Abstract
Recent fMRI studies in several labs have produced mixed results in measuring the response of V1 to manipulations of global phase coherence in natural images. We have performed a series of experiments that separately examine the role of feature density and spatial frequency in determining V1 BOLD fMRI sensitivity to the phase structure of images. Stimuli were formed in two different spatial frequency bands by placing randomly oriented Gabor patches at three different densities in images; each of these coherent images was contrasted against its phase-scrambled counterpart, equivalent to band-pass noise. A computational model was built to predict the spatial distribution of neural activity and resulting BOLD fMRI response. Modeled neural activity was compared against neural activity inferred from psychophysical measurements of contrast detection and discrimination thresholds; both indicated higher local activity for stimuli with phase coherence. The principle finding is that BOLD fMRI response, when averaged across large regions of interest, is lower when neural activity is sparse. This is the case even when the low-density coherent images have lower detection or discrimination thresholds (i.e. stronger local neural activity), demonstrating that V1 BOLD fMRI results can be misleading when spatial distribution of image features is not taken into account. Both measured and modeled BOLD fMRI responses also produce the following results: 1) sensitivity to phase coherence depends on spatial frequency; 2) population activity is frequently mixed (some voxels showing preference for coherent images; some responding more strongly to scrambled images). Therefore, careful analysis of individual voxel responses (rather than ROI averages) is required for accurate application of BOLD imaging to studies of spatial vision. This work also illustrates that local phase manipulations, rather than global phase manipulations, are more appropriate for the study of V1 responses.
NSF/IGERT DGE 9870633, University of Minnesota Graduate School Doctoral Dissertation Fellowship, R01 EB00331, P41 RR08079, the MIND Institute, and the WM KECK Foundation