基于优化零模型的脑功能超网络拓扑结构分析及多特征融合分类

Topological Structure Analysis of Brain Functional Hypernetworks Based on Optimized Null Models and Multi-feature Fusion Classification

扫码查看

原文链接

维普
万方数据

中文摘要：脑功能超网络已广泛用于脑疾病分类,但大多研究集中于改进静态超网络模型的构建来提高分类准确率,主要使用聚类系数等常见节点拓扑属性量化脑功能超网络,忽略了超网络拓扑属性所包含的有效信息.然而研究表明超边所体现的高阶模式可以更好地表征脑网络整体结构,超边间的交互则可以反映超边间的空间传递信息能力,有助于更好地理解脑疾病的神经机制.考虑到该问题,提出多组超边属性,并结合已有节点属性综合表征脑功能超网络拓扑信息.然而,由于超网络超边特性,导致拓扑信息计算代价较为昂贵,并且多个拓扑属性之间可能存在冗余性较高的问题.因此,进一步优化已有零模型使其适配脑功能超网络,并基于优化的零模型评估拓扑属性间的依赖关系,最终融合依赖性较小的拓扑属性参与分类模型的构建.结果表明,拓扑属性间的依赖关系有所不同,含有超边的拓扑属性与其余拓扑属性的依赖性较小.此外,融合依赖性较小的多拓扑属性特征的分类性能优于传统方法,达到89.39％.这表明提取脑功能超网络的拓扑信息时,不仅需要考虑节点拓扑属性,还需考虑超边信息.同时在多组拓扑属性表征脑功能超网络拓扑信息基础上,选取依赖性不强即冗余性较低的拓扑属性能够提高组间差异表征能力,最终提高脑疾病分类性能.

外文摘要：Brain functional hypernetwork is widely used for the classification of brain diseases.However,in most of the studies,the clas-sification accuracy is mainly increased by improving the construction of resting-state hypernetwork model,and common node topological at-tributes,such as clustering coefficients,are mainly used to quantify the brain functional hypernetwork.The effective information contained in the topological attributes of the hypernetwork is ignored.It is shown that the higher-order patterns embodied in the hyperedges can better characterize the overall structure of the brain network,and the interactions among the hyperedges can reflect the spatial transfer of infor-mation among the hyperedges,which can help to better understand the neural mechanisms of brain diseases.Considering the problem,multiple sets of hyperedge attributes were proposed and combined with existing node attributes to comprehensively characterize the topolog-ical information of the brain functional hypernetwork.However,the topological information is more expensive to compute due to the hyper-edge property of the hypernetwork,and the high redundancy between multiple topological attributes may exist.Therefore,the existing null model was further optimized to adapt to the brain functional hypernetwork,and the dependencies among topological attributes were evalua-ted based on the optimized null model.Finally the topological attributes with less dependency were fused to participate in the construction of the classification model.The results show that the dependencies among topological attributes are different,and the topological attributes containing hyperedge has less dependency with the rest of topological attributes.In addition,the classification performance of fusing mul-tiple topological attribute features with less dependency is better than the traditional method,reaching 89.39％.This indicates that not only node topological attributes but also hyperedge information need to be considered when extracting the topological information of brain functional hypernetwork.Meanwhile,on the basis of multiple topological attributes characterizing the topological information of brain functional hypernetwork,the topological attributes with less dependency,i.e.,lower redundancy,can be selected to improve the ability to characterize the differences between groups,and the performance of classifying brain diseases can ultimately be improved.

外文关键词：

resting-state functional magnetic resonance imagingbrain functional hypernetworknull modelhyperedge topological attributesdepressionclassification

作者：

王春燕、郭浩、杨艳丽、黄晓妍、相洁、李瑶

展开 >

作者单位：

太原理工大学计算机科学与技术学院(大数据学院),晋中 030600

太原理工大学软件学院,晋中 030600

关键词：

静息态功能磁共振成像脑功能超网络零模型超边拓扑属性抑郁症分类

基金：

国家自然科学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山西省科技厅基础研究计划

项目编号：

618761246187317862376184202103021230992021030212312920210302124166202203021222095

出版年：

2024

DOI：

10.12404/j.issn.1671-1815.2400581

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(27)