Prediction of the skeletal pattern of orthodontic patients

Abstract

It is important to know the skeletal pattern of anterior crossbite in orthodontic patients to overcome barriers with unnecessary radiation exposure, the high cost for the patients as well as to reduce time spent by the clinicians for the analysis of cephalogram in order to determine the skeletal pattern. Therefore, the main objective of this study is to propose a model to predict the skeletal pattern of anterior cross bite (ACB) in orthodontic patients. The association between skeletal pattern, mandibular displacement, tooth wear of involved teeth, the gingival recession of involved teeth, and degree of tooth mobility were checked with ACB using chi-squared test and Fisher’s exact test. Besides multinomial logistic regression model, multinomial probit model, and naïve Bayesian classification methods were used to predict the skeletal patterns and best model was selected based on the model accuracy, area under curve (AUC) value and F1-score. Multinomial logistic regression model (76%) and naïve Bayesian classifier (74%) has high accuracy when compared to multinomial probit model (64%) and further both these models show more than 95% of accuracy when predicting skeletal pattern II. The proposed multinomial logistic regression model has a higher AUC value and F1-score, and hence it is appropriate for predicting the skeletal pattern of ACB in orthodontic patients to maximize the treatment outcome, eliminate the unbearable cost on radiographs and to serve clinicians time.