric landmark detection with artificial intelligence reliable? - a systematic review with meta-analysis of SNA, SNB, ANB, and WITS cephalometric measures

Cephalometric analysis stands out due to its relevance in orthodontics for diagnosing and treating malocclusion and skeletal discrepancies. A systematic review (SR) with meta-analysis is required to evaluate the reliability of using artificial intelligence (AI) for cephalometric landmark detection in lateral cephalometric radiographs. This study searched PubMed, Scopus, Embase, Web of Science, LILACS, and Google Scholar databases using the PICOS (Population, Intervention, Comparison, Outcome, and Study design) search strategy. The treatment effects were defined as the standardized mean difference (SMD), and 95% confidence intervals (CI) were established. The Joanna Briggs questionnaire for non-randomized studies evaluated the risk of bias. The GRADE tool assessed the quality of evidence of the systematic review. The search detected 1041 articles, of which 32 full texts were selected, and 14 were included in the systematic review after a careful analysis. Five of these 14 studies were included in the meta-analysis to assess SNA, SNB, ANB, and Wits cephalometric measures. This analysis found Tau2=0.04, Chi2=31.70, and p=0.01, indicating statistical heterogeneity, and I2=50%, showing moderate heterogeneity. The total SMD was -0.05, with a confidence interval from -0.19 to 0.08, presenting no statistical differences between the manual and automatic (performed with AI) measurements but with a small effect size. AI results were similar to the control in most studies on the precision of cephalometric landmark identification.