CBCT radiomic features-based machine learning prediction models and nomogram for treatment decision-making regarding surgery-first approach in skeletal Class III malocclusion

This study aims to develop and validate a predictive nomogram and a deep learning model to determine the treatment modality of orthodontic-first approach (OFA) or surgery-first approach (SFA) for patients with skeletal Class III malocclusion using CBCT images. A nomogram was developed using CBCT images from 313 skeletal Class III patients (191 OFA vs. 122 SFA), through univariable and multivariable analyses to identify key predictors for SFA selection. Predictive accuracy was assessed using ROC analysis. CBCT images were split into training and validation sets for 5-fold cross-validation, and various deep learning models (Simple-CNN, DenseNet121, MobileNetV2, EfficientNet-B0, ResNet10, and SEResNet50) were applied to classify OFA vs. SFA. Key predictors for the nomogram included maxillary crowding, maxillary arch form asymmetry index (AI), mandibular AI, and U1-SN angle. The nomogram showed calibration with an AUC of 0.995. The Simple-CNN model achieved the best performance with sensitivity of 0.979, specificity of 0.800, precision of 0.960, accuracy of 0.909, F1-score of, 0.969 and AUC of 0.896. This study introduces a nomogram for predicting SFA selection in skeletal Class III malocclusion patients and demonstrates the effectiveness of the Simple-CNN model for automated classification of treatment approaches.