基于单通道心电信号的家用睡眠分期方法  

Household sleep staging method based on single channel ECG signal

作　　者：缪竟鸿[1,2] 韩旭[1] Tasmia Avouka 胡猛 王慧泉[1,2] 赵晓赟[3] 韦然[1,2] MIAO Jinghong;HAN Xu;AVOUKA Tasmia;HU Meng;WANG Huiquan;ZHAO Xiaoyun;WEI Ran(School of Life Science,Tiangong University,Tianjin 300387,China;Tianjin Key Laboratory of Optoelectric Detection Technology and System,Tiangong University,Tianjin 300387,China;Department of Respiratory and Critical Care Medicine,Chest Disease Hospital of Tianjin City,Tianjin 300222,China)

机构地区：[1]天津工业大学生命科学学院,天津300387 [2]天津工业大学天津市光电检测技术与系统重点实验室,天津300387 [3]天津市胸科医院呼吸与危重症医学科,天津300222

出　　处：《天津工业大学学报》2023年第4期83-88,共6页Journal of Tiangong University

基　　金：天津市科技计划项目(18ZXRHSY00200);天津市教委科研计划项目(2019KJ024)。

摘　　要：为了实现居家环境下的睡眠健康监测,提出基于可穿戴式单通道心电的不同睡眠阶段识别分析方法,研究了心电信号的27个心率变异性(HRV)特征分别在序列前向选择(SFS)、序列后向选择(SBS)和浮动序列前向选择(SFFS)3种不同方式下最优特征集的选取,设计了以堆栈式自编码器(SAE)建立的睡眠分期神经网络模型系统。结果表明:基于SFS-SAE的睡眠分期模型方法的分类效果最好,且相邻RR间期序列差值大于50 ms百分比(pNN50)、相邻RR间期序列差值的绝对中位差(MADRR)和庞加莱散点图中椭圆短轴(SD1)3种HRV特征都能在此模型系统中有效用于各睡眠阶段的识别,在清醒-睡眠(WAKE-SLEEP)分类、非快速眼动-快速眼动(NREM-REM)分类和浅睡-深睡(N1N2-N3)分类下的平均准确率分别为82%、80%和81%,基本满足家用睡眠分期判别,可用于睡眠疾病的日常筛查,是对多导睡眠图睡眠分析方法的有效补充。In order to realize sleep health monitoring in the home environment,a method based on wearable single channel ECG is proposed to identify and analyze different sleep stages.27 heart rate variability(HRV)features of ECG signal are studied in three different ways:sequence forward selection(SFS),sequence backward selection(SBS)and floating sequence forward selection(SFFS),a sleep staging neural network model system based on stack self encoder(SAE)is designed.The results show that the sleep stage model based on SFS-SAE has the best classification effect,and the three HRV features of the difference between adjacent RR interval sequences is greater than 50 milli-second percentage(pNN50),the absolute median difference of the difference between adjacent RR interval sequences(MADRR)and the elliptical minor axis(SD1)in Poincare scatter diagram can be effectively used in the recognition of each sleep stage in this model system.The average accuracy of WAKE-SLEEP classification,NREM-REM classification and N1N2-N3 classification are 82%,80%and 81%respectively,which basically meet the requirements of household sleep staging and can be used for daily screening of sleep diseases and it is an effective supplement to polysomnography sleep analysis.

关 键 词：睡眠分期 心率变异性 特征选择 神经网络 单通道心电信号 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]