EEG Research On Working Memory Based On Improved K-order Propagation Number

Working memory(WM)is a system in the brain for temporary storage and processing of information during the execution of complex cognitive tasks.Its capacity and accuracy are highly correlated with the level of human intelligence,and it is also the first function to be impaired by neurological diseases such as Alzheimer’s disease,and a key link in cognitive activities.Since the brain always has multiple functional regions working together to form a network to accomplish cognitive activities,this thesis analyzes WM EEG data from the perspective of brain networks to explore the intrinsic mechanism of WM.Since the traditional algorithms have certain shortcomings,this thesis proposes a new node importance assessment algorithm applicable to directed networks based on the weighted K-order propagation number(WKPN)algorithm,i.e.,the cross K-order propagation number(CKPN)algorithm.The CKPN introduces importance adjustment factor λ to fuse nodes sending and receiving information perspective to assess the node importance,and introduces propagation time K-factor to achieve a comprehensive utilization from local to global network topology information.Various algorithms are used to explore the node importance of ARPA network,and the CKPN evaluates more reasonable results than the comparison algorithms.The node importance of the Residence hall and Celegansneural networks is analyzed using a weighted SIR model and a deliberate attack strategy for important nodes,and the results show that the node importance sequence obtained by the CKPN fits best with the node infectivity evaluated by the weighted SIR and reduces the global efficiency of the network by 90% with only a minimum number of attacks approximately,which prove the superiority of the CKPN.The improved K-order propagation number algorithms-WKPN and CKPN-are applied to the WM study.In this thesis,an n-back experimental paradigm based on English letter sequences is designed to acquire WM EEG data,and preprocessing operations are performed on the collected EEG raw data.Undirected brain networks are constructed based on phase locking value(PLV)and the topological properties of the networks are analyzed.The results show that in WM readout state brain networks have higher global and local efficiency.Meanwhile,the important nodes of the PLV brain network are evaluated using WKPN,and the results show that the important brain regions of WM update and readout states are mainly distributed in the frontal and parieto-occipital lobes,and the important nodes are more distributed in the right side of the frontal lobe and parieto-occipital region as the task difficulty increased,implying that the brain tends to use the right half of the brain for better reading of memory targets.Directed brain networks are also constructed by using the phase transfer entropy(PTE)index,and their node importance are evaluated by applying CKPN.The important nodes of PTE networks in WM update and readout are mainly distributed in the frontal and parietooccipital regions.The information flow direction between important nodes indicates that there is a control relationship from the frontal lobe to the parieto-occipital region in WM update state.To explore good classification features to improve the classification accuracy of WM states,considering that the network domination entropy is sensitive to the subtle changes in network structure,this thesis first extracts the PLV significant node network using WKPN,and then constructs the domination entropy classification features and feeds them into the support vector machine to classify the WM update and readout states.The cross-subject classification accuracy is 78.73% and the classification time cost is the lowest,which proves that domination entropy is an effective and high-quality classification feature.