A Comparative Evaluation of Deep Machine Learning Models in Orthodontic Clinical Outcomes: A Scoping Review.

Background and Aim: This review explores how applying deep machine learning models improves orthodontic diagnostic accuracy and treatment outcomes compared to traditional predictive models or manual human approaches alone. Methods: This scoping review followed the PRISMA-ScR checklist. We searched PubMed, Google Scholar, Cochrane Library, Science Direct, and DOAJ (Directory of Open Access Journals) for relevant studies from 2014 to September 2024 that were included, focusing on AI in orthodontic diagnosis, treatment planning, and treatment outcomes. Keywords used include: 'Artificial Intelligence', 'AI', 'Machine Learning', 'Deep Learning', 'Dentistry', 'Oral Health', 'Orthodontics', 'Braces', 'Clear Aligners'. Studies underwent multiple screenings, extracted relevant data, and conducted bias risk assessments. Results: Eleven studies were reviewed, showing AI's impact on improving diagnostic accuracy and treatment outcomes in orthodontics. Deep learning models report similar diagnostic accuracy with orthodontic gold standards and beat other traditional models regarding precision and accuracy scores. The accuracy scores of deep learning models ranged from 66.4% to 99.3%. Deep learning models also show potential to track tooth movement, decrease time spent by clinicians analyzing data, and ultimately improve experiences. However, challenges such as data scarcity, bias, and variability in performance during mixed dentition stages were identified. Conclusion: AI offers transformative potential in orthodontics, particularly in diagnosis and treatment planning. However, limitations regarding data availability and algorithmic reliability need to be addressed. Collaboration among dental professionals and AI researchers is crucial to overcoming these barriers and optimizing AI's application in orthodontic care.