判别性子图挖掘方法及其在MCI分类中的应用  

Discriminative subgraph mining with application in MCI classification

作　　者：费飞[1] 王立鹏[1] 接标[1] 张道强[1] 

机构地区：[1]南京航空航天大学计算机科学与技术学院,南京210016

出　　处：《南京大学学报（自然科学版）》2015年第2期328-334,共7页Journal of Nanjing University（Natural Science）

基　　金：江苏省自然科学基金杰出青年基金(BK20130034);高等学校博士学科点专项基金(20123218110009);南京航空航天大学基本科研业务费(NE2013105);中央高校基本科研业务专项资金(NZ2013306)

摘　　要：最近,脑连接网络已经被用于神经退行性疾病(如阿尔茨海默病AD以及轻度认知障碍MCI)的诊断和分类.以往典型方法是从脑连接网络中提取一些特征(如局部聚类系数等)构成一个长特征向量,并用其训练一个分类器用于最终的分类.然而,上述方法的一个缺点是未能充分考虑网络的拓扑结构信息,因而限制了分类性能的进一步提升.提出一种基于判别子图挖掘的脑连接网络分类方法.首先分别从正类训练样本集和负类训练样本集中挖掘频繁子网络(即频繁子图);然后利用基于图核的方法来衡量频繁子网络的判别性能,并选择那些最具判别性的频繁子网络作为判别子网络用于后续的分类;最后,在真实MCI数据集上的实验验证了该方法的有效性.Brain connectivity networks have been recently used for classification of neurodegenerative diseases,e.g.,Alzheimer＇s disease（AD）and mild cognitive impairment（MCI）.Accurate diagnosis of AD,as well as its prodromal stage（MCI）,is very important for possible delay and early treatment of the disease.In conventional connectivity-networks-based classification method,features（e.g.,local clustering coefficient,etc.）are often extracted from connectivity-networks and concatenated into a long vector to train a classifier for final classification.However,one disadvantage of those methods is that some useful network topological information was not fully considered,which limits the further improvement of classification performance.Accordingly,in this paper,we propose a novel brain connectivitynetwork classification method based on discriminative subgraph mining,which can reflect the intrinsic disease pathology.Specifically,in preprocessing stage,we firstly use the specific application tool to original fluoro-deoxy-glucosepositron emission tomography（FDG-PET）images data from 27 subjects.For each preprocessed original FDG-PET image,90 regions of interest（ROIs）are labeled by an atlas warping algorithm.And then,we construct the connectivity networks according to the fiber matter between these brain regions.After,we extract a set of frequentsubnetworks（i.e.,subgraphs）with using frequent subgraph mining algorithm from each of the two groups（i.e.,MCI and NC）,respectively.Finally,we measure the discriminative ability of those frequent subnetworks using graph-kernel-based classification method and select the most discriminative subnetworks for subsequent classification.The classification experiment scheme is tested on the MCI dataset by leave-one-out（LOO）cross-validation method.The experimental results show the efficacy of our proposed method with comparison to the state-ofthe-art method for connectivity-networks based MCI classification,and our method can gain a better insight in the disease patho

关 键 词：轻度认知障碍 结构连接网络 频繁子图挖掘 图核 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]