摘要
目前,大多数关于轻度认知障碍的分类是以单个脑区在整个脑网络中的局部相关性作为分类特征的,其忽略了多个脑区间相互作用的信息,并且当前绝大部分研究是基于无权脑网络的,丢失了脑区间相互作用的强度。为了更好地捕捉到能区分脑网络的关键拓扑结构信息,本文提出了一种关键子网选择算法,首先计算加权脑网络中的最短路径,并根据正负类样本中最短路径出现的次数差异来选择差异性显著的特征,然后再根据选择出的最短路径上的节点来构建关键子网,最后利用图核对子网数据进行分类。通过将实现了所提出算法的加权脑网络、无权脑网络以及全脑网络进行对比实验,验证了该方法在轻度认知障碍分类上的有效性。
At present, most of classifications of mild cognitive impairment are based on the local correlation of a single brain region in the whole brain network, which ignores the information of interaction between multiple brain regions. Moreover, most of the current researches are based on the unweighted brain network, losing the intensity of interaction between brain regions. In order to better capture the key topology information that can distinguish brain networks, this paper proposes a key subnet selection algorithm. Firstly, the shortest path in weighted brain network is calculated, and features with significant difference are selected according to the difference of the number of the shortest path in positive and negative samples. Then, the key subnets are constructed according to the nodes on the selected shortest paths, and finally graph kernel is used to classify subnet datasets. By comparing the weighted brain network, the unweighted brain network that implement the pro-posed algorithm and the whole brain network, the effectiveness of this method in the classification of mild cognitive impairment is verified.
出处
《计算机科学与应用》
2020年第9期1571-1579,共9页
Computer Science and Application
关键词
加权脑网络
最短路径
子网选择
图核
Weighted Brain Network
Shortest Path
Subnet Selection
Graph Kernel
