Predicting Skeletal Landmarks from Soft-Tissue Landmarks Using Machine Learning: A Study on Nasion Localization

Cephalometric analysis plays a crucial role in orthodontics and maxillofacial surgery, traditionally relying on X-ray imaging to assess skeletal structures. However, reduced Field of View (FOV) CBCT scans, which minimize radiation exposure, often exclude key reference points such as the skeletal nasion. This study investigates the feasibility of predicting the nasion position using machine learning (ML) models based on soft-tissue landmarks. A dataset of 137 CBCT scans was analyzed, with the Sellion, Tragus (left and right), and Alare (left and right) landmarks used as predictors. Three ML models-Linear Regression, Random Forest, and Feedforward Neural Network (FFNN)-were evaluated through 10-fold cross-validation. The Linear Regression model demonstrated the highest accuracy, achieving a mean Euclidean error of 1.452 ± 1.077 mm between predicted and ground truth landmarks. These findings suggest that ML models can reliably estimate skeletal landmarks from soft-tissue features, introducing the possibility to evaluate a less-invasive alternative for cephalometric analysis. This approach has the potential to reduce reliance on full-cranium CBCT scans, thereby lowering patient radiation exposure while maintaining diagnostic accuracy.Clinical Relevance- The use of ML to predict skeletal landmarks from soft-tissue references offers a radiation-minimizing alternative for cephalometric analysis. By integrating reduced-FOV CBCT with facial scan data, clinicians can accurately estimate the nasion position without requiring full-cranium imaging. This approach can enhance orthodontic and surgical planning, particularly in pediatric and radiation-sensitive patients. Further development of AI-driven landmark prediction may lead to broader applications in non-invasive craniofacial diagnostics, improving patient safety and accessibility to cephalometric analysis.