Abstract
BACKGROUND: Radix entomolaris (RE), a distolingual supernumerary root in mandibular first molars, presents significant challenges in endodontic treatment due to its complex and variable anatomy. Conventional radiographic techniques often fail to detect such intricacies, increasing the risk of missed canals and treatment failure. OBJECTIVE: This study aimed to characterize the morphometric complexity of RE using cone-beam computed tomography (CBCT) and to develop machine learning (ML) models for classifying root canal morphotypes and predicting anatomical bifurcation. METHODS: One hundred extracted mandibular first molars with confirmed RE were scanned using high-resolution CBCT. Morphometric parameters-including canal curvature, cross-sectional area, roundness index (a circularity metric), volumetric size, and root fusion status (fused vs. separate roots)-were extracted at multiple apical levels. These features were used to train supervised and unsupervised ML models. A decision tree classifier predicted bifurcation presence using four anatomical features, while K-means clustering (k = 2) stratified morphotypes. RESULTS: The decision tree classifier achieved an F1-score of 0.87, with a sensitivity of 85.7 % and specificity of 89.4 %. Volumetric canal size was the strongest predictor of bifurcation (AUC = 0.81). K-means clustering identified two morphotypes: simple (round, single orifice) and complex (irregular, bifurcated, or C-shaped canals). Notably, roundness index decreased coronally, and 27 % of samples exhibited mid-root bifurcation. CONCLUSION: CBCT-derived features, particularly canal volume and curvature, effectively predict RE complexity. The integration of these features into machine learning models provides a clinically valuable framework for morphotype classification and personalized endodontic planning. These findings support the adoption of AI-assisted diagnostics in managing anatomically complex root canal systems.