Visual representations supporting category-specific information about visual objects in the brain

Abstract

Over recent years, multivariate pattern analysis (“decoding”) approaches have become increasingly used to investigate “when” and “where” our brains conduct meaningful processes about their visual environments. Studies using time-resolved decoding of M/EEG patterns have described numerous processes such as object/face familiarity and the emergence of basic-to-abstract category information. Surprisingly, no study has, to our knowledge, revealed “what” (i.e. the actual visual information that) our brain uses while these computations are examined by decoding algorithms. Here, we revealed the time course at which our brain extracts realistic category-specific information about visual objects (i.e. emotion-type & gender information from faces) with time-resolved decoding of high-density EEG patterns, as well as carefully controlled tasks and visual stimulation. Then, we derived temporal generalization matrices and showed that category-specific information is 1) first diffused across brain areas (250 to 350 ms) and 2) encoded under a stable neural pattern that suggests evidence accumulation (350 to 650 ms after face onset). Finally, we bridged time-resolved decoding with psychophysics and revealed the specific visual information (spatial frequency, feature position & orientation information) that support these brain computations. Doing so, we uncovered interconnected dynamics between visual features, and the accumulation and diffusion of category-specific information in the brain.