Prediction of Favorability of Maxillary Canine Impaction Using Artificial Intelligence Algorithm

Abstract

Objective: To test the accuracy of an artificial intelligence (AI) algorithm in predicting the favorability of maxillary canine impaction as compared to the conventional manual tracing method using orthopantomograms. Materials and methods: Orthopantomograms of 437 canine impactions were included in this study. Six parameters (sector classification, three angular, and two linear) on orthopantomograms were used to assess the severity of the maxillary canine impaction. The most advanced convolutional neural network model implemented using MATLAB program was used in assessing the favorability of maxillary canine impaction. The outcome of the parameters assessed by the implemented AI model and conventional manual tracing were compared. Receiver operating characteristic (ROC) curves, diagnostic accuracy, sensitivity, specificity, and positive and negative predictive values were used to assess the performance of AI model in comparison to manual tracing for the angular and sector parameters. Paired t-tests were used for linear measurements. Results: The overall clinical performance exceeded 90% for all the angular parameters including sector classification except for the angle between long axis of canine with lateral incisor which had a specificity score of 55%. The value and the area under the ROC curve were more than 0.9 for all the parameters. The distance from the canine cusp tip to the occlusal plane and midline was statistically significant between the groups ( p = .000). Conclusion: The proposed AI algorithm had higher accuracy in predicting the favorability of eruption in maxillary canine impactions compared to conventional manual tracing.