Abstract
High-level perceptual functions depend on complex interactions within networks of brain regions. However, the typical approach in high-level vision is to localize brain regions that respond selectively to particular categories of images and to probe response properties focusing on these specific regions. Here we show that to understand the computations performed by high-level visual regions, it is essential to analyze the connectivity of the region and model responses in the context of the larger network in which the region resides. We used fMRI to localize word-selective (VWFA-1, VWFA-2) and face-selective (FFA-1, FFA-2) regions in ventral temporal cortex (VTC). We then measured responses in these regions to a wide range of carefully controlled grayscale images while subjects performed different tasks that isolate bottom-up, stimulus-driven responses from top-down, task-dependent modulations. Our measurements demonstrate that VTC is not only sensitive to stimulus properties (e.g. image category) but is also substantially affected by the task performed on the image. Importantly, we find that the observed task sensitivity is predictable based on the pattern of white-matter connections between each region and the rest of the brain. VWFA-1 and FFA-1 are directly connected with the intraparietal sulcus (IPS) through the vertical occipital fasciculus (VOF), and we find that task-related response modulation in VWFA-1 and FFA-1 can be predicted with a model that incorporates the evoked response in IPS to each stimulus. In contrast, VWFA-2 is directly connected with Broca's area through the arcuate fasciculus, and we find that this connection imparts substantial word selectivity when subjects perform tasks involving linguistic analysis of visually presented words. These results show that anatomical connections between VTC and parietal and frontal cortex support functional interactions that fundamentally shape response properties in VTC.
Meeting abstract presented at VSS 2016