Abstract:

Objective To investigate the risk factors associated with mandibular third molar eruption in adolescents undergoing orthodontic treatment and to construct a predictive model utilizing a Logistic regression-back propagation neural network (LR-BPNN) algorithm. Methods A retrospective cohort study was conducted, which included 565 adolescent patients who had completed orthodontic treatment and underwent two consecutive cone-beam computed tomography (CBCT) scans at the Affiliated Stomatological Hospital of Nanjing Medical University. Parameters related to mandibular third molar eruption were collected. Predictors with statistical significance were first selected through Logistic regression analysis and then served as input features for a back propagation neural network (BPNN), thereby forming the LR-BPNN hybrid model. Its performance was evaluated and compared on an independent validation set using metrics such as the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity. Results Logistic regression analysis identified six core influencing factors. The established LR-BPNN hybrid model demonstrated the best predictive performance, achieving an accuracy of 89.94% and an AUC of 0.901, which was significantly higher than those of the conventional Logistic regression model (AUC=0.755) and the full variable BPNN model (AUC=0.893) (P<0.01). Conclusion This study successfully developed and validated an LR-BPNN hybrid model for predicting mandibular third molar eruption. The model exhibits high predictive accuracy and clinical practicality, providing an objective and quantitative decision support tool for assessing impaction risk and formulating personalized extraction and space management strategies in orthodontic practice.

Key words: mandibular third molar eruption, Logistic regression-back propagation neural network (LR-BPNN), intelligent clinical decision support