Machine learning prediction of canal transportation using micro-CT data

Root canal transportation remains a significant complication in endodontic treatment because current assessment methods cannot predict transportation risk prior to instrumentation. Therefore, it is of interest to develop and validate machine learning models to predict the magnitude and direction of canal transportation using pre-operative micro-CT-derived morphometric features. Hence, a total of 120 mandibular molars with moderate-to-severe canal curvature were scanned pre- and post-instrumentation and seventeen morphometric variables were used to train four machine learning algorithms with five-fold cross-validation. The gradient boosting model demonstrated the best performance, with a coefficient of determination of 0.87, mean absolute error of 0.031 mm and root mean square error of 0.042 mm in predicting apical transportation. Thus, machine learning models based on pre-operative micro-CT data can accurately predict canal transportation and may aid in risk assessment and selection of optimal instrumentation strategies in endodontic practice.