A neural pooling rule for attentional selection in human visual cortex

Abstract

To characterize sensory and decisional processes enabling attention to enhance behavioral performance, human observers performed contrast discrimination judgments following two types of attentional cues, while cortical activity was measured with fMRI. Four sinusoidal gratings were presented, one in each visual quadrant, at 8 different “pedestal” contrasts. Stimuli were shown in two 600-ms intervals separated by a 200-ms blank interval, one of which (randomized across trials) had a near threshold contrast increase across the 2 intervals. After stimulus offset, an arrow at fixation indicated the target location. Observers maintained central fixation and pressed 1 of 2 buttons to indicate the interval with higher contrast. The three non-target locations had different contrasts that remained unchanged across intervals. Half the trials were preceded by a focal attention cue (arrow at fixation indicating target location), and half were preceded by a distributed cue (4 arrows indicating four possible target locations). fMRI response amplitudes were measured in each of several visual cortical areas, separately for each visual quadrant, pedestal contrast, and attentional condition and then combined across quadrants.

Robust increases in fMRI responses and behavioral performance improvements were observed with focal versus distributed attention cues. The changes in fMRI responses could account for the improved behavioral performance only assuming that focal cues caused a 4-fold noise reduction. Whereas sensory noise reduction could account for part of this effect, neither our data nor previous studies support the full 4x reduction we found. Rather, the data were well-fit by a model in which most of the ostensible noise reduction was attributed to the selection and pooling of sensory signals into a decision, utilizing a max-pooling decision rule. We conclude that increases in neural activity with attention in early visual cortex enhance performance by selecting relevant sensory signals for decision.

Footnotes

 5T32-MH05174 & Fellowship by The Italian Academy for Advanced Studies in America to FP R01-MH69880 to DH R01-EY016200 to MC Burroughs Wellcome Fund Career Award in Biomedical Sciences to JG.