基于因果卷积的儿童睡眠分期方法  

A causal convolution-based sleep staging method for children

作　　者：赵德春[1] 李玲 舒洋 陈欢 侯筱蓉 ZHAO Dechun;LI Ling;SHU Yang;CHEN Huan;HOU Xiaorong(School of Bioinformatics,Chongqing University of Posts and Telecommunications,Chongqing 400065;School of Automation,Chongqing University of Posts and Telecommunications,Chongqing 400065;School of Medical Informatics,Chongqing Medical University,Chongqing 400016)

机构地区：[1]重庆邮电大学生物信息学院,重庆400065 [2]重庆邮电大学自动化学院,重庆400065 [3]重庆医科大学医学信息学院,重庆400016

出　　处：《北京生物医学工程》2023年第6期604-611,共8页Beijing Biomedical Engineering

基　　金：重庆市研究生科研创新项目(CYS22460J)资助。

摘　　要：目的睡眠分期能为儿童睡眠问题的诊断提供客观的评价标准,有利于提前发现和诊断儿童睡眠疾病。方法研究了一种基于因果卷积的儿童自动睡眠分期方法。首先,对原始的脑电信号进行带通滤波处理以减少噪声干扰,再利用具有不同尺寸卷积核的双分支模块提取信号的时频特征;然后利用膨胀因果卷积模块完成时序特征的提取;最后,通过全连接层和Softmax分类器对学习到的抽象特征进行分类。在20折交叉验证下,采用准确率、召回率、以及F1分数和科恩系数等指标对模型的分类性能进行评价。结果采用美国国家儿童医院的164名2~10岁临床儿童数据集,并设计了2~10岁、2~6岁、6~10岁3个年龄组,实验结果表明所提的因果卷积模型的儿童睡眠分期准确率分别为81.7%、80.0%、82.4%,科恩系数分别为0.75、0.73、0.76。结论基于因果卷积的睡眠分期方法对儿童数据有良好的分类能力,同时具有较快的收敛速度,可作为儿童睡眠疾病诊断的有效辅助工具。Objective Sleep staging can provide objective evaluation criteria for the diagnosis of children’s sleep problems and facilitate early detection and diagnosis of pediatric sleep disorders.Methods An automatic sleep staging method for children based on causal convolution is investigated.First,the raw EEG signal is band⁃pass filtered to reduce noise interference,and then the time⁃frequency features of the signal are extracted by using a two⁃branch module with convolution kernels of different sizes;then,the extraction of temporal features is completed by using the dilated causal convolution module;finally,the abstractfeatures learned by the network are classified through fully-connected layers and Softmax classifier.Under 20 fold cross validation,accuracy,recall,F1-score,and Cohen’s Kappa coefficient are used to evaluate the classification performance of the model.Results Child sleep data are obtained from 164 clinical child subjects aged 2-10 years at National Children’s Hospital.And three age groups are designed for children’s sleep data:2-10 years old,2-6 years old,and 6-10 years old.The experimental results show that the accuracy of the proposed causal convolution model for children’s sleep staging are 81.7%,80.0%,and 82.4%,with Cohen’s coefficients of 0.75,0.73,and 0.76,respectively.Conclusions The causal convolution-based sleep staging method has good classification ability for children’s data and also has a fast convergence speed,which can be used as an effective aid for children’s sleep physicians’diagnosis.

关 键 词：儿童睡眠 睡眠分期 因果卷积 时间序列 脑电 

分 类 号：R318.04[医药卫生—生物医学工程] R741.044[医药卫生—基础医学]