Spatially-specific feature tuning drives response properties of macaque IT cortex

Neurons in primate visual cortex respond to stimuli within local regions of visual space, i.e. receptive fields. Traditionally, receptive fields are mapped by characterizing neuronal or population activity in response to low-level, high-contrast stimuli, such as bars or gratings, presented at various locations. This approach has proven effective for mapping receptive fields in early visual cortex; however, it is notably more difficult to map receptive fields in high-level regions of visual cortex, such as inferior temporal (IT) cortex, due to the complex stimulus selectivity exhibited by neurons in these regions. Here, to determine whether there is an interaction between the spatial and featural selectivity of IT neurons, we measured multiunit electrophysiological responses in fMRI-defined face-selective patches of IT cortex of rhesus macaques. Monkeys fixated while viewing a diverse set of naturalistic images, including faces, face parts, hands, objects, and scenes, in a grid of locations, spaced apart by 1-2 visual degrees, spanning the central 17 degrees around fixation. We modeled IT receptive field structure as a two-dimensional Gaussian with compressive spatial summation (Dumoulin & Wandell, 2008; Kay et al., 2013) and were able to estimate reliable receptive field positions for face-selective units. We observed apparent shifts in receptive field position depending on stimulus type: specifically, a vertical shift for inverted compared to upright faces, consistent with prior human fMRI results (Poltoratski et al., 2021). However, this apparent shift can be explained by the spatially non-homogeneous contribution of face parts/features to the model responses, which we determined by modeling receptive field position as a function of eye or mouth position. These preliminary findings, that receptive fields of face cells in IT exhibit spatially specific feature tuning for face parts, indicate the need to jointly consider both spatial and featural selectivity when mapping receptive fields in high-level visual cortex.