Applying artificial intelligence to predict the outcome of orthodontic treatment

Abstract

Objectives: The study aimed to train an algorithm to predict facial and dental outcomes following orthodontic treatment using artificial intelligence (AI). In addition, the accuracy of the algorithm was evaluated by four distinct groups of evaluators. Material and Methods: The algorithm was trained using pre-treatment and post-treatment frontal smiling and intraoral photographs of 50 bimaxillary patients who underwent all first bicuspid extraction and orthodontic treatment with fixed appliances. A questionnaire was created through Google form and it included 10 actual post-treatment and AI-predicted post-treatment images. The accuracy and acceptability of the AI-predicted outcomes were analyzed by four groups of 140 evaluators (35 orthodontists, 35 oral maxillofacial surgeons, 35 other specialty dentists, and 35 laypersons). Results: The Style-based Generative Adversarial Network-2 algorithm used in this study proved effective in predicting post-treatment outcomes using pre-treatment frontal facial photographs of bimaxillary patients who underwent extraction of all first bicuspids as part of their treatment regimen. The responses from the four different groups of evaluators varied. Laypersons exhibited greater acceptance of the AI-predicted images, whereas oral maxillofacial surgeons showed the least agreement. The base of the nose and the chin demonstrated the most accurate predictions, while gingival visibility and the upper lip-to-teeth relationship exhibited the least prediction accuracy. Conclusion: The outcomes underscore the potential of the method, with a majority of evaluators finding predictions made by the AI algorithm to be generally reliable. Nonetheless, further research is warranted to address constraints such as image tonicity and the proportional accuracy of the predicted images.