Predicting case difficulty in endodontic microsurgery using machine learning algorithms.

Area of Science:

• Dentistry
• Machine Learning
• Surgical Planning

Background:

• Preoperative assessment is vital for identifying risks in endodontic microsurgery.
• Accurate case difficulty prediction enhances surgical planning and patient outcomes.

Purpose of the Study:

• To develop and validate machine learning models for predicting case difficulty in endodontic microsurgery.
• To assist clinicians with preoperative analysis and decision-making.

Main Methods:

• Cone-beam computed tomographic images from 341 teeth were analyzed.
• Linear regression (LR), support vector regression (SVR), and extreme gradient boosting (XGBoost) models were developed.
• Model performance was evaluated using metrics like MAE, MSE, EVS, and R² with five-fold cross-validation.

Main Results:

• The XGBoost model exhibited the lowest error metrics (MAE, MSE, MedAE) and highest performance scores (EVS, R²).
• Key predictors for case difficulty included lesion size, proximity to anatomical structures, and root filling characteristics.
• Feature importance analysis identified critical factors for assessing endodontic microsurgery complexity.

Conclusions:

• The XGBoost model significantly outperformed LR and SVR models in predicting endodontic microsurgery case difficulty.
• This model can enhance preoperative analysis, offering a valuable tool for clinicians.
• Identified feature importance can inform the development of standardized case difficulty scoring systems.