Research And Application Of High-order Functional Connectivity Network Construction Method Based On Rs-fMRI

The functional connectivity network(FCN)based on resting-state functional magnetic resonance imaging(rs-f MRI)can provide an objective basis for the diagnosis of neurological diseases.Although FCN has proven to be an effective auxiliary diagnostic tool,it still faces many challenges,such as how to effectively estimate the complex high-order functional connectivity(FC),and how to effectively combine dynamic FC and complex high-order FC to improve the diagnostic performance.To address those challenges,this research has carried out the following three tasks:(1)This paper proposed a method for constructing multi-view high-order FCN.Firstly,the FC time series in low-order dynamic FCN(Lo-D-FCN)are grouped into several clusters by clustering algorithm.Then,the FC time series in each cluster are integrated into central moment feature series by central moment method,and the correlation of the central moment feature series is calculated to estimate high-order FCN.The multi-view high-order FCN can be constructed by changing the order of the central moment,and each view reflects a type of high-level FC interaction pattern.This method has the following advantages:(1)This method takes both the dynamic characteristics of FC and the complexity of interaction patterns into account;(2)Using FC time series in Lo-D-FCN to estimate high-order FCN will produce large-scale high-order network,while clustering strategy can effectively reduce the scale of high-order FCN;(3)By fusing multiple views of high-order FCN,the performance of highorder FCN can be effectively improved,which is conducive to the comprehensive use of high-order FC information for disease diagnosis.(2)This paper proposed a method for constructing high-order FCN based on the central moment feature of the dynamic network.In this paper,Lo-D-FCN is regarded as a set of dynamic FC time series.The central moment method is used to extract the multi-order central moment feature matrix from Lo-D-FCN.By calculating the correlation between rows(or columns)of the feature matrix,multiple high-order FCNs can be estimated.This type of high-order FCN reflects the interaction of dynamic FC at the moment-level.The advantages of this method are:(1)the high-order FC interaction pattern based on central moment feature is helpful to deeply understand the dynamic characteristics of FC;(2)because the Lo-D-FCN has temporal sensitivity,while the central moment has translation invariance.Therefore,the high-order FCN based on the central moment features can effectively avoid the time sensitivity problem,thus allowing the consistent and meaningful comparisons across different subjects.(3)This paper proposed a method for constructing dynamic functional connection network for dimension reduction based on clustering.The main idea of this method is to integrate FC time series with similar characteristics in the dynamic functional connection network through the clustering method to discover the common interaction pattern inherent in the FC time series within the cluster.Then,the average value of the FC time series in each cluster is used as the new network node.Based on this strategy,a small-scale dynamic FCN can be constructed.The method is suitable for Lo-D-FCN and dynamic FCN with similar network structure as Lo-D-FCN.This method has the following advantages:(1)It can effectively integrate similar FC time series in dynamic FCN;(2)It can greatly reduce the scale of dynamic FCN,thereby reducing the redundant features in the network.The three FCN frameworks proposed in this paper are applied to the classification research of autism spectrum disorders,and obtains good classification results,which prove the effectiveness of the three functional connection networks proposed in this paper.