Systematic effects of retinotopic biases and category selectivity across human occipitotemporal cortex

Abstract

The organization of human visual cortex has traditionally been studied using two different methods: retinotopic mapping and category-selectivity mapping. Retinotopic mapping has identified a large number of systematic maps of the visual field, while category-selectivity mapping has identified clusters of neural populations that reliably respond more strongly to specific image categories such as objects, faces, scenes and body parts compared to other categories. While early investigations seemed to suggest that these two organizing principles were largely separated in the brain, with retinotopic maps in posterior visual cortex and category-selective regions in anterior visual cortex, recent work shows that category-selective regions overlap with retinotopic maps, giving rise to spatial visual field biases within these regions. Here, we collected fMRI responses whilst performing both retinotopic and category mapping within the same participants, allowing detailed comparison of neural tuning for space and category at the single voxel level. We use these data to evaluate two previous proposals of how retinotopic biases relate to category-selectivity: 1) complementary quadrant biases (upper vs. lower contralateral visual field) inherited from early visual cortex explain the presence of paired regions selective for the same category across lateral and ventral occipitotemporal cortex (lOTC, vOTC); and 2) eccentricity biases (center vs. periphery of the visual field) explain the presence of selectivity for different categories, specifically differentiating face-versus scene-selectivity within the ventral surface. Confirming and extending previous findings for a comprehensive set of face-, scene-, object, and place-selective regions of interest, we provide robust evidence that category-selective regions do not sample visual space uniformly, exhibiting systematic biases towards either the upper or lower field (all category regions) and center vs. periphery (face vs. place regions). Consistent with 1), we find that quadrant biases differ systematically between lateral and ventral OTC, with lateral regions showing systematic lower field biases and ventral regions showing upper field biases, differentiating regions selective for the same category in terms of their spatial bias. However, contrary to 2), we find that eccentricity tuning does not strongly predict the strength of face-or scene category-selectivity in a given voxel. Specifically, highly face-selective voxels are not solely confined to the fovea, and while most scene-selective voxels show peripheral tuning, highly scene-selective voxels actually show strong foveal tuning, particularly in anterior medial-ventral cortex. Collectively, these results demonstrate that spatial biases in category-selective cortex are widespread and robust, whilst also suggesting there is no simple relation between spatial tuning and category-selectivity.