Abstract
Behavioral data suggest that visual scene categorization draws heavily upon analysis of scene spatial properties, such as three-dimensional layout (Greene & Oliva, 2009). At the same time, the perceived category of a scene is strongly influenced by the kinds of objects scenes contain (Davenport & Potter, 2004; Joubert et al., 2007; MacEvoy and Epstein 2011). We recently used fMRI (Linsley & MacEvoy, VSS 2013) to demonstrate that these two routes to scene recognition converge, at least partially, in parahippocampal place area (PPA), an area of ventral-temporal cortex previously shown to be sensitive to scenes' spatial properties. Along PPA pattern dimensions yoked to scene spatial properties, scenes possessing extreme spatial properties were encoded as more similar to their category average when category-informative objects were visible versus masked. This "centripetal" bias may improve scene recognition accuracy by bringing scenes' encoded spatial properties into register with those expected from their object contents. In the present study, we applied a novel information-based functional connectivity analysis to identify brain regions participating in the generation of centripetal bias. Cortical volumes collected during presentation of extremely small and large bathrooms, both with informative objects visible and masked, were passed to an iterative whole-brain MVPA searchlight procedure to identify voxel clusters containing information about object-masking state. For each such cluster, we asked how well trial-by-trial scores extracted along pattern dimensions corresponding to scenes' object-masking state explained trial-wise variability of PPA centripetal bias. Similar to previous studies of PPA functional connectivity, our analysis revealed participation by clusters in the visual system and default-mode network (Baldassano et al., 2013). Among visual areas, the greatest contribution to PPA was made by clusters within lateral occipital complex (LOC), a region linked to object processing. These results reveal a functional network supporting crosstalk between object- and spatial property based routes to scene categorization.
Meeting abstract presented at VSS 2014