基于MSE-PCA的脑电睡眠分期方法研究

Research on sleep staging method of EEG based on MSE-PCA

针对传统的自动睡眠分期准确率不足问题,提出一种将多尺度熵(MSE)和主成分分析(PCA)联合使用的自动睡眠分期方法。以8例受试者睡眠脑电(EEG)监测数据及专家人工分期结果作为样本,首先使用MSE表征受试者脑电信号不同睡眠期的非线性动力学特征;然后使用PCA的前两个主成分向量代替MSE特征进行降维,实现降低数据冗余的同时保留绝大多数EEG非线性特征;最终将新向量的特征参数输入到反馈神经网络(BPNN)分类器中实现MSE-PCA模型的脑电睡眠状态的自动识别分类。实验结果表明,自动分期准确率可达到87.9%,kappa系数0.77,该方法能提高脑电自动睡眠分期系统的准确率和稳定性。

Aiming at the problem of insufficient accuracy of traditional automatic sleep staging, a new method of automatic sleep staging based on a fusion algorithm, multi-scale entropy( MSE) and principal component analysis( PCA), is proposed. In this work,the data of sleep EEG monitoring and the expert staging of 8 subjects are utilized as samples. Firstly, MSE is used to extract the nonlinear dynamic features from sleep stages. Then this features are replaced by the first two principal component vectors of PCA.The purpose is reduce the data dimension redundancy, as well as retaining the vast majority of EEG non-linear features. After that the new vector are entered into the BPNN classifier to implement the MSE-PCA model of automatic sleep staging. The experimental results show that the accuracy of automatic staging can reach to 87. 9 % and kappa coefficient is 0. 77, which can improve the accuracy and stability of automatic EEG sleep staging system.