Abstract
Functional magnetic resonance imaging (fMRI) studies have yielded significant insights about the nature of representations across the ventral visual pathway. Amongst them is the finding of category-selective regions that respond, for example, more strongly to presentations of faces than objects. These types of comparisons, while powerful, are fundamentally limited by the need to average across presumed categories. This method yields contrasts that reflect differences in the central tendencies of the response to all of the individual exemplars with no direct measure of the actual amount of variance accounted for by the categories. Here we take advantage of the increased signal available at 7 Tesla to investigate the response to each of 768 individual stimuli presented a single time. The stimuli were drawn randomly from a commercial database, constrained only to span the entire range of image types within the database. Each stimulus was presented a single time in an event-related design. We also collected standard localizers in separate runs and identified face-, body-, word-, scene-, and object-selective regions in each participant. Even with only a single event-related trial for each image we were able to recover selectivity within these predefined regions, with individual images that contained the preferred stimulus evidencing both strong activity and grouping in the multivariate response. There was also strong correlation in the rank ordering of responses across the individual images (selectivity profile) within particular regions between participants. The design also afforded us the flexibility to compare the selectivity profiles of individual voxels across the brain in a manner analogous to traditional functional connectivity analyses. This analysis revealed both local regions and networks that shared selectivity without the need to resort to any contrast. We conclude that there is enough signal in fMRI to investigate the response to single trials, enabling far more powerful and flexible designs.
Meeting abstract presented at VSS 2012