High-order Brain Network Analysis And Depression Recognition Based On Resting State Electroencephalography Data

At present,the number of people suffering from depression is rising year after year,and the disease has brought about significant mental and economic burden to patients and their families.A large number of studies have been conducted to construct brain functional networks based on brain timing signals to explore abnormal neural mechanisms and topological structures in patients with depression,which is of great significance for understanding the pathology of depression and for clinical diagnosis and treatment,and is also one of the current research hotspots in brain science.EEG signals provide extensive data support for this type of research due to their advantages of high temporal resolution and low acquisition costs.Brain functional connectivity network is an important tool to reveal the interaction between different brain regions,but most of the existing functional connectivity networks constructed based on EEG data are second-order networks,ignoring the high-order correlation between brain regions.How to construct a functional brain network that simultaneously reflects the interactions between multiple different brain regions is important for the diagnosis of depression and the analysis of brain abnormalities.In addition,functional connectivity within the brain changes over time,and this dynamic fluctuation is associated with cognitive and emotional functions of the brain.The abnormal functional connectivity pattern is also one of the important characteristics of patients with depression.Therefore,integrating the dynamic change information of functional connections into the construction of brain functional brain networks using the idea of functional network dynamic reconstruction is of great significance for exploring the brain topological structure of depressed subjects,and also provides a new idea for the construction of brain functional networks.The main work and innovation of this study are as follows:1.The weighted hyper-network model was proposed to construct the brain function network based on the resting state EEG data,and according to this model,depression patients were identified and their brain networks were analyzed.In this method,the sparse representation method of Least Absolute Shrinkage and Selection Operator(LASSO)is used to establish a hyper-network model and effectively represent the high order relationships of brain regions.On this basis,aiming at the problem that the weight of hyper-edge is not considered and single graph theory attributes are extracted in the existing research of hyper-network,the study add the correlation coefficients between the electrodes included in the hyper-edge to realize the hyper-edge weighting,and integrate hyper-edge weights into the calculation of characteristic path length,local efficiency and global efficiency graph theory attributes.Finally,the extracted graph theory attributes are used to classify and statistically analyze depression and normal controls.The experiment uses leave-one cross validation and multiple classification algorithms to train and evaluate the classifier.The classification results show that compared with traditional coupling methods,the network topology attributes extracted on weighted hyper-network achieve the best accuracy,with an 72.73%.The statistical results of network topology attribute showed that the Hypergraph clustering coefficient2 was significantly different between depressed subjects and normal controls.In addition,through the analysis of critical nodes and critical hyper-edges,it was found that the left temporal lobe,right frontal lobe,left parietal-occipital lobe and central lobe were highly correlated with depression.These findings may help to explore the pathological mechanism of depression,while weighted hyper-network methods also provide a new direction for the construction of brain networks.2.In order to capture the dynamic change characteristics of the brain on a time scale and reduce the complexity of directly analyzing dynamic functional connectivity networks,this study uses a sliding window method to construct a dynamic functional connectivity matrix on source location data to obtain a temporal sequence of functional connectivity.Based on the LASSO method for the first time,a hyper-network was constructed with functional connections as nodes to characterize the brain network topology,and three clustering attributes were extracted for statistical analysis.The study made the following findings: The differences in network attributes between depressed subjects and normal controls were mainly long-term connections across brain regions,in which the dorsal attention network and default mode network were highly involved,and these connections were mainly distributed in the frontal,temporal and parietal lobes.At the same time,there is no distribution of differential functional connections within the salient network,visual network and auditory network.Depressed subjects were significantly lower than normal controls on both cluster attributes,especially in the HCC2 attribute,indicating a weakening of information connections between brain regions in depressed subjects.The statistical results of network topology attributes also show that smaller or larger hyper-network parameters can not better reflect the differences in network structure.The above results also prove that constructing high-order networks based on functional connections can provide a method for exploring the brain mechanism of depression.The above study verifies the validity of using the weighted hypernetwork method and dynamic functional connectivity to characterize the brain functional network of depressed subjects on the EEG data,and provides a new idea for the construction of brain network and the selection of nodes in the network.