Abstract
Responses to binocular disparity are widespread throughout the visual, temporal and parietal cortices. However, the circuits that transform local binocular disparities to coherent three-dimensional (3D) representations remain unknown. Here we use human fMRI to test for cortical areas that represent global 3D surfaces as opposed to local disparities. We measured fMRI responses while participants viewed random dot stereograms depicting slanted planes (±7.5 to 52.5 deg in 15 deg steps). Using multi-voxel pattern classification analysis, we find that (i) slant can be decoded reliably across visual areas and (ii) there is pronounced fMRI pattern-tuning for slant in early visual (V1, V2) and dorsal (V3, V3A) areas. To determine whether this fMRI pattern information relates to surface slant per se or rather lower-level features (e.g. spatial extent or mean disparity of the stimuli), we measured responses when: (a) the spatial extent of the planes was fixed as slant was varied; (b) the mean disparity was fixed as slant was varied. We quantified the similarity between fMRI responses across conditions using a cross-correlation analysis. We find that fMRI responses in area V3A are highly similar when the same slant is depicted, irrespective of changes in spatial extent or overall disparity of the stimuli. However, this is not true in areas V1 and V2. Moreover, opposite slants evoke highly dissimilar responses in area V3A, while there is some similarity between responses in V1. Finally, training the classifier on data from the main experiment and testing on the control conditions (transfer test) demonstrates reduced decoding performance in V1 and V2. Together these results suggest V1 and V2 responses relate to low-level features (disparity-defined edges and distribution of disparities) while responses in V3A are compatible with the representation of surface slant.
JSPS Postdoctral Fellowships for Research Abroad (290).