An Artificial Intelligence-Based Approach With Photoplethysmogram and Heart Rate Variability for Sleep Bruxism Diagnosis


Eris O., Bozkurt M. R., Eris S. B., Bilgin C.

IEEE Access, 2025 (SCI-Expanded) identifier

  • Yayın Türü: Makale / Tam Makale
  • Basım Tarihi: 2025
  • Doi Numarası: 10.1109/access.2025.3546720
  • Dergi Adı: IEEE Access
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, INSPEC, Directory of Open Access Journals
  • Anahtar Kelimeler: Artificial intelligence, biomarkers, feature extraction, feature selection, heart rate variability, leave-one-out, machine learning, photoplethysmogram, principal component analysis, sleep bruxism
  • Bursa Uludağ Üniversitesi Adresli: Evet

Özet

Bruxism is jaw muscle activity that can cause functional and aesthetic changes in the jaw and dental structures of individuals. It can be observed during sleep or while awake. Owing to the disadvantages of using polysomnography (PSG) for the definitive diagnosis of bruxism, it is important to develop alternative and reliable diagnostic systems. In this study, we propose a noninvasive and practical solution for diagnosing sleep bruxism using photoplethysmogram (PPG) and heart rate variability (HRV). We created a database by extracting features from PPG and HRV. We used Principal Component Analysis (PCA) to reduce the data size and Fisher's feature selection algorithm to identify the most important features. Using four artificial intelligence (AI) algorithms, we built classification models that distinguished bruxism labels from control labels. The models were optimized and tested using unseen data. We evaluated the performance of the models using six criteria and validated them using leave-one-out (LOO). Singular Value Decomposition (SVD) of HRV is the most important biomarker for separating bruxism from control data. In addition, the durations of the falling and rising edges of the PPG and the amplitude values of these durations in some percentiles are also important for increasing classification success. The results, methodology used, and easy acquisition of PPG and HRV make it possible to apply the proposed model to embedded systems. The use of the proposed model in clinical evaluation provides significant advantages. This study is innovative in the literature and sheds light on future studies.