Using Mechanomyograms for Estimating Median Frequency and Root Mean Square of Electromyograms With Hybrid Deep Neural Network and CatBoost Model

Abstract

The aim of this study is to estimate the median frequency (MDF) and root mean square (RMS) of surface electromyogram (sEMG) signal using a hybrid deep neural network (DNN) and CatBoost model from mechanomyogram (MMG) signal as a simple alternative approach. sEMG is measured from the skin over muscle to capture the electrical activity of motor units during muscle contraction, with MDF and RMS commonly used to assess the degree of muscle activity. MMG, on the other hand, reflects the mechanical oscillations of the skin resulting from muscle contractions, and is conveniently measured using a three-axis accelerometer. Twenty subjects were recruited to perform isometric contractions of the extensor digitorum in the left hand and the biceps brachii in the right hand. Five parameters-MDF, RMS, skewness, kurtosis and standard deviation-were extracted from the MMG to estimate MDF and RMS obtained from the sEMG. The results show that the Pearson correlation coefficients (PCCs) for MDF and RMS estimations were approximately 0.98 and 0.92, respectively. Therefore, the study concludes that accelerometer-based MMG can be used to estimate the MDF and RMS of sEMG, making it a viable tool for monitoring muscle activity during isometric contractions.

Keywords

• Electromyogram
• mechanomyogram
• hybrid deep neural network
• median frequency
• root mean square