基于卷積長短時記憶網(wǎng)絡(luò)的心律失常分類方法

柯麗; 王丹妮; 杜強(qiáng); 姜楚迪

doi:10.11999/JEIT190712

基于卷積長短時記憶網(wǎng)絡(luò)的心律失常分類方法

doi: 10.11999/JEIT190712 cstr: 32379.14.JEIT190712

沈陽工業(yè)大學(xué)電氣工程學(xué)院沈陽 110870

基金項(xiàng)目: 國家自然科學(xué)基金(51377109)，遼寧省自然科學(xué)基金(2019-ZD-0204)

詳細(xì)信息

作者簡介:
柯麗：女，1977年生，博士，教授，博士生導(dǎo)師，研究方向?yàn)樯镫姽づc阻抗成像技術(shù)

王丹妮：女，1995年生，碩士生，研究方向?yàn)獒t(yī)學(xué)信號處理與分析

杜強(qiáng)：男，1975年生，博士，講師，研究方向?yàn)樯镝t(yī)學(xué)信號檢測與處理

姜楚迪：女，1996年生，碩士生，研究方向?yàn)獒t(yī)學(xué)電磁工程及醫(yī)療儀器

通訊作者:
柯麗　keli@sut.edu.cn

中圖分類號: TP391; R540.41
計(jì)量
- 文章訪問數(shù): 5343
- HTML全文瀏覽量: 2955
- PDF下載量: 181
- 被引次數(shù): 0
出版歷程
- 收稿日期: 2019-09-16
- 修回日期: 2020-02-20
- 網(wǎng)絡(luò)出版日期: 2020-03-23
- 刊出日期: 2020-08-18

Arrhythmia Classification Based on Convolutional Long Short Term Memory Network

School of Electrical Engineering, Shenyang University of Technology, Shenyang 110870, China

Funds: The National Natural Science Foundation of China (51377109), The Natural Science Foundation of Liaoning Province (2019-ZD-0204)

摘要

摘要:
心律失常等慢性心血管疾病嚴(yán)重影響人類健康，采用心電信號(ECG)實(shí)現(xiàn)心律失常自動分類可有效提高該類疾病的診斷效率，降低人工成本。為此，該文基于1維心電信號，提出一種改進(jìn)的長短時記憶網(wǎng)絡(luò)(LSTM)方法實(shí)現(xiàn)心律失常自動分類。該方法首先設(shè)計(jì)深層卷積神經(jīng)網(wǎng)絡(luò)(CNN)對心電信號進(jìn)行深度編碼，提取心電信號形態(tài)特征。其次，搭建長短時記憶分類網(wǎng)絡(luò)實(shí)現(xiàn)基于心電信號特征的心律失常自動分類?；贛IT-BIH心律失常數(shù)據(jù)庫進(jìn)行的實(shí)驗(yàn)結(jié)果表明，該方法顯著縮短分類時間，并獲得超過99.2%的分類準(zhǔn)確率，靈敏度等評價參數(shù)均得到不同程度的提高，滿足心電信號自動分類實(shí)時高效的要求。
- 心電信號 /
- 心律失常 /
- 深度學(xué)習(xí) /
- 卷積神經(jīng)網(wǎng)絡(luò) /
- 長短時記憶網(wǎng)絡(luò)
Abstract:
Chronic cardiovascular diseases such as arrhythmia seriously affect human health. The automatic classification of ElectroCardioGram(ECG) signals can effectively improve the diagnostic efficiency of such diseases and reduce labor costs. To tackle this problem, an improved Long-Short Term Memory (LSTM) method is proposed to achieve automatic classification of one dimensional ECG signals. Firstly, deep Convolutional Neural Network (CNN) is designed to deeply encode the ECG signal, and ECG signal morphological features are extracted. Secondly, the LSTM classification network is used to realize automatic classification of arrhythmia of ECG signal features. Experimental studies based on the MIT-BIH arrhythmia database show that the training duration is significantly shortened and more than 99.2% classification accuracy is obtained. Sensitivity and other evaluation parameters are improved to meet the real-time and efficient requirements for automatic classification of ECG signals.
- ElectroCardioGram (ECG) signal /
- Arrhythmia /
- Deep learning /
- Convolutional Neural Network (CNN) /
- Long Short-Term Memory (LSTM)

HTML全文

圖 1 心電信號預(yù)處理

下載: 全尺寸圖片幻燈片

圖 2 C-LSTM網(wǎng)絡(luò)結(jié)構(gòu)

下載: 全尺寸圖片幻燈片

圖 3 各類別心電信號分段結(jié)果

下載: 全尺寸圖片幻燈片

圖 4 CNN提取到的信號特征

下載: 全尺寸圖片幻燈片

圖 5 網(wǎng)絡(luò)訓(xùn)練和驗(yàn)證性能圖

下載: 全尺寸圖片幻燈片

圖 6 網(wǎng)絡(luò)測試集混淆矩陣

下載: 全尺寸圖片幻燈片

表 1 CNN模型的細(xì)節(jié)和參數(shù)

層數(shù)	層名稱	卷積核大小	卷積核個數(shù)	激活函數(shù)	步長	參數(shù)	輸出大小
0	輸入	–	–	–	–		300×1
1	1維卷積	5×1	16	ReLU	1	96	300×16
2	批歸一化	–	–	–	–	128	300×16
3	1維卷積	5×1	16	ReLU	1	1424	300×16
4	批歸一化	–	–	–	–	1456	300×16
5	最大池化	2	16	–	2	32	150×16
6	1維卷積	3×1	32	ReLU	1	3024	150×32
7	批歸一化	–	–	–	–	3088	150×32
8	1維卷積	3×1	32	ReLU	1	6192	150×32
9	批歸一化	–	–	–	–	6256	150×32
10	最大池化	2	32	–	2	64	75×32
11	1維卷積	5×1	64	ReLU	1	16560	75×64
12	批歸一化	–	–	–	–	16688	75×64
13	1維卷積	5×1	1	ReLU	1	28800	75×1
14	批歸一化	–	–	–	–	28928	75×1
15	最大池化	2	1	–	2	2	38×1

下載: 導(dǎo)出CSV

表 2 LSTM模型的細(xì)節(jié)和參數(shù)

層名稱	隱含單元	激活函數(shù)	參數(shù)
長短時記憶層	32	–	12
全連接	256	ReLU	9996
全連接	5	Softmax	11024

下載: 導(dǎo)出CSV

表 3 AAMI標(biāo)準(zhǔn)在心電信號分類中描述

AAMI類別	類別數(shù)量	MIT-BIH心跳節(jié)拍類別
Normal(N)	89972	正常(NOR)
		左束支傳導(dǎo)阻塞(LBBB)
		右束支傳導(dǎo)阻塞(RBBB)
		房性逸搏(AE)
		結(jié)性逸搏(NE)
Supraventricular(S)	2758	房性早搏(AP)
		異常房性早搏(aAP)
		交界性早搏(NP)
		室上性早搏(SP)
Ventricular(V)	7140	室性早搏(PVC)
Ventricular(V)	7140	室性逸搏(VE)
Fusion(F)	800	心室融合心跳(fVN)
Unknown(Q)	30	起搏心跳(P)
		起搏融合心跳(fPN)
		未分類心跳(U)

下載: 導(dǎo)出CSV

表 4 LSTM網(wǎng)絡(luò)和C-LSTM網(wǎng)絡(luò)測試集的相關(guān)評價參數(shù)(%)

網(wǎng)絡(luò)	評價參數(shù)	模型類別
網(wǎng)絡(luò)	評價參數(shù)	N	S	V	F	Q
LSTM	Acc	99.54	99.62	99.44	99.71	99.97
	Sen	99.87	91.06	95.66	80.19	0
	Spe	96.91	99.86	99.76	99.58	99.99
	PPV	99.61	95.09	97.00	78.70	0.00
C-LSTM	Acc	99.52	99.61	99.51	99.84	99.97
	Sen	99.78	92.11	96.63	88.52	0
	Spe	98.36	99.83	99.73	99.93	99.99
	PPV	99.68	94.08	96.45	91.53	0.00

下載: 導(dǎo)出CSV

表 5 自動檢測心律失常分類結(jié)果性能比較

研究	類型	分類器	信號長度	性能(%)
研究	類型	分類器	信號長度	Acc	Sen	Spe	PPV
文獻(xiàn)[19]	4	FFNN	250 samples (0.69 s)	96.94	97.78	96.31	–
文獻(xiàn)[17]	17	KNN	360 samples (1.00 s)	97.00	97.10	96.90	–
文獻(xiàn)[18]	5	SVM+RBF	200 samples (0.56 s)	98.91	98.91	97.85	–
文獻(xiàn)[4]	14	NPE+SVM	300 samples (0.83 s)	98.51	98.51	–	98.51
文獻(xiàn)[11]	5	CNN	360 samples (1.00 s)	94.03	96.71	91.54	97.86
文獻(xiàn)[20]	4	SVM	8×1071	98.39	96.86	98.92	96.85
文獻(xiàn)[12]	5	CNN	73×73	98.42	72.06	97.83	65.91
文獻(xiàn)[15]	5	FCMDBN	200 samples (0.56 s)	96.54	94.55	93.31	93.91
文獻(xiàn)[14]	2	CNN+RNN	211×24	95.76	87.85	87.85	94.99
本文方法	5	LSTM	300 sample (0.83 s)	99.14	91.70	99.22	92.60
本文方法	5	C-LSTM	300 →38 samples(0.83 s)→(0.12 s)	99.23	94.26	99.57	95.44

下載: 導(dǎo)出CSV

參考文獻(xiàn)(20)

World Health Organization. Cardiovascular diseases[EB/OL]. https://www.who.int/health-topics/cardiovascular-diseases/#tab=tab_1, 2017.

YE Can, KUMAR B V K V, and COIMBRA M T. Heartbeat classification using morphological and dynamic features of ECG signals[J]. IEEE Transactions on Biomedical Engineering, 2012, 59(10): 2930–2941. doi: 10.1109/tbme.2012.2213253

YILDIRIM ?. ECG beat detection and classification system using wavelet transform and online sequential ELM[J]. Journal of Mechanics in Medicine and Biology, 2019, 19(1): 1940008. doi: 10.1142/S0219519419400086

高興姣, 李智, 陳珊珊, 等. 基于近鄰保持嵌入算法的心律失常心拍分類[J]. 生物醫(yī)學(xué)工程學(xué)雜志, 2017, 34(1): 1–6. doi: 10.7507/1001-5515.201605045

GAO Xingjiao, LI Zhi, CHEN Shanshan, et al. Arrhythmia heartbeats classification based on neighborhood preserving embedding algorithm[J]. Journal of Biomedical Engineering, 2017, 34(1): 1–6. doi: 10.7507/1001-5515.201605045

AHMED R and ARAFAT S. Cardiac arrhythmia classification using hierarchical classification model[C]. The 6th International Conference on Computer Science and Information Technology (CSIT), Amman, Jordan, 2014: 203–207. doi: 10.1109/CSIT.2014.6806001.

BALOUCHESTANI M, SUGAVANESWARAN L, and KRISHNAN S. Advanced K-means clustering algorithm for large ECG data sets based on K-SVD approach[C]. The 9th International Symposium on Communication Systems, Networks & Digital Sign (CSNDSP), Manchester, UK, 2014: 177–182. doi: 10.1109/CSNDSP.2014.6923820.

LI Duan, ZHANG Hongxin, and ZHANG Mingming. Wavelet de-noising and genetic algorithm-based least squares twin SVM for classification of arrhythmias[J]. Circuits, Systems, and Signal Processing, 2017, 36(7): 2828–2846. doi: 10.1007/s00034-016-0439-8

YIN Xi and LIU Xiaoming. Multi-task convolutional neural network for pose-invariant face recognition[J]. IEEE Transactions on Image Processing, 2018, 27(2): 964–975. doi: 10.1109/TIP.2017.2765830

王斐, 吳仕超, 劉少林, 等. 基于腦電信號深度遷移學(xué)習(xí)的駕駛疲勞檢測[J]. 電子與信息學(xué)報(bào), 2019, 41(9): 2264–2272. doi: 10.11999/JEIT180900

WANG Fei, WU Shichao, LIU Shaolin, et al. Driver fatigue detection through deep transfer learning in an electroencephalogram-based system[J]. Journal of Electronics &Information Technology, 2019, 41(9): 2264–2272. doi: 10.11999/JEIT180900

MENG Huanhuan and ZHANG Yue. Classification of electrocardiogram signals with deep belief networks[C]. The 17th International Conference on Computational Science and Engineering, Chengdu, China, 2014: 7–12. doi: 10.1109/CSE.2014.36.

ACHARYA U R, OH S L, HAGIWARA Y, et al. A deep convolutional neural network model to classify heartbeats[J]. Computers in Biology and Medicine, 2017, 89: 389–396. doi: 10.1016/j.compbiomed.2017.08.022

ZHAI Xiaolong and TIN C. Automated ECG classification using dual heartbeat coupling based on convolutional neural network[J]. IEEE Access, 2018, 6: 27465–27472. doi: 10.1109/ACCESS.2018.2833841

CHENG Maowei, SORI W J, JIANG Feng, et al. Recurrent neural network based classification of ecg signal features for obstruction of sleep apnea detection[C]. 2017 IEEE International Conference on Computational Science and Engineering (CSE) and IEEE International Conference on Embedded and Ubiquitous Computing (EUC), Guangzhou, China, 2017, 199–202. doi: 10.1109/CSE-EUC.2017.220.

TAN J H, HAGIWARA Y, PANG W, et al. Application of stacked convolutional and long short-term memory network for accurate identification of CAD ECG signals[J]. Computers in Biology and Medicine, 2018, 94: 19–26. doi: 10.1016/j.compbiomed.2017.12.023

吳志勇, 丁香乾, 許曉偉, 等. 基于深度學(xué)習(xí)和模糊C均值的心電信號分類方法[J]. 自動化學(xué)報(bào), 2018, 44(10): 1913–1920. doi: 10.16383/j.aas.2018.c170417

WU Zhiyong, DING Xiangqian, XU Xiaowei, et al. A method for ECG classification using deep learning and fuzzy C-means[J]. Acta Automatica Sinica, 2018, 44(10): 1913–1920. doi: 10.16383/j.aas.2018.c170417

HANNUN A Y, RAJPURKAR P, HAGHPANAHI M, et al. Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network[J]. Nature Medicine, 2019, 25(1): 65–69. doi: 10.1038/s41591-018-0268-3

PARK J, LEE K, and KANG K. Arrhythmia detection from heartbeat using k-nearest neighbor classifier[C]. 2013 IEEE International Conference on Bioinformatics and Biomedicine, Shanghai, China, 2013: 15–22. doi: 10.1109/BIBM.2013.6732594.

ELHAJ F A, SALIM N, HARRIS A R, et al. Arrhythmia recognition and classification using combined linear and nonlinear features of ECG signals[J]. Computer Methods and Programs in Biomedicine, 2016, 127: 52–63. doi: 10.1016/j.cmpb.2015.12.024

GüLER I and üBEYLI E D. ECG beat classifier designed by combined neural network model[J]. Pattern Recognition, 2005, 38(2): 199–208. doi: 10.1016/j.patcog.2004.06.009

SAHOO S, KANUNGO B, BEHERA S, et al. Multiresolution wavelet transform based feature extraction and ECG classification to detect cardiac abnormalities[J]. Measurement, 2017, 108: 55–66. doi: 10.1016/j.measurement.2017.05.022

相關(guān)文章

施引文獻(xiàn)

資源附件(0)

訪問統(tǒng)計(jì)