基于改进多元多尺度熵的癫痫脑电信号自动分类

发布时间：2018-01-21 01:18

本文关键词： 癫痫脑电改进多元多尺度熵小波包分解分类　出处：《燕山大学》2014年硕士论文　论文类型：学位论文

【摘要】：癫痫是一种常见的神经系统疾病，约80%患者在发病时伴有癫痫样放电，这种放电也是目前癫痫诊断的主要依据。为正确诊断癫痫病症，往往需要对病患进行长时间的脑电监测。脑电信号数据量庞大，导致对脑电信号的分析判别成为一项繁重而又低效的工作。癫痫检测结果容易受到医生主观因素的影响，因此对癫痫脑电信号的自动分类识别就显得尤为重要。在研究国内外癫痫脑电信号常用分析方法后，重点介绍了改进多元多尺度熵算法的发展过程。多元多尺度熵作为多尺度熵在多元信号上的推广，是非线性动态相关性的一种反映。但是传统的多元多尺度熵计算量大，，对于通道数较多的系统需要耗费大量的时间和空间，并且无法准确的反应变量间的相关性。本文提出的改进多元多尺度熵，将传统多元多尺度熵针对单个变量的嵌入模式改为对所有变量同时嵌入，不但解决了通道数增加内存溢出问题，也更适用于实际多变量信号分析。本文中应用改进多元多尺度熵与小波包分解方法对癫痫脑电信号进行分类。癫痫脑电信号分类算法大致可分为时域、频域、时频和非线性域。其中时频分析中的小波包变换不仅能够反映信号的频率特性又能很好表征信号的局部信息，但是利用小波包特征进行分类需要耗费大量的时间和空间。本文提出的改进多元多尺度熵不仅保有原来多元多尺度熵对多通道数据并行处理、多尺度分析等特点，还大大降低了原有方法的复杂度与计算冗余。通过对癫痫脑电信号进行多次尺度分解，将改进多元多尺度熵与小波包变换结合，对其进行统计分析与分类。该方法既避免了由于特征数据量大而引发的大量时空消耗，又避免了传统时频分析中高频信号的干扰，更有利于实际应用。对于改进多元多尺度熵，针对GAERS大鼠癫痫脑电和波恩癫痫脑电数据进行实验，结果表明该方法能够有效提取癫痫脑电特征，具有很好的统计特性和分类精度。
[Abstract]:Epilepsy is a common nervous system disease, about 80% patients in the onset accompanied by epileptoid discharge, this discharge is also the main basis for the diagnosis of epilepsy, for the correct diagnosis of epilepsy. Patients often need to be monitored for a long period of time, EEG data volume is huge. As a result, the analysis and discrimination of EEG signal becomes a heavy and inefficient work. The results of epilepsy detection are easily affected by the subjective factors of doctors. Therefore, the automatic classification and recognition of epileptic EEG signals is particularly important. After studying the common analysis methods of epileptic EEG signals at home and abroad, the development process of improved multiscale entropy algorithm is introduced emphatically, which is the extension of multiscale entropy in multivariate signals. It is a reflection of nonlinear dynamic correlation, but the traditional multivariate multi-scale entropy calculation is large, and it needs a lot of time and space for the system with more channels. And the correlation between variables can not be accurately reflected. This paper proposes an improved multivariate multi-scale entropy, the traditional multi-scale entropy for a single variable embedding model instead of all variables at the same time. It not only solves the problem of increasing memory overflow, but also is more suitable for multivariable signal analysis. In this paper, improved multi-scale entropy and wavelet packet decomposition are used to classify epileptic EEG signals, which can be divided into time domain and frequency domain. The wavelet packet transform in time-frequency analysis can not only reflect the frequency characteristics of the signal but also represent the local information of the signal. However, using wavelet packet features to classify requires a lot of time and space. The improved multivariate multi-scale entropy proposed in this paper not only retains the original multi-scale entropy to process multi-channel data in parallel. Multi-scale analysis also greatly reduces the complexity and computational redundancy of the original method. The improved multi-scale entropy and wavelet packet transform are combined by multi-scale decomposition of epileptic EEG signals. The method not only avoids a large amount of space-time consumption caused by the large amount of characteristic data, but also avoids the interference of high-frequency signals in traditional time-frequency analysis. For improving multivariate multi-scale entropy, the GAERS rat epileptic EEG and Bonn epileptic EEG data were tested. The results show that this method can effectively extract epileptic EEG features. It has good statistical characteristics and classification accuracy.
【学位授予单位】：燕山大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：R742.1

【参考文献】