Research On Construction Method Of High-order Brain Network Based On EEG

In recent years,depression,as a mental disorder,has seen a growing prevalence in our country,making it imperative to conduct research on depression.A viable research method,Brain Function Networks(BFN),has been proposed and widely applied in the diagnosis of Major Depressive Disorder(MDD)based on Electroencephalography(EEG).Traditional methods generate brain function networks for analysis by calculating Functional Connectivity(FC)between paired electrodes.However,the human brain is a complex system,and even a simple brain activity cannot be adequately analyzed using features from a single brain area or the correlation between two brain areas.We refer to the traditional brain network as Low-order BFN(Lo-BFN),and the novel brain function network that reflects the connections between multiple channels as High-order BFN(Ho-BFN).In this thesis,two tasks are carried out for the construction and analysis of this network:(1)A method based on Matrix Variate Normal Distribution(MVND)is proposed to construct Ho-BFN.This framework can simultaneously generate Lo-BFN and Ho-BFN based on the features of EEG time series signals and has a clear mathematical explanation.Specifically,the entire time series is divided into multiple segments,and for each segment,the BFN of that period is constructed by calculating the phase lag indices between different EEG channels.Finally,using the MVND method,the distribution of BFN changes in the time series is estimated,obtaining both Lo-BFN and Ho-BFN.(2)A method based on the Central Moment to capture the time-varying information for constructing Ho-BFN is proposed.Specifically,after dividing the EEG time series into different segments,we construct BFNs within these segments and then use the Central Moment method to capture the changing relationships of these BFNs over time,obtaining the Central Moment-based BFN.Finally,by using the "correlation of correlation" theory,the relationships between multiple channels are captured to construct Ho-BFN.The experimental results show that the high-order brain function networks proposed in this study provide an innovative high-order information perspective for the diagnosis of depression.These high-order information complement the low-order information obtained by traditional methods,offering a more in-depth and comprehensive theoretical basis for the accurate diagnosis of depression.Meanwhile,we analyze the BFN differences between MDD patients and healthy controls,and the results find that the main differences occur in the left temporal and frontal lobes,which may provide potential biomarkers for the clinical diagnosis of MDD patients.