DOI
https://doi.org/10.25772/76E5-DT10
Defense Date
2025
Document Type
Thesis
Degree Name
Master of Science in Dentistry
Department
Dentistry
First Advisor
Eser Tufekci
Second Advisor
Caroline K. Carrico
Third Advisor
Parthasarathy A. Madurantakam
Abstract
Background: Artificial Intelligence (AI) and machine learning (ML) may offer innovative solutions for the virtual planning of clear aligner therapy, enhancing precision and efficiency in orthodontic treatment. The purpose of this study is to assess the accuracy and reliability of an AI-based clear aligner system's virtual treatment simulations (predictions).
Purpose: The primary objective of this observational clinical study is to assess the accuracy of a virtual treatment plan utilizing ML algorithms by comparing the predicted tooth movements with the actual clinical movements at three and six months. The secondary objective is to determine whether the use of cone beam computed tomography and machine learning algorithms accurately predicts alveolar bone defects at six months of treatment.
Methods: The models of the actual Crown-Only positions at three months and the Crown-Only and Crown-Root positions at six months were superimposed on the digital images of predicted tooth positions using Geomagic Control X Software. The mean absolute difference (MAD), root mean square (RMS), and percent within tolerance between the models was calculated to examine how well the software predicted tooth movements. Descriptive analysis was used to estimate the presence of bone defects.
Results: Nine patients were recruited for this study, but only six completed and were included in the analyses, yielding a total of 12 arches. The mean absolute difference and root mean square were significantly higher for 6m Crown-Only than Crown-Root models. The percent within tolerance was higher at three months than six months for the Crown-Only models (71% vs. 63%), and higher for the Crown-Only models compared to the Crown-Root models at 6m (63% vs. 57%). The ML software predicted 14 instances of bone defects (8%) out of 168 sites. ML software predicted defects in 11 sites where none were present clinically (7%) and failed to predict defects in 3 sites where they were observed clinically (2%).
Conclusions: AI-based clear aligners predicted tooth movements with a slight improvement in accuracy when compared to traditional clear aligners that do not utilize this technology when developing a virtual treatment plan. The predictions using Crown-Root model superimpositions were less accurate than those using the Crown-Only model superimpositions at six months. ML software demonstrated a tendency to overestimate the presence of bone defects.
Rights
© The Author
Is Part Of
VCU University Archives
Is Part Of
VCU Theses and Dissertations
Date of Submission
5-6-2025