Development and accuracy of artificial intelligence-generated prediction of facial changes in orthodontic treatment: a scoping review.

Jiajun Zhu; Yuxin Yang; Hai Ming Wong

Return

Development and accuracy of artificial intelligence-generated prediction of facial changes in orthodontic treatment: a scoping review.

Author: Jiajun ZHU ¹ ; Yuxin YANG ² ; Hai Ming WONG ³
Author Information

1. Stomatology Hospital, School of Stomatology, Zhejiang University School of Medicine, Zhejiang Provincial Clinical Research Center for Oral Diseases, Key Laboratory of Oral Biomedical Research of Zhejiang Province, Cancer Center of Zhejiang University, Engineering Research Center of Oral Biomaterials and Devices of Zhejiang Province, Hangzhou 310000, China.
2. Faculty of Dentistry, The University of Hong Kong, Hong Kong, China.
3. Faculty of Dentistry, The University of Hong Kong, Hong Kong, China. wonghmg@hku.hk.
Publication Type:Journal Article
Keywords: Artificial intelligence (AI); Facial morphology; Orthodontic treatment; Soft-tissue changes
MeSH: Artificial Intelligence; Reproducibility of Results
From: Journal of Zhejiang University. Science. B 2023;24(11):974-984
CountryChina
Language:English
Abstract: Artificial intelligence (AI) has been utilized in soft-tissue analysis and prediction in orthodontic treatment planning, although its reliability has not been systematically assessed. This scoping review was conducted to outline the development of AI in terms of predicting soft-tissue changes after orthodontic treatment, as well as to comprehensively evaluate its prediction accuracy. Six electronic databases (PubMed, EBSCOhost, Web of Science, Embase, Cochrane Library, and Scopus) were searched up to March 14, 2023. Clinical studies investigating the performance of AI-based systems in predicting post-orthodontic soft-tissue alterations were included. The Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) and Joanna Briggs Institute (JBI) appraisal checklist for diagnostic test accuracy studies were applied to assess risk of bias, while the Grading of Recommendation, Assessment, Development, and Evaluation (GRADE) assessment was conducted to evaluate the certainty of outcomes. After screening 2500 studies, four non-randomized clinical trials were finally included for full-text evaluation. We found a low level of evidence indicating an estimated high overall accuracy of AI-generated prediction, whereas the lower lip and chin seemed to be the least predictable regions. Furthermore, the facial morphology simulated by AI via the fusion of multimodality images was considered to be reasonably true. Since all of the included studies that were not randomized clinical trials (non-RCTs) showed a moderate to high risk of bias, more well-designed clinical trials with sufficient sample size are needed in future work.