Predictive modeling of soft tissue changes after orthodontic treatment with premolar extractions using frontal facial photographs

The present study aimed to determine the accuracy of machine learning in predicting soft tissue changes after orthodontic treatment after the extraction of premolars using frontal facial photographs.

Pretreatment (T0) and posttreatment (T1) frontal photographs of 100 patients (70 females and 30 males) aged 17-28 years who underwent orthodontic treatment with premolar extraction were collected according to the inclusion and exclusion criteria. Images were digitized, and landmarks were annotated using the Visual Geometry Group image annotator (Visual Geometry Group, Department of Engineering Science, University of Oxford, Oxford, United Kingdom). A high-resolution network model was developed to extract and predict key points using the affine transformation technique. The mean difference between the actual and predicted models in the frontal measurements was assessed. The accuracy of the predicted model in assessing soft tissue changes was evaluated using the mean absolute error and the mean absolute percentage error.

No statistically significant difference was found between the actual and predicted models in the soft tissue changes from T0 to T1. The overall accuracy of the predicted model in assessing soft tissue changes was 85.36%. However, the accuracy for predicting interlabial gap was notably lower (52.75%).

Soft tissue treatment changes were predicted with moderate-to-high accuracy based on 2-dimensional frontal photographs, offering a noninvasive and clinically applicable solution. The high-resolution network model and 2-dimensional affine transformation technique ensured accuracy and reproducibility, highlighting the potential for artificial intelligence-driven automation in orthodontic diagnostics.