Determination of cervical vertebral maturation using machine learning in lateral cephalograms.

Area of Science:

• Orthodontics
• Radiology
• Machine Learning

Background:

• Accurate timing of orthodontic growth modification treatments is essential for optimal outcomes.
• Traditional methods like hand-wrist radiographs and cephalogram interpretation have limitations.
• Skeletal maturation assessment is key for effective orthodontic interventions.

Purpose of the Study:

• To develop a semi-automated machine learning approach for determining skeletal maturation status.
• To utilize cervical vertebral dimensions (CVD) for enhanced accuracy in growth assessment.
• To provide a more reliable method for timing orthodontic growth modification treatments.

Main Methods:

• Collected 980 lateral cephalograms and selected 8 landmarks on cervical vertebrae (C3, C4).
• Employed a ratio-based approach with an auto_error_reduction (AER) function for landmark accuracy.
• Utilized data augmentation and a stacking model trained and tested on cephalometric data.

Main Results:

• The developed machine learning model achieved a high accuracy of 99.49% on the test set.
• The model demonstrated a minimal loss value of 0.003, indicating strong predictive performance.
• Feature engineering and data augmentation contributed to the model's effectiveness.

Conclusions:

• A novel, semi-automated model accurately assesses skeletal maturation using cervical vertebral dimensions.
• The method simplifies landmark selection and enhances accuracy through AER and data augmentation.
• This machine learning approach holds promise for precise clinical applications in orthodontics.