B-mode ultrasound radiomics of the masseter muscle for differentiating individuals with and without clinically defined probable bruxism.
Authors
Affiliations (2)
Affiliations (2)
- Department of Oral and Maxillofacial Radiology, Faculty of Dentistry, Istanbul Medeniyet University, Istanbul, Turkey. [email protected].
- Department of Oral and Maxillofacial Radiology, Faculty of Dentistry, Istanbul Medeniyet University, Istanbul, Turkey.
Abstract
To determine whether two-dimensional B-mode ultrasound radiomics of the masseter muscle can differentiate individuals with a clinically defined probable bruxism phenotype from those without clinical evidence of bruxism. Eighty-six participants were included: 44 with clinically defined probable bruxism and 42 without clinical evidence of bruxism. Bilateral masseter ultrasound images were obtained at rest and during clenching, yielding four images per participant and 344 manually segmented regions of interest. Two-dimensional radiomic features were extracted from calibrated exported B-mode images and aggregated into participant-level representations. LASSO logistic regression was used for exploratory feature selection and internal validation. Intraobserver segmentation reproducibility was assessed in 80 images using Dice similarity coefficients and feature-wise intraclass correlation coefficients. Feature extraction was completed for all 344 images without failed cases or missing values. The highest-ranking configuration used the participant-level overall mean of bilateral rest and clench images with texture plus two-dimensional shape features. This configuration achieved a nested cross-validation AUC of 0.926 ± 0.090. Aggregated participant-level cross-validated predictions for the primary model yielded an AUC of 0.941 (95% CI 0.887-0.982), accuracy of 0.884, sensitivity of 0.864, and specificity of 0.905. The demographic comparator model showed limited discrimination (AUC 0.553). Intraobserver reproducibility analysis showed excellent spatial mask overlap (Dice similarity coefficient, 0.91 ± 0.03), and 80 of 102 radiomic features (78.4%) demonstrated good or excellent reproducibility. Participant-level B-mode ultrasound radiomics showed promising internal discriminatory performance in distinguishing individuals with clinically defined probable bruxism from those without clinical evidence of bruxism in this single-centre cohort. Texture and two-dimensional shape features provided greater discriminatory information than age and sex alone. The workflow should be interpreted as an exploratory adjunctive imaging-phenotyping approach and requires independent external validation before clinical implementation.