Machine Learning Algorithms for the Diagnosis of Class III Malocclusions in Children

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging

Background:

• Artificial intelligence (AI) has advanced medical diagnosis but hasn't been applied to classifying Class III malocclusions in children.
• Class III malocclusions require accurate classification for effective treatment planning.

Purpose of the Study:

• To develop and validate an innovative machine learning (ML) model for the automated classification of dental, skeletal, and functional Class III malocclusions in children.
• To identify key cephalometric features predictive of Class III malocclusion severity.

Main Methods:

• Utilized cephalometric data from 666 children (ages 4-14), analyzing 46 measurements.
• Applied Recursive Feature Elimination (RFE) to select 14 significant features from the initial 46.
• Developed and evaluated 10 ML models, with Gaussian Process Regression (GPR) selected as the optimal model based on AUC and interpretability analysis using SHAP.

Main Results:

• The Gaussian Process Regression (GPR) model achieved the highest Area Under the Curve (AUC) of 0.879.
• No significant statistical difference was found between GPR and other top-performing models (p > 0.05).
• Key features identified for classification included SN-GoMe, U1-NA, Overjet, ANB, and AB-NPo.

Conclusions:

• ML models utilizing cephalometric data can effectively assist dentists in classifying pediatric Class III malocclusions.
• Cephalometric indicators like SN-GoMe, U1-NA, and Overjet are valuable for predicting malocclusion severity.