Background and Objectives: We aimed to develop a predictive model for the outcome
of bruxism treatments using ultrasonography (USG)-based machine learning (ML) techniques.
This study is a quantitative research study (predictive modeling study) in which different
treatment methods applied to bruxism patients are evaluated through artificial intelligence.
Materials and Methods: The study population comprised 102 participants with bruxism
in three treatment groups: Manual therapy, Manual therapy and Kinesio Tape or Botulinum
Toxin-A injection. USG imaging was performed on the masseter muscle to calculate muscle
thickness, and pain thresholds were evaluated using an algometer. A radiomics platform
was utilized to handle imaging and clinical data, as well as to perform a subsequent
radiomics statistical analysis. Results: The area under the curve (AUC) values of
all machine learning methods ranged from 0.772 to 0.986 for the training data and
from 0.394 to 0.848 for the test data. The Support Vector Machine (SVM) led to excellent
discrimination between bruxism and normal patients from USG images. Radiomics characteristics
in pre-treatment ultrasound scans of patients, showing coarse and nonuniform muscles,
were associated with a greater chance of less effective pain reduction outcomes. Conclusions:
This study has introduced a machine learning model using SVM analysis on ultrasound
(USG) images for bruxism patients, which can detect masseter muscle changes on USG.
Support Vector Machine regression analysis showed the combined ML models can also
predict the outcome of the pain reduction.
