ECG ARTIFACT REMOVAL FROM SINGLE-CHANNEL SURFACE EMG USING FULLY CONVOLUTIONAL NETWORKS Kuan-Chen Wang1Kai-Chun Liu2Sheng-Yu Peng3Yu Tsao2

2025-05-03 0 0 624.88KB 5 页 10玖币

侵权投诉

ECG ARTIFACT REMOVAL FROM SINGLE-CHANNEL SURFACE EMG USING FULLY

CONVOLUTIONAL NETWORKS

Kuan-Chen Wang1Kai-Chun Liu2Sheng-Yu Peng3Yu Tsao2

1Graduate Institute of Communication Engineering, National Taiwan University, Taiwan

2Research Center for Information Technology Innovation, Academia Sinica, Taiwan

3Department of Electrical Engineering, National Taiwan University of Science and Technology, Taiwan

r10942076@ntu.edu.tw, t22302856@citi.sinica.edu.tw, sypeng@mail.ntust.edu.tw, yu.tsao@citi.sinica.edu.tw

ABSTRACT

Electrocardiogram (ECG) artifact contamination often occurs in

surface electromyography (sEMG) applications when the measured

muscles are in proximity to the heart. Previous studies have de-

veloped and proposed various methods, such as high-pass ﬁltering,

template subtraction and so forth. However, these methods remain

limited by the requirement of reference signals and distortion of

original sEMG. This study proposed a novel denoising method to

eliminate ECG artifacts from the single-channel sEMG signals using

fully convolutional networks (FCN). The proposed method adopts a

denoise autoencoder structure and powerful nonlinear mapping ca-

pability of neural networks for sEMG denoising. We compared the

proposed approach with conventional approaches, including high-

pass ﬁlters and template subtraction, on open datasets called the

Non-Invasive Adaptive Prosthetics database and MIT-BIH normal

sinus rhythm database. The experimental results demonstrate that

the FCN outperforms conventional methods in sEMG reconstruction

quality under a wide range of signal-to-noise ratio inputs.

Index Terms—Deep neural network, ECG artifact removal,

fully convolutional network, single channel, surface electromyog-

raphy

1. INTRODUCTION

Surface electromyography (sEMG) noninvasively measures the ac-

tivation potentials of human muscles by attaching electrodes to the

skin. sEMG has been widely adopted in certain applications such

as in rehabilitation [1], stress monitoring [2], neuromuscular sys-

tem investigation [3], and prosthesis control [4]. During the data

collection, sEMG would be contaminated by the electrocardiogram

(ECG) if the measured muscles are in proximity to the heart [5, 6].

ECG contamination distorts the amplitude and frequency spectrum

of sEMG, which may further deteriorate the effects of sEMG appli-

cations or hinder the determination of relevant information in sEMG

signals. Hence, it is crucial to develop denoising techniques and dis-

card ECG artifacts from the sEMG signals.

The general frequency bands of sEMG (10-500 Hz) and ECG

(0-100 Hz) partially overlap [7]. This triggers difﬁculties in dis-

carding ECG artifacts without deteriorating the sEMG quality. To

address this problem, various methods have been developed, includ-

ing high-pass ﬁlters (HP) with different cutoff frequencies, template

subtraction (TS), adaptive ﬁlter (AF) and independent component

analysis (ICA). Although these methods have been comprehensively

compared and analyzed in previous studies [8, 9], they remain lim-

ited. For instance, applying high-pass ﬁlters would eliminate the

low-frequency part of sEMG signals; TS works effectively because

sEMG is assumed to be a zero-mean Gaussian distribution that is

not fully satisﬁed in an actual environment. Other denoising meth-

ods require the use of reference signals. For example, AF requires

an additional clean ECG reference [10], and ICA is more suitable for

applications with multiple sEMG channels [11].

Recently, neural networks (NNs) have been widely applied in

the development of signal enhancement and denoising approaches

in different applications owing to their powerful nonlinear map-

ping capability, such as acoustic signals [12, 13], ECG [14, 15], and

EEG [16, 17] noise removal. In these studies, different deep-learning

models were developed for signal enhancement. Some commonly

adopted models are the multilayer perceptron (MLP) [12, 17], con-

volutional neural networks (CNNs) [17], fully convolutional neural

networks (FCNs) [13, 15, 18], and long short-term memory mod-

els [17, 19]. These NN-based signal enhancement methods have

achieved extraordinary results in improving signal quality when

compared with conventional denoising methods.

Although previous research has utilized NN to process sEMG

signals for other classiﬁcation tasks (e.g., gesture classiﬁcation [20,

21] or noise-type identiﬁcation [22, 23]), few studies have explored

the feasibility of NN for contamination removal in sEMG [24].

Therefore, this study adopted an NN and proposed an FCN-based

denoising method to eliminate ECG artifacts from sEMG signals.

FCN can handle single-channel sEMGs without other reference sig-

nals, e.g., other sEMG channels or an ECG channel. Furthermore,

the proposed approach can directly process raw sEMG signals with-

out further data transformation, such as short-time Fourier transform

(STFT), which reduces the computational complexity and facilitate

near-real-time applications. The experimental results indicate that

the FCN-based denoising approach exhibits a better ﬁlter ability than

conventional methods in the removal of ECG artifacts from sEMG

signals. Owing to these advantages, the proposed method may be a

suitable choice for eliminating ECG contamination in sEMG.

2. RELATED WORK

2.1. Conventional ECG artifact removal methods

Several methods have been developed to suppress ECG artifacts

in sEMG, including the HP, TS, AF and ICA [8, 9, 25]. HP di-

rectly eliminates the low-frequency part of noisy sEMG signals,

where most ECG frequency components exist. TS is based on

the quasiperiodic properties of ECG and the assumption that the

sEMG has a zero-mean Gaussian distribution. It attempts to elimi-

nate the ECG waveforms from the noisy sEMG signals in the time

domain. To develop ECG waveform templates, ECG waveform

detection is initially applied to noisy sEMG signals. Subsequently,

arXiv:2210.13271v1 [eess.SP] 24 Oct 2022

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

ECGARTIFACTREMOVALFROMSINGLE-CHANNELSURFACEEMGUSINGFULLYCONVOLUTIONALNETWORKSKuan-ChenWang1Kai-ChunLiu2Sheng-YuPeng3YuTsao21GraduateInstituteofCommunicationEngineering,NationalTaiwanUniversity,Taiwan2ResearchCenterforInformationTechnologyInnovation,AcademiaSinica,Taiwan3DepartmentofElectricalEnginee...

展开>> 收起<<

ECG ARTIFACT REMOVAL FROM SINGLE-CHANNEL SURFACE EMG USING FULLY CONVOLUTIONAL NETWORKS Kuan-Chen Wang1Kai-Chun Liu2Sheng-Yu Peng3Yu Tsao2.pdf

共5页,预览1页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

ECG ARTIFACT REMOVAL FROM SINGLE-CHANNEL SURFACE EMG USING FULLY CONVOLUTIONAL NETWORKS Kuan-Chen Wang1Kai-Chun Liu2Sheng-Yu Peng3Yu Tsao2

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: