基于心电和脑电多模态特征组合的自动睡眠分期方法研究  

Automatic Sleep Staging Method based on Electroencephalogram and Electrocardiogram Multi-modal Features Combination

作　　者：吕君同 史文彬 张楚婷 李凡 郭睿琦 陈一川 周聪睿 叶建宏 LV Juntong;SHI Wenbin;ZHANG Chuting;LI Fan;GUO Ruiqi;CHEN Yichuan;ZHOUCongrui;YEH Chien-Hung(School of Information and Electronics,Beijing Institute of Technology,Beijing,100081;XUTELI School,Beijing Institute of Technology,Beijing,100081)

机构地区：[1]北京理工大学信息与电子学院,北京100081 [2]北京理工大学徐特立学院,北京100081

出　　处：《生命科学仪器》2023年第1期41-49,共9页Life Science Instruments

基　　金：国家自然科学基金项目(62171028、62001026);国家高层次人才基金项目(3050012222022)

摘　　要：准确的睡眠分期能够评估睡眠质量,在睡眠紊乱或疾病诊断干预中起关键作用。本文利用多尺度熵分析和经验模态分解方法获取多通道脑电信号于睡眠状态下的非线性动力学特征,利用心率变异度时频域指标及多尺度样本熵构建了心电信号睡眠特征。基于最大相关-最小冗余特征选择算法及主成分分析降维,实现了高效多模态特征组合构建。多模态特征组合驱动的多种传统机器学习自动睡眠分期模型在ISRUC-S3数据集上达到了最高84.05%准确率,Kappa系数最高为0.7810,表明所提出多模态特征组合的有效性及准确性。Accurate sleep staging supports evaluating the quality of sleep,which plays an essential role in the clinical diagnosis and intervention of sleep disorders and related diseases.This study applied the multiscale entropy analysis and empirical mode decomposition to extract the information of nonlinear dynamics in multi-channel electroencephalogram signals during sleep,and employed heart rate variability including time and frequency measures,as well as the sample entropy in different scales for electrocardiogram signals.By applying the maxrelevance and min-redundancy feature selection method and principle component analysis for dimensionality reduction,an efficient multi-modal features combination was constructed.The multi-modal features combination was then fed into multiple traditional machine-learning-based automatic sleep classification models,which achieved the highest accuracy of 84.05%and the highest kappa value of 0.7810 on the ISRUC-S3 dataset,indicating the high effectiveness and high precision of our multi-modal features combination in automatic sleep staging.

关 键 词：自动睡眠分期 多尺度熵分析 经验模态分解 心率变异度 多模态 

分 类 号：R540[医药卫生—心血管疾病]