Manuscript_OyildirimR2ClearWord文档下载推荐.docx

1、 March 2018DOI: 10.1016/pbiomed.2018.03.016CITATIONS118READS5,079Some of the authors of this publication are also working on these related projects:Smart Grid View projectPower quality monitoring systems View projectAll content following this page was uploaded by zal yldrm on 22 July 2018.The user h

2、as requested enhancement of the downloaded file.A NOVEL WAVELET SEQUENCES BASED ON DEEP BIDIRECTIONAL LSTM NETWORK MODEL FOR ECG SIGNAL CLASSIFICATIONzal YILDIRIM（*corresponding author）1 Computer Engineering Department, Engineering Faculty, Munzur University, Tunceli, Turkey* Corresponding author. T

3、el.: +90-428 -2131794; fax: +90-428-2131861.E-mail address: oyildirimmunzur.edu.tr, yildirimozal29A Novel Wavelet Sequences Based on Deep Bidirectional LSTM Network Model for ECG Signal Classificationzal YILDIRIMComputer Engineering Department, Munzur University, Tunceli, Turkeyoyildirimmunzur.edu.t

4、r, yildirimozalAbstractLong-short term memory networks （LSTMs）, which have recently emerged in sequential data analysis, are the most widely used type of recurrent neural networks （RNNs） architecture. Progress on the topic of deep learning includes successful adaptations of deep versions of these ar

5、chitectures. In this study, a new model for deep bidirectional LSTM network-based wavelet sequences called DBLSTM-WS was proposed for classifying electrocardiogram （ECG） signals. For this purpose, a new wavelet-based layer is implemented to generate ECG signal sequences. The ECG signals were decompo

6、sed into frequency sub-bands at different scales in this layer. These sub-bands are used as sequences for the input of LSTM networks. New network models that include unidirectional （ULSTM） and bidirectional （BLSTM） structures are designed for performance comparisons. Experimental studies have been p

7、erformed for five different types of heartbeats obtained from the MIT-BIH arrhythmia database. These five types are Normal Sinus Rhythm （NSR）, Ventricular Premature Contraction （VPC）, Paced Beat （PB）, Left Bundle Branch Block （LBBB）, and Right Bundle Branch Block （RBBB）. The results show that the DB

8、LSTM-WS model gives a high recognition performance of 99.39%. It has been observed that the wavelet-based layer proposed in the study significantly improves the recognition performance of conventional networks. This proposed network structure is an important approach that can be applied to the simil

9、ar signal processing problems.Keywords: Long-Short Term Memory, Recurrent Neural Networks, Deep Learning, ECG signals.1. IntroductionImprovements in the field of machine learning have allowed for the development of computer- based intelligent decision support systems and their implementation more ef

10、ficiently in many areas. The development of intelligent systems in the health field is attractive in terms of the amount of data and the importance of the data it contains. Researchers in different disciplines, specifically in terms of applications, frequently study biomedical data, including image

11、and signal data. Electrocardiogram （ECG） signals are one of the most frequently studied data. ECG signals include information about heart function and heart conditions. Therefore, monitoring and recognition of ECG signals is a remarkable topic in the biomedical field 1.Studies performed on ECG signa

12、ls are generally examined in two parts: detection and classification. Studies on detection concentrate on the problem of determining heartbeat within the ECG data obtained for a certain period of time. These detected beats are marked in the rawdata for analysis. It is an area particularly emphasized

13、 for the development of wearable biomedical monitoring systems 2, 3. Threshold-based methods 4, digital filter-based methods 5, 6, and wavelet transform （WT） 7-10 have been applied in heartbeat detection scenarios.The classification of detected ECG signals is another important step. At this stage, a

14、utomatic identification of segmented heartbeat signals is provided. There are generally several steps for classification. Since direct classifier execution on segmented signals is not efficient in terms of performance, the distinguishing features of these signals are hand-crafted and used as input data. Wavelet transform is one of the commonly used methods to obtain ECG signal features 11-15. By using some statistical methods such as Principal Component Analysis （PCA） and Linear Discriminant Analysis （LDA） on these coefficient