Granger Causality Analysis And Its Application In FMR

Since the1990s, the blood oxygen level dependent functional magnetic resonanceimaging (BOLD-fMRI) technology is a representation of brain imaging technology. Anda large number of related fMRI data analysis methods emerge, which provides a newopportunity to explore the interactions of human brain neural information.Neuroimaging studies have suggested that we can indirectly record neural signalsaccording to BOLD fluctuation. New research tools are developed to use the BOLDinformation in the brain to investigate the functional integration and functionalsegregation. In the present study, we focus on the effective connectivity to explore theinteraction between distributed brain areas.This paper aims to provide supplements and improvements on current methods ofgranger causality analysis, which has been widely applied in fMRI data. Four aspects ofthis study have been put forward:1. Wavelet multispectral approach-based granger causality analysis: This paperapplies the method of the multispectral wavelet to decompose and reconstruct BOLDsignals. The method is able to strength the characteristics of the original signal viareconstruction of low frequency signals. What’s more, the accuracy of granger causalitywas improved.2. Blind deconvolution approach-based granger causality analysis: In this sectionwe aim to share the method of blind deconvolution on BOLD signals. Based on thehypothesis that spontaneous neural activity in resting-state is similar to the neuralactivity in task status, we detect the local peak whose value is higher than a giventhreshold, and further estimate parameters such as time-delay to extract hemodynamicfunction from BOLD fMRI. This approach doesn’t need prior knowledge and is verysimple for use. At the end, we stimulate the blind deconvolution approach to confirm itsvalidity in the granger causality analysis.3. Partial Conditioning approach-based granger causality analysis: Based on theinformation theory, granger causality is completely equivalent to transfer entropy underthe condition of Gaussian, thus we can calculate granger causality through mutual information indirectly. And it is effective during conditioning on a large number ofvariables.4. Finally, we have a brief application of the improved condition of partial grangercausality analysis on Board Game Experts and amateur players. The results demonstratethat there is more interaction in master, and the functional modularity is stronger thanthe amateur.