Cortical representation of animate and inanimate objects in complex natural scenes

The representations of animate and inanimate objects appear to be anatomically and functionally dissociated in the primate brain. How much of the variation in object-category tuning across cortical locations can be explained in terms of the animate/inanimate distinction? How is the distinction between animate and inanimate reflected in the arrangement of object representations along the cortical surface? To investigate these issues we recorded BOLD activity in visual cortex while subjects viewed streams of natural scenes. We then constructed an explicit model of object-category tuning for each voxel along the cortical surface. We verified that these models accurately predict responses to novel scenes for voxels located in anterior visual areas, and that they can be used to accurately decode multiple objects simultaneously from novel scenes. Finally, we used principal components analysis to characterize the variation in object-category tuning across voxels. Remarkably, we found that the first principal component reflects the distinction between animate and inanimate objects. This dimension accounts for between 50 and 60% of the total variation in object-category tuning across voxels in anterior visual areas. The importance of the animate–inanimate distinction is further reflected in the arrangement of voxels on the cortical surface: voxels that prefer animate objects tend to be located anterior to retinotopic visual areas and are flanked by voxels that prefer inanimate objects. Our explicit model of object-category tuning thus explains the anatomical and functional dissociation of animate and inanimate objects.

► A model of object-category tuning predicts BOLD activity evoked by natural scenes.
► The model is used to decode multiple objects in natural scenes.
► The model reveals the sources of variation in object-category tuning across cortex.
► The primary source of variation is preference for animate or inanimate objects.