A Data and Model-Driven Approach to Explore Inter-Subject Variability of Resting-State Brain Activity Using EEG-fMRI

In this paper, we investigate the origin of the large inter-subject-variability of EEG-fMRI correlation patterns. For that purpose, a simplified representation of the fMRI signal is obtained by using a hierarchical clustering algorithm detecting spatial patterns of mutually correlated voxels. The general-linear model is subsequently used to determine which of the identified patterns correlates significantly to the spontaneous variations of the alpha rhythm. This strategy provides insight in the nature of resting state fMRI and reduces the number of statistical tests in the GLM correlation analysis. For all 16 subjects except one, the clustering of BOLD signal yielded very consistent regions which included areas belonging to the default mode network as well as the neuronal networks involved in the generation of the alpha and mu rhythms. These BOLD clusters showed much less inter-subject variability than the Alpha-BOLD statistical parametric maps obtained on a voxel-by-voxel basis. It is shown that hierarchical clustering is applicable to whole head fMRI and that it is very appropriate to obtain data reduction thereby facilitating the comparison of the results of individual subjects. The very consistent results of BOLD clustering over subjects suggests that the large inter-subject variability observed in the Alpha-BOLD statistical parametric maps is related to the individual variations in the EEG.