A machine learning-based predictive model for mandibular third molar extraction difficulty: incorporating multimodal...

Area of Science:

• Oral and Maxillofacial Surgery
• Dental Imaging
• Machine Learning in Medicine

Background:

• Predicting mandibular third molar (MM3) extraction difficulty is crucial for surgical planning.
• Current methods often rely on subjective clinical judgment, leading to variability.

Purpose of the Study:

• To develop a rapid and accurate predictive model for MM3 extraction difficulty.
• To integrate machine learning with multimodal parameters, including CBCT imaging.

Main Methods:

• A dataset combining clinical data and automated CBCT morphological features was created.
• Six machine learning models (SVM, ANN, XGBoost, RF, KNN, Logistic Regression) were trained and optimized.
• SHAP and RFE analyses were used for feature importance and model validation.

Main Results:

• XGBoost model achieved the highest prediction accuracy (88.24%), surpassing junior clinicians (83.53%).
• Morphological features, particularly adjacent tooth angulation, contact area, and MM3 volume, were dominant predictors.
• Clinical factors like fibrinogen and prothrombin time also contributed to the predictions.

Conclusions:

• Integrating morphological and clinical features significantly enhances prediction accuracy for MM3 extraction difficulty.
• Adjacent tooth resistance emerged as the most influential factor, followed by bone resistance and mandibular canal proximity.