Multimodality Multi-Lead ECG Arrhythmia Classiﬁcation using Self-Supervised Learning 1stThinh Phan

2025-04-26 0 0 2.21MB 4 页 10玖币

侵权投诉

Multimodality Multi-Lead ECG Arrhythmia

Classiﬁcation using Self-Supervised Learning

1st Thinh Phan

Department of AI Convergence

Chonnam National University

Gwangju, South Korea 61186

phantrandacthinh2382@gmail.com

2nd Duc Le

Department of CSCE

University of Arkansas,

Fayetteville, Arkansas 72703

minhducl@uark.edu

3rd Patel Brijesh

Department of Cardiology

West Virginia University

Morgantown, WV 26506

brijesh.patel@wvumedicine.org

4th Donald Adjeroh

Department of CSEE

West Virginia University

Morgantown, WV 26506

don@csee.wvu.edu

5th Jingxian Wu

Department of ELEG

University of Arkansas,

Fayetteville, Arkansas 72703

wuj@uark.edu

6th Morten Olgaard Jensen

Department of Biomedical Engineering

University of Arkansas,

Fayetteville, Arkansas 72703

mojensen@uark.edu

7th Ngan Le

Department of CSCE

University of Arkansas,

Fayetteville, Arkansas 72703

thile@uark.edu

Abstract—Electrocardiogram (ECG) signal is one of the most

effective sources of information mainly employed for the diagno-

sis and prediction of cardiovascular diseases (CVDs) connected

with the abnormalities in heart rhythm. Clearly, single modality

ECG (i.e. time series) cannot convey its complete characteristics,

thus, exploiting both time and time-frequency modalities in the

form of time-series data and spectrogram is needed. Leveraging

the cutting-edge self-supervised learning (SSL) technique on unla-

beled data, we propose SSL-based multimodality ECG classiﬁca-

tion. Our proposed network follows SSL learning paradigm and

consists of two modules corresponding to pre-stream task, and

down-stream task, respectively. In the SSL-pre-stream task, we

utilize self-knowledge distillation (KD) techniques with no labeled

data, on various transformations and in both time and frequency

domains. In the down-stream task, which is trained on labeled

data, we propose a gate fusion mechanism to fuse information

from multimodality.To evaluate the effectiveness of our approach,

ten-fold cross validation on the 12-lead PhysioNet 2020 dataset

has been conducted. https://github.com/UARK-AICV/ECG SSL

12Lead.

Index Terms—ECG classiﬁcation, self-supervised learning, con-

trastive learning, multimodalities, multi-lead

I. INTRODUCTION

CVDs are leading causes of deaths globally. The mortality

rate can be considerably reduced by early treatment if occult

signals linked with CVD are detected by ECG. The ECG

signals, which record cardiac electrical activities, are widely

adopted to diagnose abnormal heart rhythms and intra-cardiac

conduction abnormalities.

Traditionally, cardiac feature extraction [1] and pattern

classiﬁers [2] are separated. Notwithstanding the decent per-

formance, they are not applicable to real-life cases because

of high time consumption and complexity. Recent DNNs-

based approaches have obtained amazing research progress

in various domains [3]–[5]. DNN-based ECG classiﬁcation

in general can be categorized into single lead classiﬁcation

[6], [7] or multiple lead classiﬁcation [8], [9]. Our proposed

method belongs to the second category. In this group, [8]

proposed a simple residual neural network used for classifying

6 types of abnormalities on the in-house 12-lead database,

where some of its testing metrics are better than those of

expert cardiologists. [9] trained a CNNs-based structure on

multi-lead ECG data to diagnose myocardial infraction. Due

to high complexity in higher-dimensional data of 12-lead

ECG, different architectures have been adopted to model time

correlation among ECG sample points [10], [11]. Besides

time-series data, time-frequency has played an important role

in ECG analysis. [12] used STFT-based spectrogram and 2D

CNNs for ECG arrhythmia classiﬁcation. [7] proposed using

STFT and stationary wavelet transform (SWT) transformations

to obtain two-dimensional (2-D) matrix input suitable for deep

CNNs. [13] proposed a novel wavelet sequence based on deep

bidirectional LSTM network model.

Furthermore, with the growing demand for medical exam-

ination and treatment, the healthcare industry steadily accu-

mulates innumerable amounts of data but these unlabelled

data might not be serviceable for most tasks. To address this

limitation, we leverage the recent advanced SSL techniques,

i.e., contrastive learning [14]. A common workﬂow to apply

SSL is to train the network in an unsupervised manner by

learning with a pre-stream task, and then ﬁnetuning the pre-

trained network on a target downstream task. The suitable pre-

stream tasks can be considered in four categories: context-

based [15], generation-based [16], free semantic label-based

[17], [18], and cross-modal-based [19], [20]. Recently, SSL-

based DNNs have also been applied in ECG classiﬁcation. [21]

proposed a self-designed loss to bring the representations from

the same patient closer and study the shared context of individ-

ual recordings through time and scenarios. [22] customized an

SSL model that understands the differences among segments

from individual patients and dissimilarities between patients’

recordings from same category. [23] conducted extensive ex-

periments on four SSL methods with multiple combinations of

transformations and demonstrated the improvement in macro

arXiv:2210.06297v1 [eess.SP] 30 Sep 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：

MultimodalityMulti-LeadECGArrhythmiaClassicationusingSelf-SupervisedLearning1stThinhPhanDepartmentofAIConvergenceChonnamNationalUniversityGwangju,SouthKorea61186phantrandacthinh2382@gmail.com2ndDucLeDepartmentofCSCEUniversityofArkansas,Fayetteville,Arkansas72703minhducl@uark.edu3rdPatelBrijeshDepar...

展开>> 收起<<

Multimodality Multi-Lead ECG Arrhythmia Classiﬁcation using Self-Supervised Learning 1stThinh Phan.pdf

共4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！

Multimodality Multi-Lead ECG Arrhythmia Classiﬁcation using Self-Supervised Learning 1stThinh Phan

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: