Abstract
From a neuroscientific perspective, understanding the human capacity to recognize objects requires the identification of where and when the brain processes object information. A wealth of findings shows that a hierarchical system of visual areas in the ventral visual stream enables object recognition. This system is extremely fast, and brain responses about object categories emerge in the range of 100-200ms after stimulus onset. However, most studies have limited analysis of object recognition to the level of everyday categories, leaving open questions about the representation of objects above and below the category level. In this study, we used magnetoencephalography (MEG) and multivariate pattern classification to investigate brain responses to a complex and large image set consisting of 92 images, ordered hierarchically at different levels of generality: the supra-category, category and sub-category level (Kiani et al, 2007; Kriegeskorte et al., 2008). In a first step, we investigated the time course with which information about single images emerged at different levels of generality with high temporal fidelity. We then used representational similarity analysis to link the dynamics in the evolving MEG signal to the hierarchical structure of the ventral visual stream. To this goal, we recorded brain responses to the same image set with fMRI. We showed that early MEG signals (~71ms after stimulus onset) can be linked to fMRI signals in visual area V1, whereas later on (starting at 113ms) MEG signals can be linked to human inferior-temporal cortex (IT). Taken together, our results describe the dynamics of object recognition in the human brain with high temporal fidelity, and link these dynamics quantitatively across imaging modalities to the ventral visual stream.
Meeting abstract presented at VSS 2013