Abstract
Although it is generally assumed that the human visual cortex has motion-direction-selective units analogous to those found in animals, the neurophysiological evidence is scarce. Functional neuroimaging has revealed motion-sensitive areas in humans by comparing responses to moving and static stimuli (or dynamic noise), but is thought to lack the resolution to probe into the selectivity to particular motion directions. Here we show that ensemble patterns of fMRI voxels in human visual areas (V1–V4 and MT+) exhibit robust direction selectivity that allows for accurate prediction of motion direction, when information from weakly direction-selective voxels is combined with optimal weights. We performed conventional fMRI scans (3T scanner, voxel size 3x3x3mm) while subjects viewed random dots moving in 1 of 8 directions in each 16s block. A linear decoder was trained to classify voxel intensity patterns induced by different motion directions by optimizing the voxel weights using linear support vector machines. Then, the decoder was evaluated with independent test data. Using 800 voxels from areas V1–V4, the decoder produced predictions peaking sharply at the correct direction (RMSE of 4 subjects, 64 deg). Area MT+, in which fewer voxels were available (∼100 voxels), showed a prediction performance similar to those obtained using the same number of voxels from each of the areas V1–V4. This contrasts our findings from separate studies, in which area MT+ showed markedly poorer orientation selectivity than areas V1–V4 (Kamitani & Tong, VSS/SFN 2004). Our results demonstrate that human visual cortical activity is indeed selective for different motion direction. More generally, the multi-voxel decoding analysis provides a powerful new method to characterize feature selectivity in specific areas of the human brain, and can provide an important bridge between animal and human neurophysiology.
Supported by JSPS, NICT (YK), and NIH (R01-EY14202, P50-MH62196; FT).