How bottom-up and top-down factors shape representation in word- and face-selective cortex
32.17, Sunday, 17-May, 10:45 am - 12:30 pm, Talk Room 1
Kendrick Kay1, Jason Yeatman2; 1Department of Psychology, Washington University in St. Louis, 2Institute for Learning and Brain Sciences, University of Washington
Specific regions of ventral occipitotemporal cortex (VOT) appear to be specialized for the representation of certain visual categories: for example, the visual word form area (VWFA) for words and the fusiform face area (FFA) for faces. However, a computational understanding of how these regions process visual inputs is lacking. Modeling these regions is particularly challenging because responses depend on both bottom-up stimulus properties and top-down cognitive processes. To develop a model of VOT, we measured BOLD responses in VWFA and FFA to a wide range of carefully controlled grayscale images while subjects performed different behavioral tasks. During a demanding fixation task (judge the color of a small central dot), responses in VWFA and FFA were not strictly categorical but were sensitive to low-level properties such as image contrast and phase coherence. This suggests that category-selective regions inherit response properties present in early visual cortex. Combined with previous work demonstrating systematic receptive fields in VOT (Kay et al., VSS 2014), we suggest that a cascade model of visual processing may be able to explain a substantial component of VOT responses. During a categorization task (judge the category of the stimulus), responses in VWFA and FFA were substantially stronger than responses observed under the fixation task. Importantly, the categorization responses were not a simple linear rescaling of the fixation responses. Rather, the enhancement was disproportionately large for low-contrast stimuli, and the amount of enhancement observed for a stimulus was highly correlated with the reaction time measured for that stimulus. These results indicate that when stimuli are behaviorally relevant, top-down influences can selectively enhance VOT responses to weak stimuli, effectively increasing the amount of information conveyed by VOT. In sum, our results lay the groundwork for the development of models that quantitatively predict VOT responses.