Topological Analysis And Classification Research Of High-order Uncertain Brain Network Based On Independent Component

Resting-state functional magnetic resonance imaging technology combining with complex brain network theory has become one of the importa nt methods to study current nerve diseases.This method can accurately diagnose various mental sickn ess including depression.However,resting-state functional networks do not take into account dynamic uncertain functional con nectivities b etween brain regions.At the same time,traditional brain network analy sis often requires the conversion of uncertain graphs into certain graphs by setting th resholds,but there is no consensus on the threshold setting range in the current literatures.Previous researchers have used the minimum span ning tr ee method to analyze the brain networks.However,the minimum spanning tree might underestimate the im portance of low-weight connectivities and clusters in brain network inform ation processing,and does not fully consider the information about the network topology,which limits the further improvement of classification performance.In order to solve these problems,this paper proposes a method the high-order uncertain resting-state functional brain networks based on the independent component.The method does not need to rely on the priori brain map templates,nor does it need to select thresholds.It also takes into account the time-varying characteristics of the scan time,which can more precisely model the brain networks and retain more functional connectivity network details.At the same time,in order to find discriminative subgraph features more precisely,several new discriminative feature selection methods are proposed.The classification results show that the high-order uncertain resting-state functional brain networks based on the independent component effectively improves the accuracy of depression diagnosis.The main researches of this paper are as follows:Firstly,resting-state functional connectivity network based on independent component analysis is constructed.Independent component analysis does not need to rely on the priori brain map templates,which avoids the difference in analysis results caused by different brain map templates,and could fully reflect the spatial relations between voxels,obtain a more reasonable functional connectivity pattern.Secondly,the method to construct a high-order uncertain brain network is proposed,which fully consider the dynamic functional connectivities and uncertain information in the brain network,and improve the expression ability of the brain network.There is no need to make threshold selections,and the network details are preserved as much as possible,revealing a larger,more ab stract interaction.Thirdly,in view of the problem that the classification accuracy of the existing discriminative feature selection method of uncertain graph is not high,several new discriminative feature selection methods are proposed.This method calculates the statistical measures of the probabilities of occurrence of the subgraph pattern in the sample.Combining with the common discriminative score function,the most discriminative subgraphs are selected as features.This paper is the major component of the National Natural Science Foundation of China,"High-order Complex Space-Time Effects Analysis and Modeling Research of Resting-state Functional Brain Networks"(No.61876124).This study has also been supported by research grants from Natural Science Foundation of Shanxi Province(201801D121135),Scientific and Technological Innovation Programs of Higher Education Institutions in Shanxi(2016139),Key Research and Development(R&D)Projects of Shanxi Province(201803D31043)and CERNET Innovation Project(NGII20170712).This paper focuses on the construction and classification methods of high-order uncertain brain networks,and proposes several new discriminative feature selection methods in order to find reliable biological indicators and assist clinical diagnosis.