Motor Imagery Recognition Based On Functional Network And Effective Network Analysis

Motor imagery is an emerging rehabilitation medical method,which can activate motor-related brain areas just like active exercise.Through the brain-computer interface based on motor imagery,the information interaction between the brain and external devices can be realized,which brings hope to patients with movement disorders and brain injuries.By describing the functional connection and effective connection between brain regions,the brain network represents the information interaction of the brain cortex when the subject performs the task,and is an effective method for analyzing brain function of motor imagery.Therefore,thesis uses functional brain network and effective brain network as carriers to study motor imagery.The main contents and innovations of thesis are as follows:(1)Thesis proposes a bilevel functional brain network construction method based on corticomuscular coupling node selection.The electrical signals of body movements originate from the brain cortex,so a coupling relationship is sure to exist between electroencephalogram(EEG)and electromyography(EMG).Based on corticomuscular coupling analysis method,thesis quantified the coupling intensity between 32 channels EEG and 4 channels related EMG signals corresponding to hand movements.Based on the core nodes and the prior knowledge of motor sensory brain region in neurophysiology,the core nodes combinations under each movement are obtained,and the regional functional brain network composed of sensory motor core nodes is constructed.Then extract the average node degree,average clustering coefficient and average path length of the regional functional brain network.Motor imagery is the result of whole-brain coordination.Although the regional functional brain network is highly targeted,it may still miss some uncertain factors in motor imagery.Based on the full network characteristics of the minimum spanning tree,combined with the diameter and average eccentricity features of the minimum spanning tree brain network used to describe the global characteristics,the bilevel functional brain network comprehensive feature combining the minimum spanning tree and the core nodes regional functional brain network is constructed.(2)Thesis proposes an effective brain network construction method of the left and right hemispheres and the whole brain based on transfer entropy.Traditional directed transfer function(DTF)and partial directed coherence(PDC)are used to analyze Granger causality,but both DTF and PDC are based on linear models.Considering the complexity and nonlinearity of the brain,thesis chooses to use the transfer entropy,which does not need to assume the interaction model and is nonlinear in nature,to quantify the causal relationship between brain network nodes.Through the functional network analysis of left and right hand movements,it was found that body movements mainly activated the contralateral brain cortex region.According to this phenomenon,in order to clearly show the difference between left and right hand movements,thesis divides the effective brain network into left and right hemispheres effective brain network,and extracts the average node degree and average clustering coefficient features.However,considering that this analysis method ignores the global characteristics of the brain network,the core nodes in functional brain network analysis play the role of simplifying the network edges and highlighting the core nodes.Therefore,thesis selects 14 nodes to form a whole brain effective brain network by referring to the core nodes combination,and extracts the global efficiency and average path length features that describe the global attributes of the network.Thus,the effective brain network features not only highlight the difference between left and right hand movements,but also take global consideration.(3)After the brain network was established from different aspects and features were extracted,various features were input into the support vector machine(SVM)for classification experiments to verify the effectiveness of the feature extraction method described in thesis.The experimental results show that the classification accuracy of functional network and effective network analysis method is comparable.The classification accuracy of motion imagery experiment is close to that of active exercise experiment.If the features of functional network and effective network are integrated,higher motor imagery recognition accuracy can be obtained.The research results show that the combination of global and local features proposed in thesis is correct in the analysis of motor imagery brain network,which further enriches the research methods and related research results of motor imagery.