Machine learning in orthodontics: Automated facial analysis of vertical dimension for increased precision and efficiency

Abstract

The digitization of dentistry has brought many opportunities to the specialty of orthodontics. Advances in computing power and artificial intelligence are set to significantly impact the specialty. In this article, the accuracy of automated facial analysis for vertical dimensions using machine learning is evaluated. Automated facial analysis of 45 patients (20 female, 25 male) was conducted. The subjects' ages were between 15 and 25 years (mean, 18.7; standard deviation, 3.2). A python program was written by the authors to detect the faces, annotate them and compute vertical dimensions. The accuracy of the manual annotation of digital images was compared with the proposed model. Intrarater and interrater reliability were evaluated for the manual method, whereas intraclass correlation and the Bland-Altman analysis were compared with manual and automated methods. The authors found acceptable intrarater reliability and moderate to poor interrater reliability for the manual method. The agreement was found between manual and automated methods of facial analysis. The 95% confidence interval limit of agreements was<10% for the metrics assessing vertical dimension. Machine learning offers the ability to conduct reliable and easily reproducible analyses on large datasets of images. This new tool presents opportunities for further advances in research and clinical orthodontics.