Abstract
Researchers use functional magnetic resonance imaging (fMRI) to localize functional regions of interest (fROI) in ventral temporal cortex (VTC) that preferentially respond to particular categories such as faces, bodies, places, and words. In many cases, this fMRI localizer cannot be obtained, for example, in patient populations. The goal of this project is to generate a probabilistic functional atlas that can be used to predict the location of fROIs in people's brains based on anatomy alone. Twelve subjects were scanned with fMRI using a localizer showing images of characters, bodies, faces, places, and objects (Stigliani et al., 2015). From these data, six fROIs in VTC from each subject were identified: mFus-faces/FFA-2, pFus-faces/FFA-1, CoS-places/PPA, OTS-bodies/FBA, pOTS-characters/VWFA-1, and mOTS-characters/VWFA-2 based on significantly higher fMRI activation for preferred vs. non-preferred stimuli. fROIs were projected to each subject's cortical surface, cortex-based aligned to a common cortical surface of the FreeSurfer average template, and then density maps were created, reflecting the percentage overlapping subjects for each fROI at each vertex of the average template. Probabilistic ROIs (pROIs) were created by thresholding these maps at 33% to remove outliers and then eliminating vertices shared by multiple fROIs (Figure 1). To validate the spatial predictability of pROIs, we implemented a leave-one-out cross-validation procedure. Our results show that place-selective fROIs are consistently predicted (78%±2%) in independent subjects and more so than face-, body- and character-selective fROIs (main effect of ROI, F(5,64)=20.2; p< 10-6). Thus, predictability is best for fROIs that have the most consistent localization relative to cortical folding (Weiner et al. 2014; Grill-Spector & Weiner 2014). We are evaluating additional methods such as hyperalignment (Haxby et al., 2011) that may improve the predictability of this functional atlas.
Meeting abstract presented at VSS 2016