Use of artificial intelligence to predict outcomes of nonextraction treatment of Class II malocclusions

Abstract

Maxillary molar distalization is a standard option for nonextraction treatment of Class II malocclusions. The purpose of this study was to develop an automated, artificial intelligence (AI) system to predict the dental, skeletal, and soft tissue changes after nonextraction treatment. Retrospective data and images were obtained for 284 patients who had had nonextraction treatment using modified C-palatal plates. Pre and posttreatment cephalograms were superimposed using the sella and the nasion as reference points. Cephalometric changes as a result of treatment were used for learning using convolutional neural networks (CNN). From the results, heatmaps were generated in the form of lateral cephalograms, representing the predicted amount of change for each landmark as a result of treatment. Six landmarks had a prediction error of < 1.0 mm: the A point, anterior nasal spine, subnasale, soft tissue A, pronasale, and columella. Ten landmarks, including the upper lip, labrale superius, and stomion superius, showed errors of 1.0-2.0 mm. Eleven landmarks, including the lower lip, soft tissue B, and stomion inferius, had errors of 2.0-3.0 mm. Only the soft tissue pogonion and menton landmarks had errors > 3.0 mm. Our AI model based on CNN architecture shows that it is possible to predict the cephalometric changes resulting from nonextraction treatment. The predicted changes expressed in the form of lateral cephalometric images may be useful visual guides for clinicians as well as patients when considering nonextraction treatment.