PREDICTION OF CERVICAL DISC HERNIATION DISEASE UTILIZING TRAPEZIUS sEMG SIGNALS WITH MACHINE LEARNING TECHNIQUES BASED ON FREQUENCY DOMAIN FEATURE EXTRACTION

Abstract

Cervical disk herniation (CDH) is a disease that affects the quality of life of many people due to the neck pain it causes. The aim of this study was to develop an automatic prediction system to aid in diagnosis by evaluating the change in the surface electrical activity of the trapezius muscle in SDH disease in order to find an answer to the question: 'Can the surface electromyogram (sEMG) recorded from the trapezius muscle be an effective indicator for the diagnosis of SDH disease?'. To this end, a dataset will be created using preprocessing and feature extraction methods from sEMG signals from CDH patients and healthy individuals. In the first step, the Savitsky-Golay filter is used to denoise the sEMG signals and the dominant frequency signals between 20 and 150 Hz are included in the study using the Butterworth filter design. Twenty PSD-based features in the frequency domain were then obtained from the signals to which we applied the Burg method. Eleven of the most significant features based on the information gain, gain ratio, and Gini values are selected to be submitted to the classifiers. 80% of all new feature areas are used for classification and the rest for prediction. The best classification accuracy of 91.6% was obtained with the Tree classifier using 10-fold cross-validation for classification. In addition, neural networks and CN2 rule inducer provided 87.5% classification accuracy for prediction using 20% of the remaining data that the classifiers had not seen before. The experimental results demonstrate that the trapezius muscle has different surface electrical activity in CDH patients and healthy subjects and that the frequency domain characteristics of this activity are important for disease prediction.