Neural priority maps encode behavioral relevance independently across multiple attended locations

Abstract

Retinotopic cortical regions encode both the physical salience and behavioral relevance of visual stimuli, suggesting that these areas support neural "priority maps" of the visual field. These maps, which are instantiated as distributed neural codes across visual, parietal, and frontal areas, guide covert attention and overt motor plans. Previous work has characterized how these maps encode the task relevance of a single covertly-attended visual stimulus among multiple stimuli. However, it remains unknown how relevance-related modulations scale with the number of task-relevant locations. If modulations reflect limitations in attention-related enhancements of relevant neural populations, attending to more stimuli should reduce the magnitude of modulations. Instead, if relevance-related modulations reflect the selection of task-relevant populations by long-range inputs to optimize coding for further processing, each relevant stimulus location would be modulated by an equivalent amount, independent of other stimuli. We collected functional magnetic resonance imaging (fMRI) data while participants performed a demanding covert attention task requiring they monitor zero, one, or two peripheral task-relevant stimuli. Using a spatial inverted encoding model, we reconstructed neural priority maps from activation patterns across retinotopic cortex. Strikingly, relevance-related enhancement at each attended location was equivalent whether one or two locations were cued, with no evidence for attenuation when multiple stimuli were relevant. Moreover, task-related modulations were spatially focal and disjoint, localized to each cued stimulus location. These results suggest a model whereby behavioral relevance is encoded categorically in neural priority maps: BOLD signal from neural populations at each task-relevant location are equivalently and independent enhanced regardless of how many other locations are concurrently relevant. This modulatory profile reveals a relevance map that signals which populations require selective augmentation to guide decision-making through the plethora of local neural computations associated with covert attention, with performance limitations arising at later processing stages rather than within early sensory representations.