脑电信号的多尺度特性分析

发布时间：2019-04-03 16:57

【摘要】：脑电信号具体的产生机制目前仍处于研究阶段，它包含了大量神经元以不同形式组织活动的信息，具有非线性、多尺度、多分辨性等特性。多尺度性是指人体神经电生理信号包含了多种时间尺度成份及多种空间解剖结构尺度。其多尺度性是传统的单尺度熵方法难以描述的，也不是傅里叶线性多频率叠加理论可以完全解决的。其尺度的数目从理论上来说是无限多的，这是脑电信号这种极端复杂的非线性系统所共有的特征。因此，多尺度特性是急待研究的重点问题。近年来，多尺度熵方法已经逐渐成为探究多尺度特性及描述神经生理机制的工具。虽然有多种多尺度分解方法，如粗粒化过程，移动平均过程，小波变换和集合经验模态分解，但是很少有对这些方法的系统性的评估。本文的目的是找到描述神经生理机制的最佳多尺度熵指标。研究思路是综合比较多种多尺度熵的性能，进而应用到实际信号中分析其多尺度特性。首先，将四种熵方法，香农熵、样本熵、排序熵和递归熵与四种多尺度分解方法,粗粒化过程，移动平均过程，最大重叠率离散小波变换和集合经验模态分解相结合，生成16种多尺度熵方法。然后，应用一个基于不同参数的神经群模型来产生一个强度可变的神经元群，输出类似于正常脑电信号和癫痫棘波信号的模拟信号。添加不同密度的高斯白噪声到神经群模型中来量化每一个多尺度熵的抗噪性；使用预测概率分析来评估每一种多尺度熵的有效性；分别绘制不同信号状态下的各个尺度下的多尺度熵的值，量化每一种多尺度熵在各个尺度下的区分度。最后，将多尺度熵方法应用到实际的癫痫信号和麻醉信号中，，找到跟踪癫痫信号的癫痫状态强度和麻醉信号的麻醉深度的最佳多尺度熵方法。
[Abstract]:The specific generation mechanism of EEG signal is still in the research stage. It contains a large number of information of neurons in different forms of tissue activity, which has the characteristics of nonlinear, multi-scale, multi-resolution and so on. Multi-scale means that the electrophysiological signal of human nerve contains a variety of time-scale components and spatial anatomical structure scales. Its multi-scale property is difficult to describe by traditional single-scale entropy method, nor can it be completely solved by Fourier linear multi-frequency superposition theory. The number of scales is infinite in theory, which is the common characteristic of EEG signal, which is a very complex nonlinear system. Therefore, multi-scale characteristics is an urgent issue to be studied. In recent years, multi-scale entropy method has gradually become a tool for exploring multi-scale characteristics and describing neurophysiological mechanism. Although there are many multi-scale decomposition methods, such as coarse-grained process, moving average process, wavelet transform and set empirical mode decomposition, few of these methods are systematically evaluated. The aim of this paper is to find the best multi-scale entropy index to describe the neurophysiological mechanism. The research idea is to synthesize and compare the performance of multi-scale entropy, and then apply it to analyze the multi-scale characteristics of real signals. Firstly, four entropy methods, Shannon entropy, sample entropy, ordering entropy and recursive entropy, are combined with four multi-scale decomposition methods, coarse-grained process, moving average process, maximum overlap discrete wavelet transform and set empirical mode decomposition. Sixteen kinds of multi-scale entropy methods are generated. Then, a neural group model based on different parameters is used to generate a group of neurons with variable intensity, which outputs analog signals similar to normal EEG signals and epileptic spike signals. Different density Gaussian white noise is added to the neural group model to quantify the anti-noise property of each multi-scale entropy, and the prediction probability analysis is used to evaluate the effectiveness of each multi-scale entropy. The value of multi-scale entropy in each scale of different signal states is plotted, and the discrimination degree of each multi-scale entropy in each scale is quantified. Finally, the multi-scale entropy method is applied to the actual epileptic signals and anesthetic signals, and the optimal multi-scale entropy method is found to track the state-of-epilepsy intensity of epileptic signals and the anesthetic depth of anesthetic signals.
【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】：TN911.6

【参考文献】