Condylar High-dimensional Image Feature Analysis for Skeletal Class Ⅲ Malocclusion

Abstract

<p>Objectives: This study aimed to detect the high-dimensional image features of mandibular condyles specific to patients with skeletal Class Ⅲ malocclusion.</p><p>Methods: Lateral cephalograms (LC) and craniofacial cone beam computed tomography (CBCT) images of 51 adult patients were collected. The patients were then diagnosed with skeletal Class Ⅰ, Ⅱ, or Ⅲ malocclusion according to the measurement on LCs. In CBCT images, the condyles were segmented using 3D Slicer. Among 102 condyles, 81 of them were randomly selected as the training set for the deep learning (DL) model and the remaining for the test set. The ViT3D model for the skeletal Class Ⅲ malocclusion identification task was developed. Additionally, following image segmentation, radiomic features were then extracted and selected. Those features with intraclass correlation (ICC) over 0.9 were included in the repeated measure analysis to compare the feature difference among groups at the significance level of 0.05.</p><p>Results: According to the cephalometric analysis, 18, 15 and 18 patients were diagnosed with skeletal Class Ⅰ, Ⅱ, and Ⅲ, with the ANB angle of 2.43±1.24, 7.87±1.91 and -2.21±1.52, respectively (P<0.001). The confounding variables, including age, gender, vertical facial pattern and mandibular deviation, were shown to have no significant difference among groups. For skeletal Class Ⅲ malocclusion identification, the VIT3D model showed acceptable accuracy (0.81) and area under the curve (AUC) (0.74). For the radiomic features, among 443 features included, 109 radiomic features showed significant differences among groups, which were related to the shape, intensity and texture.</p><p>Conclusions: In the DL model using CBCT images, the mandibular condyle may serve as a potential diagnostic target for identifying skeletal Class Ⅲ malocclusion. Further, variations in quantitative high-dimensional image features on condyles exist among different malocclusions, aiding in refining diagnostic models and enhancing our understanding of mandibular growth mechanisms.</p>