An explainable predictive model of direct pulp capping in carious mature permanent teeth

Area of Science:

• Dentistry
• Machine Learning
• Biomedical Informatics

Background:

• Direct pulp capping (DPC) is a conservative treatment for carious mature permanent teeth.
• Predicting DPC success is crucial for patient outcomes and treatment planning.
• Current prediction methods lack personalization and interpretability.

Purpose of the Study:

• To develop and validate an explainable machine learning (ML) model for predicting the personalized success probability of DPC treatment.
• To enhance clinical decision-making through interpretable AI in endodontics.

Main Methods:

• A retrospective cohort of 393 teeth from 372 patients undergoing DPC was analyzed.
• Six ML models were trained and validated, with feature importance assessed using Shapley Additive Explanation (SHAP).
• The top-performing XGBoost model was refined and translated into a web application.

Main Results:

• The DPC treatment failure rate was 9.67% at 1-year follow-up.
• The XGBoost model achieved an Area Under the Curve (AUC) of 0.86 for 1-year success prediction.
• An 11-feature interpretable XGBoost model was developed, prioritizing key predictive factors.

Conclusions:

• The developed XGBoost model provides a user-friendly tool for predicting personalized DPC success probability.
• Explainable AI (SHAP) offers transparent insights, supporting improved clinical decision-making and patient counseling.
• This approach enhances personalized dental care by integrating demographic and clinical data.