Study On Network Feature Extraction Of Static EEG Source Network Based On Network Entropy

At present,EEG technology is developing rapidly.As a widely used non-invasive neuroimaging technology,it is widely used in various neuroscience research and clinical data diagnosis.Therefore,it is a very important and meaningful work to study the extraction of more effective features from EEG signals for brain research.A common method in traditional EEG feature extraction is to construct a scalp brain network through multi-channel EEG signals and extract network parameters from it,but the existence of volume effect leads to inaccurate scalp network information.And the traditional scalp network parameter features cannot measure the network complexity well.EEG source location can effectively solve the volume effect and obtain spatial information at the cortical level.At the same time,thanks to the network entropy method,which can measure the distribution of network structure information and network complexity,this thesis proposes to use the network entropy method to extract features from the source localization brain network at the cortical level.Network entropy features were extracted from EEG and compared with a variety of traditional EEG features on EEG datasets with three visual stimulation tasks.The main work is as follows:1.A new generation of EEG source localization algorithm BC-VARETA is used to calculate the brain source and map the brain source to the brain area.The brain area is used as the node and the partial coherence matrix between the sources is used as the edge to form a static EEG source localization brain network?2.Four network entropy algorithms are introduced to extract the brain network feature,Node Degree Entropy is used to measure the network heterogeneity,Link Length Variation Entropy is used to characterize the difference in the weight change of the edge links in the undirected weighted graph,and using Average Link Length-Based Link Influence Entropy to characterize the average length change in the network after any edge link is deleted to measure the importance of the deleted edge link,and using Average Path Length-Link Influence Entropy to characterize the importance of special edge links such as bridging links?3.After the line graph transformation of the EEG source network,the activation value（node value）of the brain area is added to the network while retaining the network topology information,and the network entropy algorithm is used to extract the features again?4.Using the stable classification algorithm to classify the network entropy features and compare them horizontally with the traditional EEG features,it is found that the network entropy feature classification F1 score of the BC-VARETA static source brain network reaches 0.766,which is 0.08 higher than the traditional feature classification.Then,the network entropy features of the static EEG source brain network are sorted by importance under the optimal classification parameters,and it is found that the alpha frequency band features have the highest contribution to the classification.Starting from the static brain network after source location,this thesis innovatively uses the network entropy algorithm to extract topological complexity measurement features from the brain network at the cortical level.achieved excellent performance.The method of extracting static EEG source network features based on the network entropy can provide guidance for further related researches using EEG to analyze brain cognition and disease diagnosis in the future.