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

Area of Science:

• Orthodontics and Maxillofacial Surgery
• Medical Imaging
• Machine Learning

Background:

• Cephalometric analysis is vital for orthodontics and maxillofacial surgery, traditionally using X-rays.
• Reduced Field of View (FOV) Cone Beam Computed Tomography (CBCT) minimizes radiation but may omit key skeletal landmarks like the nasion.
• Accurate nasion identification is crucial for effective treatment planning.

Purpose of the Study:

• To investigate the feasibility of predicting the skeletal nasion position using machine learning (ML) models.
• To assess the accuracy of ML models in estimating nasion from soft-tissue landmarks.
• To explore a less-invasive alternative for cephalometric analysis, reducing radiation exposure.

Main Methods:

• Analyzed a dataset of 137 CBCT scans.
• Utilized soft-tissue landmarks (Sellion, Tragus, Alare) as predictors.
• Evaluated Linear Regression, Random Forest, and Feedforward Neural Network (FFNN) models via 10-fold cross-validation.

Main Results:

• Linear Regression achieved the highest accuracy with a mean Euclidean error of 1.452 ± 1.077 mm.
• ML models demonstrated reliable estimation of skeletal landmarks from soft-tissue features.
• The findings support the use of ML for predicting nasion position.

Conclusions:

• ML models can accurately predict skeletal landmarks from soft-tissue references, offering a radiation-minimizing approach to cephalometric analysis.
• This method enables nasion estimation without full-cranium CBCT, enhancing safety, especially for pediatric and radiation-sensitive patients.
• Further AI development in landmark prediction can advance non-invasive craniofacial diagnostics and improve patient care.