Divided attention for feature-based selection in visual cortex

Abstract Feature-based attention can select relevant features such as colors or directions of motion from the visual field irrespective of the spatial position. In visual cortex not only do we see feature-specific attention affecting responses in neurons with receptive fields at an attended location, but that effect also spreads to neurons with receptive fields beyond the spatially relevant location. When only one feature is task relevant, the spread of activity across space can act to facilitate perception of behaviorally relevant stimuli. However, when multiple features are relevant, what is the effect on behavior and brain activity? We tested this question by having observers divide attention between two patches (left and right of fixation) of moving dot stimuli, each containing overlapping upward and downward motion fields. In one condition, subjects performed a task on motion fields moving in the same direction in both patches. In another condition attention was divided between opposite directions of motion. Replicating a previous behavioral study, we found that observers showed better performance when dividing attention to the same directions of motion than opposing directions of motion. We analyzed the BOLD responses while observers performed this task using an inverted encoding model approach that provides estimates of responses to each of the four component dot fields. We found larger responses to the attended components, replicating previous studies of spatial and feature-based attention. However, these effects were much larger when attention was divided to between the same directions of motion than to opposing directions. Our fMRI results in area hMT+ predict our behavioral results by extending the normalization model of attention to include a global feature-based attention component in that leads to suppressed responses to attended stimulus components when attention is directed to opposing directions of motion.