Abstract
Traditional accounts hold that selective attention facilitates perception by increasing the gain of sensory neurons that are maximally responsive to task relevant stimulus features. In contrast, recent theoretical and empirical work suggests that attention strives to maximize performance on the current perceptual task, even if this means applying sensory gain to neurons tuned away from the relevant features (Navalpakkam and Itti, 2007; Scolari and Serences, 2009). For example, when discriminating a 90° oriented line from a set of distractors oriented at 92°, sensory gain should be applied to neurons tuned to flanking orientations because they undergo a larger change in firing rate in response to the target and distractors. Given the high-density of orientation-selective neurons in primary visual cortex, we hypothesized that such “off-channel” gain in V1 should predict performance on a difficult orientation discrimination task. We used fMRI and voxel tuning functions to determine if correct trials were associated with more off-channel sensory gain when compared to incorrect trials. Subjects completed a 2AFC task in which a grating (sample stimulus) was presented at one of 10 possible orientations for 2s, followed by a 400ms delay, and then a second grating which was rotated clockwise or counterclockwise from the sample (the size of the offset was determined on a subject-by-subject basis, and ranged from 1-4.75°). Subjects exhibited a larger BOLD response in V1 in the most informative voxels (e.g., ±36° offset from the sample) on correct trials than on incorrect trials; this pattern was not observed in other visual areas (V2v, V3v, and V4v). These results indicate that performance on demanding perceptual tasks is not always predicted by the gain of the maximally responsive sensory neurons. Instead, the magnitude of gain in the most informative neurons predicts perceptual acuity, even if these neurons are tuned away from the relevant feature.
National Institutes of Health Grant R21-MH083902 (J.T.S.).