基于CNN和LSTM的睡眠呼吸暂停检测算法

doi:10.16180/j.cnki.issn1007-7820.2021.02.004

摘要/Abstract

摘要：

睡眠呼吸暂停是一种常见的睡眠障碍,与多种疾病发生有关。文中提出了使用深度学习方法来实现睡眠呼吸暂停事件检测。心电信号两个阶段的预处理分别作为CNN和LSTM模型的输入:CNN模型以原始ECG信号作为输入,通过卷积自动提取特征来识别睡眠呼吸暂停;LSTM模型使用心电的间接信号作为输入,从RR间期和呼吸信号中自动提取特征。实验表明,LSTM模型准确度较高为87.4%,与传统方法性能接近。文中所提方法结合了人工特征提取和深度学习优点,比传统分类方法更具适用性。

关键词: 卷积神经网络, 长期短期记忆, 睡眠呼吸暂停, 心电信号, 预处理, 特征提取

Abstract:

Sleep apnea is a common sleep disorder, which is associated with multiple diseases. This study proposes deep learning methods to detect sleep apnea events. The two stages of ECG signal preprocessing are used as inputs to the CNN and LSTM models, respectively. The CNN model takes the original ECG signal as input and automatically extracts features through convolution to identify sleep apnea. The LSTM model uses ECG's indirect signal as input and automatically extractes features from RR intervals and respiratory signals. Experiments show that the LSTM model has a high accuracy of 87.4%, which is close to the performance of the traditional methods. The proposed method combines the advantages of artificial feature extraction and deep learning, and is more applicable than the traditional classification methods.

Key words: CNN, LSTM, sleep apnea, electrocardiograph signal, preprocessing, feature extraction

中图分类号:

TP183

葛靖,刘子龙. 基于CNN和LSTM的睡眠呼吸暂停检测算法[J]. 电子科技, 2021, 34(2): 21-26.

GE Jing,LIU Zilong. The Algorithm Based on CNN and LSTM for Sleep Apnea Syndrome Detection[J]. Electronic Science and Technology, 2021, 34(2): 21-26.

图/表 7

图1

图2

表1

表2

图3

表3

表4

参考文献 19

[1]	余晓敏, 涂岳文, 黄超, 等. 基于心电信号的睡眠呼吸暂停综合征检测算法[J]. 生物医学工程学杂志, 2013,30(5):999-1002.
	Yu Xiaomin, Tu Yuewen, Huang Chao, et al. An algorithm based on ECG signal for sleep apnea syndrome detection[J]. Journal of Biomedical Engineering, 2013,30(5):999-1002.
[2]	Mendonça F, Mostafa S S, Ravelo-García A G, et al.A review of obstructive sleep apnea detection approaches[J]. IEEE Journal of Biomedical and Health Informatics, 2018,23(2):825-837. doi: 10.1109/JBHI.2018.2823265 pmid: 29993672
[3]	韦张跃昊, 钱升谊. 基于滤波重构和卷积神经网络的心电信号分类[J].电子科技, 2019(11):1-6.
	Wei Zhangyuehao, Qian Shengyi. ECG signal classification based on filtering-reconstruction and convolutional neural network[J].Electronic Science and Technology, 2019(11):1-6.
[4]	Penzel T, Moody G B, Mark R G, et al. The apnea-ECG database [C].Cambridge:Computers in Cardiology, 2000.
[5]	郭永丛, 司玉娟. 基于心电信号提取呼吸信号的算法[J]. 吉林大学学报(信息科学版), 2016,34(3):28-34.
	Guo Yongcong, Si Yujuan. Respiratory signal extraction algorithm based on ECG[J]. Journal of Jilin University(Information Science Edition), 2016,34(3):28-34.
[6]	林剑华. 基于心电信号的呼吸检测算法及其在睡眠分析中的应用研究[D]. 武汉:华中科技大学, 2017.
	Lin Jianhua. The study of ECG-based respiration detection algorithm and its application in sleep analysis[D]. Wuhan:Huazhong University of Science and Technology, 2017.
[7]	张异凡, 黄亦翔, 汪开正, 等. 用于心律失常识别的LSTM和CNN并行组合模型[J]. 哈尔滨工业大学学报, 2019,51(10):76-82.
	Zhang Yifan, Huang Yixiang, Wang Kaizheng, et al. Arrhythmia classification using parallel combination of LSTM and CNN[J]. Journal of Harbin Institute of Technology, 2019,51(10):76-82.
[8]	黄永锋, 江依鹏, 杨树臣. 基于卷积神经网络的非接触式呼吸暂停算法研究[J]. 智能计算机与应用, 2019,9(4):104-106,111.
	Huang Yongfeng, Jiang Yipeng, Yang Shuchen. Research on non-contact apnea algorithm based on convolutional neural network[J]. Intelligent Computer and Applications, 2019,9(4):104-106,111.
[9]	武悦. 基于心电信号对睡眠呼吸暂停综合征判别算法的研究[D]. 长春:吉林大学, 2018.
	Wu Yue. Research on discriminant algorithm of sleep apnea syndrome based on electrocardiographic signal[D]. Changchun:Jilin University, 2018.
[10]	Erdenebayar U, Kim Y J, Park J U, et al. Deep learning approaches for automatic detection of sleep apnea events from an electrocardiogram[J].Computer Methods and Programs in Biomedicine, 2019(18):105-115.
[11]	De Chazal P, Sadr N. Sleep apnoea classification using heart rate variability, ECG derived respiration and cardiopulmonary coupling parameters [C].Orlando: IEEE Engineering in Medicine and Biology Society, 2016.
[12]	Kiranyaz S, Ince T, Gabbouj M. Real-time patient-specific ECG classification by 1-D convolutional neural networks[J]. IEEE Transactions on Biomedical Engineering, 2015,63(3):664-675. doi: 10.1109/TBME.2015.2468589 pmid: 26285054
[13]	Banluesombatkul N, Rakthanmanon T, Wilaiprasitporn T. Single channel ecg for obstructive sleep apnea severity detection using a deep learning approach [C].Jeju Island:IEEE Region 10 Conference, 2018.
[14]	Choi S H, Yoon H, Kim H S, et al. Real-time apnea-hypopnea event detection during sleep by convolutional neural networks[J].Computers in Biology and Medicine, 2018(4):123-131.
[15]	Yu H, Guo X. The Detection of Sleep Apnea Hypopnea Syndrome based on Improved BP Neural Network [C]. Shenzhen:IEEE the Third International Conference on Signal and Image Processing, 2018.
[16]	Pathinarupothi R K, Rangan E S, Gopalakrishnan E A, et al. Single sensor techniques for sleep apnea diagnosis using deep learning [C].Park City:IEEE International Conference on Healthcare Informatics, 2017.
[17]	Song C, Liu K, Zhang X, et al. An obstructive sleep apnea detection approach using a discriminative hidden Markov model from ECG signals[J]. IEEE Transactions on Biomedical Engineering, 2015,63(7):1532-1542. doi: 10.1109/TBME.2015.2498199 pmid: 26560867
[18]	Martín-González S, Navarro-Mesa J L, Juliá-Serdá G, et al.Heart rate variability feature selection in the presence of sleep apnea: An expert system for the characterization and detection of the disorder[J]. Computers in Biology and Medicine, 2017(91):47-58.
[19]	Li K Y, Pan W F, Li Y F, et al. A method to detect sleep apnea based on deep neural network and hidden Markov model using single-lead ECG signal[J].Neurocomputing, 2018(3):294-303.

编号	层类型	滤波器尺寸	输出维度
1	batchnorm1	-	6 000×1
2	conv_1	64@50×1	5 951×64
3	maxpool_1	2×2	2 975×64
4	dropout1	0.2	2 975×64
5	batchnorm2	-	2 975×64
6	conv_2	64@50×1	2 926×64
7	conv_3	32@30×1	2 897×32
8	maxpool_2	2×2	1 448×32
9	dropout2	0.2	1 448×32
10	conv_4	16@10×1	1 439×32
11	maxpool_3	2×2	719×16
12	flatten	-	11 504
13	dense_1	-	256
14	dense_2	-	1

编号	层类型	记忆细胞	输出维度
1	batchnorm1	-	240×2
2	1^st LSTM	240	240×128
3	2^nd LSTM	240	256
4	batchnorm2	-	64
5	1^st dense_1	-	64
6	2^nd dense_2	-	64
7	merge	-	320
8	dense_3	-	32
9	dense_4	-	1

方法	Accuracy/%	Recall/%	Precision/%
CNN	75.3	78.1	80.7
LSTM	87.4	84.2	85.5

文献来源	方法	Accuracy /%	Recall /%	Precision /%
Wu等^[9]	SVM	91.29	89.80	91.75
Martín等 ^[18]	DFA	84.76	81.45	86.82
Son等^[17]	HMM+SVM	86.20	82.60	88.40
Li等^[19]	DNN+HMM	84.70	88.90	82.10