Construction of a CNN model based on lateral cephalometric imaging features and its application in surgery-first treatment decision-making for skeletal Class III malocclusion

Selecting surgery-first or orthodontics-first for skeletal Class III malocclusion is challenging. Whether lateral cephalograms can support preliminary treatment decisions remains under-explored. A total of 511 skeletal Class III patients undergoing orthodontic–orthognathic treatment were analyzed. Twelve lateral cephalometric variables were assessed, and multivariate analysis identified independent predictors of treatment selection. Subsequently, the lateral cephalograms were split (7:1.5:1.5) into training, validation, and test sets. The test set was kept completely independent from the model development process. Four deep learning models were evaluated using five-fold cross-validation. Finally, 87 patients with skeletal Class III malocclusion were included for external validation. Multivariate analysis identified L1–MP as an independent predictor of SFA selection. All four deep learning models achieved accuracies above 0.80. MobileNetV2 demonstrated the best performance, with the best-performing fold on the internal test set achieving an AUC of 0.949, accuracy of 0.953, sensitivity of 0.957, specificity of 0.941, precision of 0.889, and an F1 score of 0.914. In the external validation cohort, MobileNetV2 achieved an AUC of 0.890, an accuracy of 0.851, a sensitivity of 0.750, a specificity of 0.909, a precision of 0.828, and an F1 score of 0.787. MobileNetV2 demonstrated high accuracy in classifying lateral cephalograms, providing an accessible preliminary decision support tool for the selection between orthodontic-first and surgery-first approaches in skeletal Class III malocclusion patients.