Abstract
Past research in macaque neurophysiology has identified mixed-selectivity neurons that respond to combinations of variables, such as task and stimulus type, in a way that is a nonlinear function of the variables considered individually. However, the prevalence of these neurons in human cortex remains largely unexplored. If neurons with mixed selectivity are distributed heterogeneously within a brain region at a spatial scale detectable by fMRI, then we should be able to measure mixed selectivity at the voxel level with fMRI, leveraging the whole-brain coverage that fMRI affords. Here, we tested the feasibility of this approach and compared the magnitude of mixed-selectivity coding across human occipital, parietal and frontal regions. Specifically, we examined data from four experiments in which human participants engaged in several tasks upon exemplars from eight object categories. In three of the experiments, participants performed a oneback task on either the shape or color of the exemplar, varying the conjunction strength between shape and color across experiments. In a fourth experiment, participants performed either a oneback or oddball task on the exemplar. In all four experiments, stimuli were equated between tasks, allowing us to model main effects of task and stimulus category, and their interaction, on voxel responses. Across all four experiments, we found that voxel responses in the frontoparietal regions were significantly better explained by task/stimulus interaction terms than the early visual and ventral regions, reflecting the presence of nonlinear mixed selectivity neurons in frontoparietal regions that enable them to effectively integrate task and stimulus information for flexible behavior. These findings demonstrate the feasibility of using fMRI to examine mixed-selectivity noninvasively in the human brain.
Meeting abstract presented at VSS 2018