The Functional Architecture of Noise Correlation in fMRI Responses from Human Visual Cortex

Abstract

When an identical stimulus is presented repeatedly, the activity of sensory cortical neurons varies from trial to trial, dubbed ‘neuronal noise’. Recent electrophysiological and imaging studies reported that the ‘noise’ is not just a random and independent deviation from signal and reflects correlated activity among local cortical sites. Here we investigated the structure of correlated ‘noises’ in early human visual areas by monitoring moment-to-moment fluctuations in fMRI responses to visual stimuli. By defining receptive fields and stimulus preferences of individual voxels, we could reveal a reliable functional architecture of noise correlation: noise correlation was high in pairs of voxels whose stimulus preferences are similar and whose receptive fields are close to each other. The analysis of residual correlation confirmed that this functionally defined structure of noise correlation could not be explained by trivial correlations due to anatomical proximity. The spectral analysis of time series revealed that the stimulus-preference-dependent correlation was maximal at a low (<0.035Hz) band of temporal frequency whereas the receptive field-dependent correlation was maximal at a medium (0.035∼0.082Hz) band. Furthermore, the functional structure of noise correlation was held true for voxel pairs within and between different visual areas, regardless of the presence or types of visual stimulation.