Classification of Bruxism Presence Using Machine Learning Based on Radiomic Features Extracted from Ultrasound Images

Sevim F. A., Büyük B., Duyan Yüksel H., Soydan Çabuk D., Evlice B.

DDS Global Congress 2025 | joint IADMFR Meeting, Venice, İtalya, 16 - 18 Ekim 2025, ss.10-11, (Özet Bildiri)

  • Yayın Türü: Bildiri / Özet Bildiri
  • Basıldığı Şehir: Venice
  • Basıldığı Ülke: İtalya
  • Sayfa Sayıları: ss.10-11
  • Çukurova Üniversitesi Adresli: Evet

Özet

Abstract:  

Classification of Bruxism Presence Using Machine Learning Based on Radiomic Features Extracted from Ultrasound Images

Dt Fatma Aybike Sevim (1),Dt Beyzanur Büyük (1), Dr Hazal Duyan Yüksel (1),Dr Damla Soydan Cabuk (1),Dr Burcu Evlice (1)

(1) Cukurova University Faculty of Dentistry, Adana, Turkey

Aim

This study aimed to develop a machine learning (ML) model to differentiate individuals with and without bruxism by analyzing radiomic features extracted from ultrasonographic images of the masseter and anterior temporalis muscles.

Materials and Methods  

The study included age- and sex-matched participants with and without a history of bruxism. Bilateral high-resolution ultrasound images of the masseter and temporalis muscles were acquired. Manual segmentation was performed using 3D Slicer software. As part of the preprocessing pipeline, high-pass and low-pass wavelet filters were applied to enhance textural detail. Radiomic features were extracted using the pyRadiomics library and subsequently standardized. Feature selection was carried out in three stages: VarianceThresholding (threshold = 0.8), SelectKBest ranking, and LASSO regression. Four supervised classification algorithms—k-nearest neighbors (KNN), logistic regression (LR), support vector machine (SVM), and random forest (RF)—were trained and evaluated. Model performance was assessed using standard metrics, including area under the curve (AUC), sensitivity, specificity, accuracy, precision, and F1-score.

Results

The optimal features selected through the feature selection pipeline demonstrated statistically significant differences between individuals with and without bruxism (p < 0.001). The machine learning models successfully identified bruxism with high diagnostic accuracy.

Conclusion

Conclusion Combining ultrasound-based radiomic analysis with machine learning presents a promising, non-invasive strategy for identifying muscle alterations associated with bruxism. Further validation studies on larger datasets are needed.

Keywords: Bruxism, Radiomics, Ultrasound Imaging, Machine Learning, Masseter Muscle, Temporalis Muscle 

DT FATMA AYBİKE SEVİM.jpg