Purchase this article with an account.
Yalda Mohsenzadeh, Caitlin Mullin, Benjamin Lahner, Aude Oliva; Neural Dynamics of Category Representations Across Space and Time in the Ventral Visual Cortex. Journal of Vision 2019;19(10):60b. doi: https://doi.org/10.1167/19.10.60b.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Previous neuroscience works have established a functional organization across ventral visual stream such that distinct categories are processed in different brain areas; e.g. the parahippocampal cortex (PHC) for places, the object-selective lateral occipital complex (LOC), the fusiform gyrus for faces and animates. However, it is still an open question how categorical representation in these specialized brain regions unfold over time. Here we extended the approach of MEG-fMRI fusion from Cichy et al. (2014,2016) to reveal the category-related spatiotemporal neural dynamics of vision in the human brain. We collected fMRI and MEG data while participants (N=15) viewed 156 natural images in four categories, namely objects, scenes, faces and animates (people and animals). We created theoretical model representational dissimilarity matrices (RDMs) representing category membership—higher similarity within a category. Using representational similarity analysis, per category content, we correlated spatially-resolved fMRI RDMs with temporally-resolved MEG RDMs while partialling out the other three category models. Then we localized the category-related neural activity in fMRI data by conducting a searchlight analysis, comparing each category model RDM with searchlight fMRI RDMs. Finally, we filtered the partial fMRI MEG representational correlations with the spatially localized category contents resulting in a spatiotemporal map per category while controlled for other categories. Scene and Object category processing started at ~80ms in early visual cortex and reached medially to PHC at ~110ms and laterally to LOC at ~100ms, respectively, both with a peak at ~135ms. Body category content emerged in LOC at ~90ms and extended to Fusiform at ~110ms. Fusiform showed significant face category content processing starting at 70ms reaching a peak at ~140ms. Together, our results revealed the spatiotemporal dynamics of category representations in the ventral stream at the millimeter and millisecond resolution.
This PDF is available to Subscribers Only