AI in Orthodontics, Where Are We And Where Are We Going 10 MINUTE SUMMARY

orthodontics In summary

Join me for a podcast summary looking at Ai in orthodonticsand its clinical application. A growing topic in orthodontics, and one of themost featured topics at this years AAO. This summary is based on 3 lectures fromthis year’s summer meeting by Juan Francisco Gonzalez & Jean Marc Retrouvey,Tarek ElShebiny , Jonas Bianchi and Lucia Cevidanes. We will look whatAi is, the way it works and its clinical application, as well as a criticalview on this young field.

What is Ai:

1. Technology that enables computers and machinesto simulate human intelligence, perform 1 task very well, e.g. voice command, Youtuberecommendations

2. Predictive modelling, makes calculations, convert information into numbers or categoriesand recognise patterns

Levels of Ai: Machine learning, Neural Networks and Deep Learning

1. Machine learning

a. The ability for a machine to learn from data andpast experience to identify patterns and make predictions

2. Neural Networks

a. Specific model which relies on interconnectednodes, which perform a mathematical calculation of associations , patterns, andprobabilities

3. Deep learning

a. Is a complex version of neural networks

Virtual patient

· CBCT segment + STL file – segmentation of theteeth and roots, with labelling of different stuctures

o Can print model, visualise ideal vector andcalculate ideal vector

o However clinician still required to establish biomechanics

· CBCT integration for aligner cases, Unpublishedthesis Khalid Alotaibi:

o Treatment planning confidence increased 50%, leastchange was treatment planning modification

Diagnostic data:

· Ai cephalometric tracing

o 46% of 24 landmarks 2.0mm within

o 4 different programmes Iortho, Webceph, Orthodc, cephx

o All landmarks had good overall agreement butvariation in identification

· Facial Analysis

· Automated 3D facial asymmetry analysis usingmachine learning Adel 2025

o Study – 7 landmarks

o Identified manually and with deep learning

o 5 accurate, 2 significant difference but notclinically relevant

Diagnostic accuracy of photos

· Clinical photos assessment by Ai, and comparedto clinical examination

· Sensitivity 72%, specificity 54% Vaughan & Ahmed2025

Growth prediction

· Poor agreement age 9

Comparison between direct, virtual and AI bonding

· DIBs – uses Ai for bonding

· Compare Ai Vs user modified indirect bonding Vsdirect bonding (gold standard), 0.5mm significant

· Incisors accurate

· Premolars and lower laterals inaccurate

Monitoring

Previous podcast exploring the accuracy of remote monitoring

o with Ferlito 2022 80%repeatability from 2 scans 44.7% repeatability and reproducibility

Bracket removal from scan and retainer fit

Tarek Assessment of virtual bracket removal by artificialintelligence and thermoplastic retainer fit AJODO 2024

o Retainers for both – clinically acceptable

FDA approval of Ai in dentistry

· FDA - Software of Medical Diagnosis

§ 4 dental:

· Dental Monitoring

· Ray Co

· X-Nav technologies

· Densply Sirona

What’s next

· More data learning to train AI model

· Robotics customising appliances per patient