Artifact Correction Method for M-response using UNet1D Network

Abstract

Translate Article

This article explores the application of convolutional neural networks (CNNs), specifically the UNet1D architecture, for filtering M-response signals − an essential component of electromyographic (EMG) diagnostics used to assess the functional state of the peripheral neuromuscular system. The goal of the study is to improve signal quality by effectively removing artifacts while preserving the amplitude-frequency characteristics critical for clinical interpretation. The model was trained on real biomedical data containing various types of noise and artifacts. Results show that the proposed model achieves high filtering accuracy with minimal deviation from the original signals (a mean amplitude deviation of only −0.04%, and a 20.75% increase in signal-to-noise ratio, SNR). The CNN-based approach demonstrated strong potential for preserving diagnostically significant features and can be effectively integrated into clinical EMG workflows to enhance diagnostic precision and reduce the impact of external interference.