Intra-V1 functional networks predict observed stimuli

Abstract

Several studies suggest that the pattern of co-fluctuations of neural activity within V1 (measured with fMRI) changes with variations in attention/perceptual organization of observed stimuli. Here we used multivariate pattern analysis of intra-V1 correlation matrices to predict the level and shape of the observed Navon letters. We examined the inter-individual stability of network topologies and then tested if they contained intra-individual information about stimulus shape or level that was tolerant to changes in the irrelevant feature. The inter-individual classification was accurate for all specific level and letter-shape tests. These results indicate that the association of V1 topologies and perceptual states is stable across participants. Intra-participant cross-classification of level (ignoring shape) was accurate but failed for shape (ignoring level). Cross-classification of stimulus level was more accurate when the stimulus-evoked response was suppressed in the fMRI time series and not present for correlations based on raw time series, stimulus-evoked beta-series, or simulations of the effects of eye movements measured in a control group. Furthermore, cross-classification weight maps evinced asymmetries of link strengths across the visual field that mirrored perceptual asymmetries. We hypothesize that feedback about level information drives the intra-V1 networks based on fMRI background activity. These intra-V1 networks can shed light on the neural basis of attention and perceptual organization.