基于线图的无权脑超网络超边学习及融合特征分类

扫码查看

原文链接

NETL
NSTL
万方数据

中文摘要：脑功能超网络已广泛用于脑疾病的分类诊断中.在现有研究中,研究人员集中于改进脑功能超网络的构建却忽略了脑超网络拓扑对于分类诊断的影响,大多使用节点特征表征脑网络的拓扑.而研究表明超边信息能够弥补超网络的特征,同时超边间的传递有助于整体学习.考虑到该问题,提出基于线图的无权脑功能超网络超边学习,以分析超边对脑功能超网络拓扑和分类性能的影响.具体来说,首先,基于功能磁共振数据,使用星型扩展方法构建脑功能超网络;其次使用线图理论构建超网络的线图模型;然后使用超边密度提取线图的局部属性特征并使用非参数检验方法进行局部特征选择;接着使用基于图的子结构模式挖掘算法提取线图的子图特征并使用频繁分数特征选择方法选取判别子图;最后分别利用支持向量机构建分类模型.结果表明,所提方法分类结果优于传统脑功能超网络分类结果,达到86.79％.这表明脑功能超网络模型的超边拓扑信息影响分类模型的构建.此外,基于线图模型所求得的融合特征优于任一单一类型的特征,达到88.68％.可见对于超边拓扑信息提取,不仅需要考虑超边的属性信息,还需考虑超边间的空间传递信息能力.

外文标题：Hyperedge Learning of Unweighted Brain Hypernetwork Based on Line Graph and Fusion Feature Classification

外文摘要：Brain functional hypernetwork is widely used in the classification diagnosis of brain diseases.In the existing studies,the construction of brain functional hypernetwork is extensively studied to improve classification accuracy,but the impact of brain functional hypernetwork topology on classification performance is overlooked.However,it was showed that hyperedge information could make up for the characteristics of hypernetworks,and the transmission between hyperedges is conducive to overall learning.In view of this problem,hyperedge learning of unweighted brain hypernetwork based on line graph was proposed to analyze the influence of hyperedge on the topology and classification performance.Specifically,firstly,based on functional magnetic resonance data,the star extension method was used to construct brain functional hypernetwork.Secondly,line graph theory was introduced to construct the line graph model.Next,hyperedge density was extracted as local feature and nonparametric test was performed to select feature.Then,the graph-based substructure pattern mining algorithm was adopted to extract subgraph and the frequent score feature selection was applied to select discriminate subgraph.Finally,support vector machine was employed to construct classification model.The results show that the classification results of the proposed method are better than those of traditional brain function hypernetwork,reaching 86.79％.It is concluded that the topological information of the brain function hypernetwork affected the construction of the classification model.In addition,the fusion features obtained based on the line graph model are better than any single type of features,reaching 88.68％.It is suggested that that for hyperedge topology information extraction,the attribute information of the hyperedge needs to be considered,and the spatial transfer information ability between hyperedges needs to be considered.

外文关键词：

resting state functional magnetic resonance imagesbrain functional hypernetworkline graphhyperedge densitymachine learningAlzheimer's disease

作者：

上官学奎、黄晓妍、王春燕、郭浩

展开 >

作者单位：

山西省信息产业技术研究院有限公司软件工程事业部,太原 030012

太原理工大学计算机科学与技术学院(大数据学院),晋中 030600

关键词：

静息态功能磁共振影像脑功能超网络线图超边密度机器学习阿尔兹海默症

基金：

国家自然科学基金国家自然科学基金山西省科技厅基础研究计划山西省科技厅基础研究计划山西省科技厅基础研究计划

项目编号：

6187612461873178202103021231292021030212416620210302123099

出版年：

2024

DOI：

10.12404/j.issn.1671-1815.2305981

科学技术与工程

中国技术经济学会

科学技术与工程

CSTPCD北大核心

影响因子：0.338

ISSN：1671-1815

年,卷(期)：2024.24(21)