22-Effect of spatial smoothing on graph analysis of fMRI data

Abstract

Introduction Smoothing is used to increase SNR. Here we examine it’s effect of smoothing on graph analysis of fMRI data. Methods We used fMRI data of 30 subjects from HCP (Human Connectome Project; 3T; motor task). We preprocessed the data using realign and unwarp, spatial normalization, filtering and smoothing with FWHM = {none, 3, 5, 8 and 11} mm, parcellated according AAL atlas, used mean and first eigenvector as representative signal. Finally, we performed graph analysis using average node strength, characteristic path length, lambda, efficiency, clustering coefficient, and gamma. Results Increase of FWHM is related to increase of clustering coefficient, node strength and efficiency, and to decrease of path length. Gamma and lambda are relatively not influenced by size of FMWH kernel. With low FHWM or no smoothing and using mean is the node strength higher and path length shorter. Discussion Increase of FWHM increases correlations in the network, i.e. weights in graph. Therefore node strength increases and characteristic path length decreases. Clustering coefficient also reflects increasing weights between neighboring nodes. Increase of correlation coefficients with higher FWHM is more prominent when using first eigenvector as representative signal. We recommend to smooth fMRI data consistently across study.