基于脑电信号的运动功能障碍检测与评估研究

发布时间：2018-05-07 13:19

本文选题：运动功能障碍 + 脑电信号　；参考：《杭州电子科技大学》2017年硕士论文

【摘要】：运动功能障碍是多种脑损伤疾病的重要表现,如癫痫疾病、脑卒中等。对患者运动功能的检测与评估是进行疾病诊断和康复治疗的重要方法。同时,脑电信号(Electroencephalogram,EEG)是直接反映大脑活动情况的电生理表现形式。通过对EEG的分析,可以发现大量的生理、病理信息。本文希望通过对EEG的分析处理,对脑受损疾病患者的运动功能障碍进行评估。而对EEG的有效处理分析是一项复杂的工程。本文选取具有运动功能障碍表现的典型疾病—癫痫患者作为本课题研究的实验对象,主要通过利用EEG分析方法对患者的癫痫特征进行提取与识别。本文着重研究EEG分析处理算法,其中包括:EEG的消噪方法、特征提取方法以及模式分类识别算法。主要研究内容和创新点包括以下几个方面:(1)提出一种基于降噪源分离的EEG消噪方法。降噪源分离是盲源分离中一种新方法,可以根据待处理信号的特点具有针对性地设计合适的降噪函数,进行源分离处理而获得源信号。本文首先设计仿真实验,选择适合癫痫患者EEG的降噪函数,再根据该降噪函数进行源分离处理和消噪处理。通过仿真实验与实际EEG处理结果表明,降噪源分离的EEG消噪算法的消噪效果优于基于独立成分分析的盲源分离算法的消噪效果。(2)提出一种基于多种熵融合的EEG特征提取方法。由于单一的一种熵值只能表达信号某一方面的复杂度,往往缺少对信号的整体度量。为了对EEG从多个角度实现更加全面的信息表达,提出一种基于多种不同物理意义的熵融合的EEG特征提取方法,提高了EEG特征的完整表达能力。这一创新性方法已经在Neural ComputingApplications(SCI源刊)发表。研究工作首先分析四种不同测量方法的熵在EEG中的物理意义及表现,然后通过实验分析多种熵融合对EEG分类效果的提升。(3)提出一种基于极限学习机的集成分类器设计。极限学习机是一种较为新型的分类器,是近年来EEG分类器设计的研究热点之一。本文针对常规极限学习机泛化能力差、分类结果不稳定的缺陷,利用集成分类的思想,根据Bagging和Adaboost两种集成方法,将多个极限学习机集成为一个强分类器。实验分析证实,本文所设计的分类器在取得较好的分类结果的同时,也改善了极限学习机分类结果的稳定性。
[Abstract]:Motor dysfunction is an important manifestation of many brain injury diseases, such as epilepsy, stroke and so on. The detection and evaluation of patients' motor function is an important method for disease diagnosis and rehabilitation. At the same time, electroencephalogram (EGG) is a direct electrophysiological expression of brain activity. Through the analysis of EEG, we can find a lot of physiological and pathological information. The purpose of this study is to evaluate motor dysfunction in patients with brain damage by analyzing and treating EEG. The effective processing and analysis of EEG is a complex project. In this paper, a typical disease with motor dysfunction, epileptic patients, was selected as the experimental object of this study, mainly by using EEG analysis method to extract and identify the epileptic characteristics of the patients. This paper focuses on the analysis and processing algorithm of EEG, including the denoising method of EEG, the feature extraction method and the pattern classification recognition algorithm. The main research contents and innovations include the following aspects: 1) A EEG denoising method based on de-noising source separation is proposed. Noise reduction source separation is a new method in blind source separation. According to the characteristics of the signal to be processed, a suitable denoising function can be designed and the source signal can be obtained by source separation processing. In this paper, a simulation experiment is designed to select the noise reduction function suitable for epileptic patients with EEG, and then the source separation and denoising processing are carried out according to the noise reduction function. The results of simulation and EEG processing show that the de-noising effect of EEG denoising algorithm based on de-noising source separation is better than that of blind source separation algorithm based on independent component analysis. (2) A EEG feature extraction method based on multi-entropy fusion is proposed. Because a single entropy value can only express the complexity of a certain aspect of the signal, it often lacks the global measurement of the signal. In order to realize more comprehensive information expression of EEG from many aspects, a EEG feature extraction method based on entropy fusion of different physical meanings is proposed, which improves the complete expression ability of EEG features. This innovative approach has been published in Neural ComputingApplications(SCI Source. In this paper, the physical meaning and performance of entropy of four different measurement methods in EEG are analyzed firstly, and then a design of integrated classifier based on limit learning machine is proposed. Extreme learning machine (LLM) is a new type of classifier, which is one of the hotspots in the design of EEG classifier in recent years. Aiming at the shortcomings of poor generalization ability and unstable classification results of conventional LLMs, this paper integrates multiple LLMs into a strong classifier by using the idea of integrated classification and according to the two integration methods of Bagging and Adaboost. The experimental results show that the classifier designed in this paper not only achieves good classification results, but also improves the stability of the classification results of the extreme learning machine.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R318;TN911.7

【参考文献】