Kanwisher Lab

The Group-Constrained Subject-Specific (GSS) method is designed to algorithmically discover functional regions of interest (fROIs) that are activated systematically across subjects. This method is similar to a random effects analysis, but is more tolerant of anatomical variability across subjects. Detailed information about this how method can be used to define ventral visual stream fROIs can be found here. For information about using this method for language localization, see Fedorenko et al. (2010) and this website.



The key steps of the GSS method for defining individual subject fROIs are as follows:
1. We overlaid individual subject's thresholded (p < 0.0001) activation maps on top of one another in a common space, creating probabilistic overlap maps. Each voxel in the overlap maps contains information about the number of subjects that show a significant effect in that voxel. We did this for four contrasts of interest (faces>objects, scenes>objects, bodies>objects, and objects>scrambled). For a faces > objects contrast, for example, the overlap map looks as follows, where the heat of each voxel corresponds to the percentage of subjects that have activation at that voxel:
2. Using a watershed image segmentation algorithm, the overlap maps were divided into functional "parcels" following the map's topography. We then considered all parcels in which ≥ 60% of subjects show the relevant activation. For faces > objects here are our parcels:
3. To use these parcels as spatial constraints to algorithmically select a fROI for each region in each subject, just intersect each parcel (black outlines) with each individual subjects' corresponding thresholded activation map. For example, for faces > objects in one subject we see:
In particular, the subject-specific fROIs were defined as the activation that falls within the boundaries of each parcel.


  • The SPM toolbox necessary to perform GSS analyses, along with instructions and compatibility information, can be downloaded here.

  • The parcels derived in Julian et al. (2012) that identify the major face-, body-, scene-, and object-selective regions in the ventral visual pathway (Kanwisher, 2010) in a large group of subjects (n=30) can be found here (click to download):
    The parcels are in the *.img format and can be viewed with any standard visualization software, like xjview.

  • Also available for download are ventral visual stream parcels defined using an improved functional analysis stream (details coming soon), and a better registration algorithm (CVS registration) that takes into account both cortical volume and surface based information. Because of this improved normalization, these parcels more accurately reflect the distribution of category-selective activation within a large group of subjects (n=40). Click to download (last updated 2/14/2012):
    These parcels are in NIfTI format and can be viewed on the CVS brain template (cvs_avg35) included with recent versions of FreeSurfer.

    You can use the parcels available here with corresponding localizer activation maps in your subjects to algorithmically define your individual-subject fROIs in a similar fashion. For the parcels derived in Julian et al. (2012), this can be performed entirely using the SPM toolbox. Please cite Julian, J.B., Fedorenko, E., Webster, J., & Kanwisher, N. (2012) when using any of the parcels provided here.