Abstract
Recently, Kamitani & Tong (2006) showed that ensemble patterns of fMRI voxels exhibit robust direction selectivity that allows for accurate prediction of motion direction. Here, we investigated whether ensemble activity patterns contain reliable information about heading direction that requires higher-level of motion processing. For heading stimuli, we used radially expanding optic flow patterns that simulated heading either left or right direction. fMRI signals were measured while subjects viewed optic flow. A linear decoder was trained to classify voxel intensity patterns induced by left or right heading directions, then evaluated with independent test data. The ensemble activity from area MT+ led to good predictions (∼90 %) and the performance was better than other cortical areas. As humans perceive accurate heading direction even when their eyes are rotated, we also tested if the performance of the decoder is generalized to optic flow with additional motion of eye-movement. In eye movement condition, subjects pursued a horizontally moving point in an optic flow display. Testing the decoder on the activity patterns of the eye movement condition, area MT+ showed less robust but still relatively high decoding performance. Subjects also viewed the optic flow display simulating additional motion of the eye-movement while they fixated a stationary point. The decoder was trained with the activity patterns of the simulated display condition and tested on those of executed eye-movement condition. Although the retinal flows were identical between training and test conditions, the decoding performance of area MT+ was worse than the decoder trained with the condition without motion information of the eye-movement. These results suggest that area MT+ is involved in heading perception and coding head-centric motion with compensation of extra-retinal information, as suggested by previous studies. More generally, our study shows that decoding technique can be used to reveal the function of the cortical activity.
This work was supported by Nissan Science Foundation.