最小生成树在精神分裂症EEG脑网络中的应用研究
发布时间：2018-09-10 10:52
【摘要】：随着社会压力的不断增加,精神类疾病已经成为导致亚健康的主要因素之一,精神分裂症就是最为常见的精神类疾病,对该疾病的早期诊断也显得尤为重要。目前的诊断手段大多基于量表,医生根据量表进行主观判断,这种诊断方式存在较多主观因素,为精神分裂症的诊断带来不可预估的影响。随着复杂网络理论和脑功能成像技术的发展,基于EEG脑网络的精神分裂症疾病研究取得了一定的进展,为该疾病的病理认知和早期诊断提供了新的思路。但是,近来也有研究指出传统复杂网络分析方法存在一定的弊端。传统复杂网络研究方法包括无权网络、加权网络。无权网络涉及到阈值选择的问题,阈值T的选择从本质上是随机的。阈值的选择可能会导致网络中存在虚假或噪声连接(T的选择较小),也可能丢弃掉网络中包含重要信息的一些弱连接(T的选择较大)。边的多少进一步影响脑网络属性值的测量。而加权网络的度量同样会被噪声连接和平均功能连接强度影响,所以,无权网络和加权网络的分析方法可能会由于方法学的问题导致无法令人确信的研究结果。因此,本研究将最小生成树算法(Minimum Spanning Tree,MST)引入到复杂脑网络的研究中,期望能解决传统网络分析方法中存在的问题,在EEG脑网络分析的方法学问题上进行有价值的探索。并在精神分裂症与正常被试的EEG脑网络中对无权网络、加权网络以及最小生成树这三种研究方法进行了对比分析与探讨。本研究所用的EEG数据为来自合作单位北京回龙观医院的40例精神分裂症患者和40例正常被试,所有数据经过预处理后,通过相位滞后指数(PLI)分别构建了无权、加权以及MST脑网络,然后将不同被试组间显著差异的网络属性作为分类特征进行了分类研究,最后从理论分析、相关性分析和分类效果上证明了MST引入EEG脑网络的研究是有效可行的,是对复杂脑网络传统分析方法的有益补充。本研究主要完成的内容有以下几点:(1)选取不同阈值构建各个稀疏度下精神分裂症与正常被试的无权脑网络,计算其属性并进行了统计分析,提取有显著性差异的无权网属性。(2)计算精神分裂症与正常被试的加权脑网络属性并对其进行了统计分析,提取有显著性差异的加权网属性。(3)采用Kruskal算法构建精神分裂症与正常被试的MST。计算精神分裂症与正常被试MST网络的属性并对其进行统计分析,提取有显著性差异的MST属性。(4)从理论分析、相关性分析和分类效果上证明了MST引入EEG脑网络的研究是有效可行的,是对复杂脑网络传统分析方法的有益补充。总之,MST比传统复杂网络分析方法在EEG脑网络研究中具有更好的表现,将其作为传统复杂网络的有益补充,能为EEG脑网络的研究提供更佳的思路与方法。
[Abstract]:With the increasing of social pressure, mental diseases have become one of the main causes of sub-health, schizophrenia is the most common mental diseases, the early diagnosis of this disease is particularly important. At present, most diagnostic methods are based on the scale, and doctors make subjective judgment according to the scale. There are many subjective factors in this diagnostic method, which has an unpredictable impact on the diagnosis of schizophrenia. With the development of complex network theory and brain function imaging technology, the research of schizophrenia disease based on EEG brain network has made some progress, which provides a new idea for the pathological cognition and early diagnosis of the disease. However, recent studies have pointed out that traditional complex network analysis methods have some drawbacks. Traditional complex network research methods include weighted network and weighted network. The selection of threshold T is random in nature. The selection of threshold may lead to the existence of false or noisy connections (the selection of T is small), or it may discard some weak connections containing important information in the network (the choice of T is large). The number of edges further affects the measurement of brain network attribute values. The measurement of weighted networks is also affected by the noise connection and the average functional connection strength. Therefore, the analysis methods of the weighted network and the weighted network may lead to the inconclusive research results due to the methodological problems. Therefore, the minimum spanning tree algorithm (Minimum Spanning Tree,MST) is introduced into the study of complex brain networks, which is expected to solve the problems existing in the traditional network analysis methods and to explore the methodology of EEG brain network analysis. In the EEG brain network of schizophrenia and normal subjects, the three research methods of weighted network, weighted network and minimal spanning tree were compared and discussed. The EEG data used in this study were 40 schizophrenic patients and 40 normal subjects from Huilongguan Hospital in Beijing. After pretreatment, the data were constructed with phase lag index (PLI). Weighted and MST brain networks were used to classify the significantly different network attributes of different subjects as classification features. Finally, the theoretical analysis was carried out. The results of correlation analysis and classification show that the research of introducing MST into EEG brain networks is effective and feasible, and it is a useful supplement to the traditional analysis methods of complex brain networks. The main contents of this study are as follows: (1) selecting different thresholds to construct the brain network of schizophrenia and normal subjects under different sparsity, calculating its attributes and making statistical analysis. (2) the weighted brain network attributes of schizophrenia and normal subjects were calculated and analyzed statistically. (3) Kruskal algorithm was used to construct MST. between schizophrenia and normal subjects. The attributes of MST network of schizophrenia and normal subjects were calculated and statistically analyzed, and the MST attributes with significant differences were extracted. (4) from theoretical analysis, The results of correlation analysis and classification show that the research of introducing MST into EEG brain networks is effective and feasible, and it is a useful supplement to the traditional analysis methods of complex brain networks. In a word, MST has better performance in the research of EEG brain network than the traditional complex network analysis method. As a useful supplement to the traditional complex network, it can provide a better way of thinking and method for the study of EEG brain network.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R749.3;O157.5
本文编号：2234233
[Abstract]:With the increasing of social pressure, mental diseases have become one of the main causes of sub-health, schizophrenia is the most common mental diseases, the early diagnosis of this disease is particularly important. At present, most diagnostic methods are based on the scale, and doctors make subjective judgment according to the scale. There are many subjective factors in this diagnostic method, which has an unpredictable impact on the diagnosis of schizophrenia. With the development of complex network theory and brain function imaging technology, the research of schizophrenia disease based on EEG brain network has made some progress, which provides a new idea for the pathological cognition and early diagnosis of the disease. However, recent studies have pointed out that traditional complex network analysis methods have some drawbacks. Traditional complex network research methods include weighted network and weighted network. The selection of threshold T is random in nature. The selection of threshold may lead to the existence of false or noisy connections (the selection of T is small), or it may discard some weak connections containing important information in the network (the choice of T is large). The number of edges further affects the measurement of brain network attribute values. The measurement of weighted networks is also affected by the noise connection and the average functional connection strength. Therefore, the analysis methods of the weighted network and the weighted network may lead to the inconclusive research results due to the methodological problems. Therefore, the minimum spanning tree algorithm (Minimum Spanning Tree,MST) is introduced into the study of complex brain networks, which is expected to solve the problems existing in the traditional network analysis methods and to explore the methodology of EEG brain network analysis. In the EEG brain network of schizophrenia and normal subjects, the three research methods of weighted network, weighted network and minimal spanning tree were compared and discussed. The EEG data used in this study were 40 schizophrenic patients and 40 normal subjects from Huilongguan Hospital in Beijing. After pretreatment, the data were constructed with phase lag index (PLI). Weighted and MST brain networks were used to classify the significantly different network attributes of different subjects as classification features. Finally, the theoretical analysis was carried out. The results of correlation analysis and classification show that the research of introducing MST into EEG brain networks is effective and feasible, and it is a useful supplement to the traditional analysis methods of complex brain networks. The main contents of this study are as follows: (1) selecting different thresholds to construct the brain network of schizophrenia and normal subjects under different sparsity, calculating its attributes and making statistical analysis. (2) the weighted brain network attributes of schizophrenia and normal subjects were calculated and analyzed statistically. (3) Kruskal algorithm was used to construct MST. between schizophrenia and normal subjects. The attributes of MST network of schizophrenia and normal subjects were calculated and statistically analyzed, and the MST attributes with significant differences were extracted. (4) from theoretical analysis, The results of correlation analysis and classification show that the research of introducing MST into EEG brain networks is effective and feasible, and it is a useful supplement to the traditional analysis methods of complex brain networks. In a word, MST has better performance in the research of EEG brain network than the traditional complex network analysis method. As a useful supplement to the traditional complex network, it can provide a better way of thinking and method for the study of EEG brain network.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R749.3;O157.5
【参考文献】
相关期刊论文 前2条
1 赵林;朱桂斌;文玉强;戚曹;;基于最小生成树的规则图像碎片复原算法[J];计算机技术与发展;2016年06期
2 杨晶;张兆鑫;王鹏;;最小生成树算法在城市基础建设中的应用[J];电子测试;2015年02期
，本文编号：2234233
本文链接：https://www.wllwen.com/yixuelunwen/jsb/2234233.html
最近更新
教材专著
