Statistical properties of BOLD magnetic resonance activity in the human brain.

We investigated the random variability of BOLD (blood oxygen level dependent) activation during rest, or null-hypothesis, conditions in which the observers were neither receiving controlled sensory stimuli nor performing cognitive tasks. The data indicate that the distributions for the BOLD variation across space are skewed, with non-Gaussian tails, while the distributions for the temporal variation within individual voxels are predominantly Gaussian. The proportion of voxels that show non-Gaussian properties is highly correlated with the magnitude of head movement of the observers. In all observers, the white matter showed less variability than the gray matter. The distributions for the spatial and the temporal variations are robust across observers despite differences in the data acquisition methods (EPI vs. spiral) and magnetic field strength (1.5 vs. 3 T). In most cases, the non-Gaussian tails of the spatial distribution can be eliminated by normalizing the amplitude in each voxel to its standard deviation before cumulating across voxels. We therefore recommend such a normalization procedure before any data manipulations are performed on fMRI data.