Predicting the Risk of Maxillary Canine Impaction Based on Maxillary Measurements Using Supervised Machine Learning

Area of Science:

• Dentistry
• Medical Imaging
• Machine Learning

Background:

• Palatally impacted maxillary canines are a common dental anomaly.
• Accurate prediction is crucial for timely orthodontic intervention.
• Current diagnostic methods may not fully capture predictive skeletal factors.

Purpose of the Study:

• To develop and evaluate supervised machine learning models for predicting palatally impacted maxillary canines.
• To identify key maxillary measurements predictive of impaction.
• To compare the performance of various machine learning algorithms.

Main Methods:

• Analysis of 138 cone beam computed tomography (CBCT) scans.
• Measurement of multiple maxillary dimensions including widths and length.
• Application of eight supervised machine learning algorithms (e.g., Gradient Boosting, Random Forest, SVM).
• Validation using 5-fold cross-validation and performance metrics (AUC, accuracy, precision, recall, F1 Score).

Main Results:

• A predictive model incorporating two dental and two skeletal measurements was developed.
• Interpterygoid width and nostril width demonstrated the largest effect sizes.
• Gradient Boosting Classifier achieved the highest performance, with AUC values up to 0.91.
• Nostril width was identified as the most important variable across algorithms.

Conclusions:

• Supervised machine learning utilizing maxillary measurements is a promising approach for predicting palatally impacted canines.
• Gradient Boosting Classifier and Random Forest Classifier showed strong predictive capabilities (AUC > 0.8).
• This method offers potential for improved early diagnosis and treatment planning.