Machine Learning (ML) and Deep Learning (DL), in general, are radically changing the way dental procedures, orthodontics in particular, are done, with the use of Artificial Intelligence (AI). These technologies improve the process of planning treatment, allow for diagnosing conditions correctly, and simplify clinical processes. Using big data, orthodontists are able to use AI to automate their daily clinical work and come up with specific treatment plans that could suit a particular patient. On the clinic side, the AI is used to support important functions, including patient recruitment, outcome optimization, and better clinical decision-making. However, several gaps remain in our knowledge of AI applications in orthodontics; namely, there will be a lack of systematic categorization and ranking of these applications. Therefore, scholars and practitioners have realized the need to have an integrated framework that categorizes AI methods and the extent of their effectiveness in clinical practice. In order to overcome these shortcomings, we organized a Systematic Literature Review (SLR) of the available literature by concentrating on methodological tendencies and AI-based applications. We have conducted a review of the current studies and a critical analysis of those studies, and then categorized them as four clinical groups: hybrid or custom systems, imaging-based applications, diagnostic tools, and treatment-planning solutions. Each category underwent a comprehensive search, with the main results being derived to help to understand the level of AI integration in the orthodontic practice. According to our findings, the main areas of AI research in orthodontics are AI-driven treatment planning and decision-making (13.8%), dental/arch-form classification (17.2%), and diagnosis and analysis (roughly 24.1%). It demonstrates that healthcare providers are seeking better ways to diagnose and treat their patients. Nearly 46.7% of approaches are ML, which is the most common type. About 33.3% of the time, DL is the second most common type. Hybrid models are used in about 20% of applications. But there are still major shortcomings: very few studies incorporate explainable AI (XAI), and most of them use local datasets (65.6%) with mostly medium (48.3%) or small (37.9%) sample sizes. This means that we are unable to utilize the results in other situations, which shows that we need bigger, more uniform datasets and better validation in clinical settings.
Sedaghat et al. (Mon,) studied this question.