Abstract
Ultra-high field fMRI provides the unique opportunity to record brain activation at sub-millimeter resolution. It is thus possible to reconstruct different cortical depth layers within primary and early visual cortices. Here we investigated how neural activity transfers across different cortical depth layers in V1. Specifically we ask whether thalamo-cortical, and cortico-cortical triggered responses lead to different inter-laminar communication patterns. To address this issue we used previously recorded data (Muckli et al., 2014), that were acquired to investigate laminar sensitivity to cortico-cortical contextual information. Using a visual occlusion paradigm these data showed that in the absence of thalamic input to V1, outermost layers hold meaningful contextual information about the immediate visual environment.The current data set encloses the BOLD signal of 4 participants while they were viewing images of natural scenes. To isolate non-thalamic-related activity, retinal input to a selected portion of V1 was blocked by occluding the bottom right quadrant of the images. Our analysis focused on a subpopulation of V1 voxels obtained by retinotopically mapping the cortical representation of the occluded area. We used two simple linear encoding models to assess the preferential tuning of individual voxels to contextual information. This approach provided additional evidence that outermost layers exhibit the largest concentration of voxels sensitive to contextual information. We also iteratively trained support vector machines (SVM) on each layer and tested on the remaining 5, independently for feed-back (i.e. neural activity triggered by occluded images) and feed-forward (i.e. neural activity triggered by non-occluded images) signals. We found feed-forward cross-layer SVM accuracy was highest in the mid layers; and feed-back cross-layer SVM accuracy was highest in the outermost layers. In line with animal models, these results suggest that mid-layer communication is prominent during feed-forward processing, while outer layers interact mostly during feed-back.
Meeting abstract presented at VSS 2015