Decoding Feature-Based Attention in Visual Cortex

FMRI were recorded from participants viewing random dot kinematograms (RDKs) varying in either color or shape. In baseline conditions, participants attended a single feature array (gray circles, red circles, or gray squares) flickered at either 8.57 Hz or 12 Hz. In selection conditions, two superimposed RDKs differing in one feature dimension were flickered at two different frequencies (8.57 Hz vs 12 Hz), and the participant was instructed to attend color or shape. For all conditions, participants indicated whether they noticed a luminance change in the attended RDK array. We found that (1) for the baseline conditions, (a) comparing red circles against gray circles and gray circles against gray squares yielded no regions of significant differential activations and (b) MVPA decoding between red circles vs gray circles and between gray circles vs gray squares revealed above chance decoding accuracy in the entire visual hierarchy, suggesting that color and shape are broadly represented in the visual cortex in the form of linearly discriminable spatial patterns and (2) for the selection conditions, (a) comparing selecting red circles in the presence of gray circles against selecting gray circles in the presence of red circles and comparing selecting gray circles in the presence of gray squares against selecting gray squares in the presence of gray circles yielded no regions of significant differential activations and (b) MVPA decoding between selecting red circles vs selecting gray circles and between selecting gray circles vs selecting gray squares revealed above chance decoding accuracy in the entire visual hierarchy, suggesting that attention selection for color and for shape occurs broadly in the visual cortex in the form of linearly discriminable spatial patterns. Overall, our data support the idea that rather than modulating neural activities only in feature-selective visual areas, FBA modulates neural activity broadly in the visual cortex.