Methods Of Meg Functional Brain Network Analysis And Toolbox Devolopment

Magnetoencephalography(MEG)is an advanced magnetic source imaging technology,it can measure the brain magnetic fields noninvasively.MEG has a quite high temporal resolution,with the help of source reconstruction algorithms,MEG signals can be projected on the cerebral cortex.Source signals can be interpolated with Magnetic Resonance Imaging(MRI),which has a relatively high spatial resolution.This paper proposed an improved method for MEG functional brain network construction.Traditional researches usually choose the source signal with max power to represent the neural activities of an entire ROI,which obviously lost a lot of information.We proposed two solutions to improve:(1)performing a check on(and adjustment of)the polarity of the time-series of neighboring voxels before averaging time-series across a ROI;(2)clustering source signals of a ROI and choose the cluster center which contains max number of voxels.We choose 20 subjects from HCP database to test these solutions.A k-means cluster was applied to the connectivity matrixes obtained from three kinds of methods,to cluster the states of working memory and motor task.When use correlation coefficient to calculate the connectivity matrix,all three methods can get a high accuracy rate.However,when use phase coherence,only the cluster-based method can get a reliable result.Furthermore,when it comes to phase lag index,all three methods resulted a relatively low accuracy rate.This paper also proposed a sliding-window-based framework for dynamic functional connectivity analysis with MEG data.We performed source reconstruction on windowed pieces of MEG data respectively,to obtain a series of connectivity matrixes,which represent the dynamic changes of brain functional connectivity.We apply this framework to 16 subjects from HCP database.The result showed that even in resting-state,the connectivity of brain changes dynamically.A k-means algorithm was applied to the connectivity matrixes,got 5 states.One state occurred most frequently and it showed similar pattern with default mode network.It can be regarded as a baseline of dynamic changes of brain network.We also developed an easy-to-use toolbox for MEG analysis based on Matlab and FieldTrip,named EasyMEG.EasyMEG contains a set of functions for MEG analysis: preprocessing,time-lock analysis,time-frequency analysis,source analysis and plotting.It can meet the analysis needs of different kinds of users.