Xiaoli Zhang¹, Taosheng Liu¹; ¹Michigan State University

Much research has shown that spatial attention increases neural activity for attended stimuli. However, whether and how spatial attention modulates neural representations for visual features remains understudied. In this study, we comprehensively characterized attentional modulation using multiple measures of neural representations along the visual hierarchy. Inside an MRI scanner, a dot field in one of four colors was shown on each trial. Participants performed a luminance change discrimination task under two attentional conditions in separate runs: (1) at the dot field (i.e., the Attended condition), (2) at the fixation (i.e., the Unattended condition). We analyzed the blood oxygenation level-dependent (BOLD) activity in brain regions involved in visuospatial attention, including visual areas and frontoparietal regions. To assess the neural representations of color features, we compared a variety of metrics between Attended and Unattended conditions: univariate activity level, representational dimensionality, multivariate pattern classification, and noise correlation (NC) within each region. Both univariate activity and representational dimensionality confirm that attention increases neural signal strength to a greater magnitude in frontoparietal than visual regions. Attention also improves color decoding in prefrontal regions, although not in early visual regions. Interestingly, attention decreases within-region noise correlation in both early visual regions as well as frontoparietal regions. Our results are generally consistent with the model of frontoparietal network controlling spatial attention and reveal interesting connections from multiple measures of neural activity. Our results suggest that top-down spatial attention modulates neural activity along the visual hierarchy not only via increasing overall neural gain, but also by preserving neural population responses from sources of noise. Spatial attention appears to employ multiple mechanisms to enhance brain representations.

Acknowledgements: Funding: NIH R01EY032071