Representative Signal Extraction in MEG Network

For the construction of MEG brain network, the source signal with the strongest power in each brain region is generally selected to represent its neural activity, which causes the loss of information. In this study, two improvement schemes were proposed for this problem: research methods based on superposition averaging and cluster-based research methods. To evaluate the scheme, 51 subjects were selected with MEG data under task state. Signals were extracted by using maximum power, superimposed average and clustering, respectively, from each band.The brain function network was constructed and clustered. Additionally, the characteristics of brain network constructed by the three methods were compared. Based on the results, the superposition averaging method acquired the highest accuracy, then the highest power method and clustering acquired lower accuracy. Brain network feature analysis found that those based on the method of superimposed average and maximum power have strong small world attributes, while brain networks constructed using clustering methods have weaker world attributes. From that, it is feasible to construct MEG brain network using the superimposed average and maximum power methods.