This thesis presents the detection and analysis of Teeth Grinding and Clenching by using Surface Electromyography (sEMG). The question of the existence of a relationship between data and conscious teeth grinding and clenching is discussed in this thesis, and will be answered in relation to the relevant research questions. Previous work has shown that sEMG along with machine learning techniquesare suitable for the detection of medical disorders e. g. gait disorders, for the development of electronic wheelchairs or even prostheses. Most techniques for the detection of bruxism, which is defined as extreme teeth grinding and clenching, focus on the determination of threshold values for automatic detection. In this thesis the focus is set on finding appropriate machine learning algorithms for the classification of EMG signals acquired from the temporalis muscle. It is classified whether teeth are being grinded and clenched or not. The potential of Logistic Regression, Support Vector Machine and Random Forest classifiers along with different sets of features is evaluated. Additionally, it is examined whether classifier calibration can improve the model in terms of generalizability. After evaluation of different techniques it is empirically shown that the random forest model with a feature set of eight time-domain features works best. It is also shown, that calibration with Isotonic Regression is better suitable for the detection of teeth grinding and clenching than calibration using Platt Scaling. However, the calibration curves presented later in this thesis show that both techniques are not optimal.




  • No labels