Machine learning models for prognosis prediction in regenerative endodontic procedures

This study aimed to establish and validate machine learning (ML) models to predict the prognosis of regenerative endodontic procedures (REPs) clinically, assisting clinicians in decision-making and avoiding treatment failure.

A total of 198 patients with 268 teeth were included for radiographic examination and measurement. Five Machine Learning (ML) models, including Random forest (RF), gradient boosting machine (GBM), extreme gradient boosting (XGB), Logistic regression (logR) and support vector machine (SVM) are implemented for the prediction on two datasets of follow-up periods of 1-year and 2-year, respectively. Using a stratified five folds of cross-validation method, each dataset is randomly divided into a training set and test set in a ratio of 8 : 2. Correlation analysis and importance ranking were performed for feature extraction. Seven performance metrics including area under curve (AUC), accuracy, F1-score, sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were calculated to compare the predictive performance.

The RF (Accuracy = 0.91, AUC = 0.94; Accuracy = 0.84, AUC = 0.86) and GBM (Accuracy = 0.91, AUC = 0.93; Accuracy = 0.84, AUC = 0.85) had the best and similar performance simultaneously in the prediction of 1-year follow-up period and 2-year follow-up period, respectively. The variables applied to predict the primary outcome in REPs were ranked accordingly to their values of feature importance, including age, sex, etiology, the number of root canals, trauma type, swelling or sinus tract, periapical lesion size, root development stage, pre-operative root resorption, medicaments, scaffold, second REPs, previous root canal filling.

RF and GBM models outperformed XGB, logR, SVM models on the overall performance on our datasets, exhibiting the potential capability to predict the prognosis of REPs. The ranking of feature importance contributes to establishing the scoring system for prognosis prediction in REPs, assisting clinicians in decision-making.