Abstract: Brain network classification methods have attracted a lot of attentions in the fields including brain science and brain disease diagnosis. However, most of existing studies on brain network classification use brain regions or the correlation between paired brain regions as classification features, which cannot reflect the topology information among multiple brain regions. To overcome the problem, this paper proposes a novel brain network classification method with subgraph selection and graph-kernel-based dimensionality reduction. Firstly, this method mines two groups of frequent subgraphs from positive and negative classes respectively, and then selects the most discriminative subgraphs using the subgraph selection algorithm based on their respective frequencies difference. Secondly, it uses the graph-kernel-based principal component analysis (GK-PCA) method to perform feature extraction on the graph data with selected subgraphs. Finally, it adopts the support vector machine (SVM) to perform classification on the data with extracted features, validates the proposed method on real brain network dataset of mild cognitive impairment (MCI), and the experimental results show the efficacy of the proposed method.

Key words: subgraph mining, feature selection, graph-kernel-based dimensionality reduction, brain network classification

摘要： 脑网络分类在脑科学研究和脑疾病诊断等领域引起了学者们的广泛关注。目前大多数有关脑网络分类的研究都是以单个脑区或成对脑区之间的相关性作为分类特征，其缺点是不能反映多个脑区之间的拓扑结构信息。为克服上述缺点，提出了一种基于子图选择和图核降维的脑网络分类方法。具体包括：（1）分别从正类训练样本组及负类训练样本组中提取多个频繁子图，进而利用基于频度差的子图选择算法选取最具判别性的子图集；（2）基于上述过程中得到的子图集，利用图核主成分分析（graph-kernel-based principal component analysis，GK-PCA）方法对经过子图选择后的图数据进行特征提取；（3）利用支持向量机（support vector machine，SVM）在特征提取后的数据上进行分类。在真实的轻度认知障碍（mild cognitive impairment，MCI）脑网络数据集上对该方法进行了验证，实验结果表明了该方法的有效性。

关键词: 子图挖掘, 特征选择, 图核降维, 脑网络分类