Inferring population responses in human visual cortex with classification analysis

Abstract

How the joint activity of many neurons with different selectivities represent visual stimuli is of key importance to understanding visual perception and visually guided behavior, yet most tools for probing these response properties in human visual cortex have been severely limited due to their coarse spatial-resolution. Conventional sized functional magnetic resonance imaging (fMRI) voxels (3×3×3 mm) are thought to encompass many dozens of cortical columns containing neurons with different specificities for basic visual stimulus properties like orientation and direction. However classification techniques have successfully shown that by combining responses across many voxels with small, noisy biases for orientation and direction, the stimulus a subject is viewing can be correctly classified. We used high-spatial resolution fMRI and classification analysis, and found that robust orientation and direction selectivity can be observed in large draining veins that would be expected to drain over an area of cortex comprising many cortical columns. Thus the success of classification techniques in fMRI may be based on sampling from easily measurable signals from large draining veins rather than biased sampling of well localized signals from cortical maps. Large veins may have specificity because they amplify a biased representation inherent to the cortical architecture or alternatively, because they might be organized to drain specifically from cortical columns that are functionally active together. Either way, our results suggest that inferring population response in human visual cortex can be made possible by measuring valid, if indirect, signals represented in large draining veins.

Footnotes

 JLG was supported by a Career Award in the Biomedical Sciences from the Burroughs-Wellcome Fund. P.S. was supported in part by a postdoctoral fellowship from the Japan Society for the Promotion of Science.