Abstract
Neural responses in visual cortex can be elicited by viewing a scene (perception), maintaining a recent percept of that scene (working memory), or retrieving a scene viewed in the past (long-term memory). fMRI studies have shown that the three types of representations share some properties; for example, both forms of memory, like perception, can be retinotopically tuned. However, until very recently there has been little investigation of systematic differences between visual cortex responses during memory and those during perception (with the exception of signal-to-noise ratios, which are consistently shown to be lower during memory than perception). We used fMRI to quantify spatial tuning in visual cortex during perception, working memory, and long-term memory. Each trial contained a 11.5-second target period during which a small, peripheral drifting Gabor target was either (1) viewed, (2) maintained in working memory following a brief pre-trial exposure, or (3) retrieved from long-term memory (which required pre-scan associative learning with cues shown at fixation). For all three conditions, participants maintained central fixation during the 11.5 seconds and then made a saccade to the target location. Targets in each trial were sampled from one of sixteen possible polar angles, always at 7ยบ eccentricity. In all three conditions, we find that BOLD responses in the V1 to hV4 maps during the target period were spatially tuned to the stimulus polar angle, confirming retinotopic tuning in memory. As expected, during perception the polar angle tuning became increasingly wide from V1 to hV4, reflecting larger receptive fields. In contrast, during both working memory and long-term memory, tuning width was nearly the same across the four maps, approximately matched to perception in hV4 but wider than perception in V1-V3. The results demonstrate systematic differences in how sensory cortex is used for both working and long-term memory compared to perception.