Deep Learning-Assisted Maxillary Canine Impaction Prediction On Panoramic Radiographs

Abstract

Aim or purpose: The purpose of this study is to create a novel predictive model for impacted canines and to assess its predictive accuracy in external validation against existing models,facilitated by a deep learning-based automated landmark detection system. Materials and methods: Patients aged 7 to 14 who underwent panoramic radiographic examinations between 2020-2023 and received a diagnosis of impacted canines were included in the study. An automatic landmark localization system was employed to assist the measurement of geometric parameters on the panoramic radiographs. Then we developed a new prediction model and compared the outcomes of existing models. The metrics of accuracy, sensitivity, specificity, precision and area under the ROC curve (AUC) were used to compare the performance of different models. A predictive model for the impaction of maxillary canine teeth was developed using multifactorial logistic regression. This model was validated in internal and external validation cohorts, to assess its diagnostic efficacy. Results: The new model demonstrated favorable diagnostic metrics, with a sensitivity of 98%, a specificity of 83% (AUC of 0.98).Margot et al. achieved a sensitivity of 95% and a specificity of 86% (AUC = 0.97), followed by the model by Arnautska et al, with a sensitivity of 93% and a specificity of 71% (AUC = 0.94) . The model by Alqerban et al. showed poor performance with an AUC of only 0.20. Conclusions: The new model achieved high accuracy and addressed the logical and inclusivity shortcomings of previous models. Deep learning has significantly facilitated research in predicting maxillary canine impaction.