AI for sagittal skeletal classification on lateral cephalograms: preliminary meta-analysis

Meta-analysis of 3 AI studies (2743 lateral cephalograms): pooled sensitivity 87%, specificity 91%, AUC 0.94. Single-center, no external validation—estimates are exploratory.

• Systematic review and meta-analysis of 3 studies (9 model arms, 2743 lateral cephalograms).
• DenseNet-based models generally performed best; Swin-T transformer weakest.
• Major limitations: single-center data, lack of external validation, high heterogeneity; pooled results are exploratory, not definitive.

RadPigeon summaries are original and for information only. They are not clinical advice.