遗传社团划分算法揭示静息态fMRI社团结构

摘要

如何从静息态fMRI(functional magnetic resonance imaging)数据构建的脑功能网络中得到脑网络的社团结构及发现脑功能网络的差异性是近来研究的焦点.现有的划分算法大多需要人为的设定参数,从而得到较好的社团结构.然而,这样的社团划分存在着较大的局限,无法实现自动划分.文中提出了一种新的脑网络社团划分方法,利用遗传算法操作来改变节点之间的连接关系,同时演化计算最大化的模块度值,无需预先划分个数就能得到最大模块度值对应的社团结构和社团个数.为了测试划分算法的有效性,首先在一系列标准的社会网络上进行了测试,证明了该方法能够划分出更好的社团结构.然后将算法应用在脑网络数据上,分别是由自闭症被试和正常被试构建而成的脑功能网络,并发现了一些有意义的社团结构.最后对得到的社团结构进行了分析,结果表明分析脑功能网络的社团结构是有效的.%How to detect the community structure of brain networks from the rs-fMRI(resting state functional magnetic resonance imaging)and their difference is the focus of the research.Most of the existing partitioning al-gorithms need to set parameters for getting a better community structure.Actually, these methods are relatively limited because they can′t detect community automatically.In this paper, we propose a novel method named GAcut to detect community structure on rs-fMRI.We use genetic algorithm to change the connection between nodes.At the same time,optimizing Modularity Q and automatic detecting the community structure without any parameters.In order to test the effectiveness of GAcut,we first use the algorithm on some real-world networks. The results indicate that GAcut is more effective than other popular methods.In rs-fMRI testing step, we use GAcut on the rs-fMRI data to compare different subjects(autism spectrum disorders and normal).We find and analyze some special communities.The results show that these communities are meaningful.