Artificial Intelligence-Assisted Prediction of Midpalatal Suture Maturation from Cervical Vertebral Maturation Assessed on Lateral Cephalometric Radiographs: Development of a Combined Diagnostic Model

Background/Objectives: This study evaluated the association between cervical vertebral maturation (CVM) and midpalatal suture (MPS) maturation and developed an artificial intelligence (AI)-assisted model for predicting MPS stage from lateral cephalometric radiographs and clinical variables. Methods: CVM and cone-beam computed tomography (CBCT)-defined MPS stages were assessed in 250 individuals. Their association was examined using Spearman correlation and multivariable ordinal logistic regression incorporating CVM stage, chronological age, and sex. A separate dataset of 1448 cephalograms was used to develop and evaluate a YOLOv11 model for automated CVM classification. The combined-model cohort comprised 325 patients, including the initial 250 and 75 additional patients, divided into training (n = 260) and held-out internal test (n = 65) subsets. The combined model was trained using reference CVM stage, chronological age, and sex and evaluated on the held-out test set using AI-predicted CVM stage in place of reference CVM. Results: CVM and MPS stages were strongly correlated (r = 0.749, p < 0.001). CVM stage and age were significantly associated with MPS stage, whereas sex was not (Nagelkerke R2 = 0.716). YOLOv11 correctly detected and classified 277/344 reference regions (80.5%). On the held-out test set, the combined model achieved 69.2% five-class accuracy (45/65; 95% CI: 57.2–79.1%) and 92.3% binary accuracy for A–C versus D–E. Repeated nested cross-validation across 325 patients yielded 58.3% mean accuracy with reference CVM, compared with 58.8% for reference CVM alone. Substituting AI-CVM only during evaluation in the 65-patient subset yielded 55.7% mean accuracy. Conclusions: The proposed AI-assisted model remains exploratory, with automated-input evaluation limited to 65 patients. The added predictive value of age and sex beyond CVM staging alone remains unproven. Its performance does not establish the ability to predict expansion success or justify replacing clinically indicated CBCT. External validation is required before clinical implementation.