Abstract
The map of category-selective regions in the human ventral visual stream provides organizational constraints to visual recognition. While some principles of the organization have been established, such as medial-lateral differentiation by stimuli that vary in their typical retinotopic eccentricity, the relative position of word and face selective areas, both of which are foveated, remains unclear. Some studies suggest that face selective areas tend to be medial to word selective areas, though the response to these stimuli have infrequently been compared to one another within the same subjects. Here we present evidence that challenges this organization from invasive neural recordings from 30 patients with intractable epilepsy using a combination of multivariate decoding accuracy, broadband gamma response, and event related potential amplitude. In these patients we found strong category-specific responses in the ventral visual stream, with house and tool specific responses distributing more medially than faces and words, consistent with previous studies suggesting a medial-lateral organization by typical retinotopic eccentricity. However, we found no medial-lateral bias for words and face selective areas at a group level. In fact, in some individual subjects, word specificity was medial to face specificity, and in others strongly face and word specific patches alternated (e.g. patches alternated face-word-face or word-face-word in the lateral-medial direction). We compare these results with similar category mapping on participants with ultra-high-resolution 7 Tesla MRI to determine if similar interdigitation of word and face selective voxels exists when looking at a fine scale using fMRI. Our results suggest an organization of the ventral stream wherein word sensitive patches and face sensitive patches, two classes of visual objects that are foveated, are finely interdigitated with little-to-no medial-lateral biases relative to each other.
Meeting abstract presented at VSS 2018