Decoded Visual Stimulus Studies Using Functional Magnetic Resonance Imaging

Abstract

In this paper, data from Kay experiment were used to explore how different brain regions involved in visual processing encode different types of visual stimuli, and whether this process has similarities to principal component analysis (PCA). By correlating brain activity with projections on principal components (PCS) of images extracted from the Kay image dataset, V1 and V2 voxel activities were found to have the highest average correlations with image projections on the top 10 PCS. Meanwhile, V2 processes "animal" images, a set of photos of animals, such as a giraffe, that most closely resemble PCA across all brain regions. Next, the brain voxels were correlated with the vectors of each image label, showing an increasing correlation between brain activity and the semantic information of the corresponding visual stimuli. This trend was further demonstrated on "artifact" images, with the exception of V4, after the same correlations were calculated based on image categories.