抑郁症静息态功能脑网络社团结构差异研究

发布时间：2018-10-10 16:35

【摘要】：对于大部分复杂网络,不论是社会网、经济网还是生物网,都潜在的具有模块性,被认为是复杂网络的一个主要组织原则。而图论为我们提供了对复杂网络进行定量的分析方法。人脑也可以抽象成为复杂网络。人脑的各个脑区可以抽象为节点;而如果脑区与脑区存在功能上的关联,那么脑区间的这种功能上的关联就可以抽象成一条边。社团结构是复杂网络的基本性质之一,通过分析网络的社团性质,可以理解脑网络的结构性质和功能关系。本文对fMRI信号低频涨落成分进行同步性分析,构建出脑功能连接网络,并基于"堆结构"的模块划分算法对其进行模块划分,并使用图论中有关社团结构以及节点的拓扑角色等相关分析方法去分析静息态下人脑功能连接网络,探索抑郁症患者在脑网络性质和连接上是否发生异常。通过实验分析研究发现抑郁症患者与正常人的脑功能连接网络在社团结构及脑区节点角色等方面有显著差异,而这些差异可能是导致抑郁症病发的深层原因。首先,我们实验表明,不论正常人还是抑郁症患者的脑网络模块度相对随机网络来讲具有显著差异。在稀疏度范围2.5%至10%内对两组的模块度做AUC,没有发现组间显著性差异,但在某特定网络稀疏度(5%,7%)下,发现模块度有显著差异。第二、对于健康组、抑郁组以及可比较的随机网络,随着脑网络稀疏度的增加,模块度(Modularity)单调下降。即在一定的阈值空间内,模块度是边数的减函数。当网络在最稀疏的100条边时,模块度最高。第三,论文中引入了"小核团"及"亲密度"的概念,通过研究并发现,对比正常组,抑郁症患者某些模块间的连边数量发生改变。第四,实验表明,两组对应的模块组成、脑区节点角色及模块角色有差异。这些差异的发现为我们在疾病诊断方面提供了一些比较重要的研究和临床应用价值。最后,对全局脑属性模块度Q以及90节点的Z值和P值进行属性选择,选择分类性能比较好的节点。并对这些属性进行训练,最终构造出分类器。在本次研究中,筛选出分类性能排名前10的属性节点,并随机选取样本容量的70%来对这些属性进行训练,构造出分类器,最后用剩下的30%的样本容量对其进行测试。测试结果表明,分类的正确率达到90%以上。
[Abstract]:For most complex networks, whether they are social networks, economic networks or biological networks, they are potentially modular and are considered as one of the main organizational principles of complex networks. Graph theory provides us with a quantitative analysis method for complex networks. The human brain can also be abstracted into complex networks. Each brain region can be abstracted as a node, and if there is a functional correlation between the brain region and the brain region, the functional correlation between the brain regions can be abstracted into a single edge. Community structure is one of the basic properties of complex networks. By analyzing the community properties of networks, the structural properties and functional relationships of brain networks can be understood. In this paper, the low frequency fluctuation component of fMRI signal is analyzed synchronously, and the brain functional connection network is constructed, and the module partition algorithm based on "heap structure" is proposed. In order to explore the abnormal nature and connection of brain network in depression patients, correlation analysis methods such as community structure and topological role of nodes in graph theory are used to analyze the brain functional connectivity network in resting state. Through the experimental analysis, it is found that there are significant differences in the association structure and the role of the brain node between the depression patients and the normal people, and these differences may be the underlying causes leading to the occurrence of depression. First of all, our experiments show that there are significant differences in the degrees of brain network modules between normal and depressed patients compared with random networks. In the range of sparsity from 2.5% to 10%, there was no significant difference in AUC, between the two groups, but there was a significant difference in modularity under a given network sparsity (5% or 7%). Secondly, for healthy group, depression group and comparable random network, the modularity (Modularity) decreased monotonously with the increase of brain network sparsity. In a certain threshold space, modularity is a minus function of the number of edges. When the network is on the thinnest 100 edges, the modularity is the highest. Thirdly, the concepts of "small nuclei" and "affinity" are introduced. Fourth, the experiment shows that there are differences between the two groups of corresponding modules, the role of the node in the brain region and the role of the module. The discovery of these differences provides us with some important research and clinical application value in disease diagnosis. Finally, the global brain attribute modularity Q and the Z and P values of 90 nodes are selected to select the nodes with better classification performance. Finally, the classifier is constructed by training these attributes. In this study, the top 10 attribute nodes were selected, and 70% of the sample size was randomly selected to train these attributes to construct the classifier. Finally, the remaining 30% sample size was used to test them. The test results show that the correct rate of classification is over 90%.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：R749.4;O157.5

【相似文献】