EMG Signal Evaluation by Graph Signal Processing & Total Variation Denoising

Abstract

Electromyography (EMG) serves as a vital diagnostic tool in medical and clinical research, enabling the monitoring and analysis of muscle electrical activity. In medical diagnostics, EMG aids in identifying and assessing neuromuscular syndromes, i.e. amyotrophic lateral sclerosis (ALS). However, EMG signals are prone to various forms of noise and interference, posing challenges to accurate data interpretation. Thus, the development of robust denoising techniques is crucial for enhancing EMG signal quality and addressing practical challenges in clinical diagnostics, rehabilitation, and neuromuscular research. This research introduces an innovative methodology integrating Variational Mode Decomposition (VMD) and Graph Signal Processing (GSP) to improve EMG signal quality. Unlike conventional approaches like Continuous Wavelet Transform (CWT), this study explores the untapped potential of VMD with Intrinsic Mode Functions (IMFs) 16 and GSP in EMG signal analysis. sEMG data collected from 10 subjects using the EMGUSB (OT Bioelettronica) underwent denoising techniques, specifically CWT, VMD, and GSP. Evaluation of noise reduction performance reveals compelling results, with GSP demonstrating superior noise reduction capabilities compared to VMD and CWT. Specifically, GSP increases the SNR by 259.15 meanwhile decreases the RMSE by 0.07. In comparison, VMD upturns SNR with 111.56 and declines RMSE of 0.15. While both VMD and GSP outperform CWT, which exhibits SNR enhancements of 90.46 and RMSE reductions by 0.15. Statistical analysis validates the significant improvements (p < 0.05) provided by VMD and GSP over CWT across varying noise levels. Notably, VMD and GSP collectively exhibit substantial enhancements in both SNR and RMSE metrics, underscoring their efficacy in preserving signal fidelity while minimizing noise and artifacts.