Elsevier

NeuroImage

Volume 60, Issue 4, 1 May 2012, Pages 2357-2364
NeuroImage

An algorithmic method for functionally defining regions of interest in the ventral visual pathway

https://doi.org/10.1016/j.neuroimage.2012.02.055Get rights and content

Abstract

In a widely used functional magnetic resonance imaging (fMRI) data analysis method, functional regions of interest (fROIs) are handpicked in each participant using macroanatomic landmarks as guides, and the response of these regions to new conditions is then measured. A key limitation of this standard handpicked fROI method is the subjectivity of decisions about which clusters of activated voxels should be treated as the particular fROI in question in each subject. Here we apply the Group-Constrained Subject-Specific (GSS) method for defining fROIs, recently developed for identifying language fROIs (Fedorenko et al., 2010), to algorithmically identify fourteen well-studied category-selective regions of the ventral visual pathway (Kanwisher, 2010). We show that this method retains the benefit of defining fROIs in individual subjects without the subjectivity inherent in the traditional handpicked fROI approach. The tools necessary for using this method are available on our website (http://web.mit.edu/bcs/nklab/GSS.shtml).

Introduction

In a common approach to analyzing functional magnetic resonance imaging (fMRI) data, functional regions of interest (fROIs) are defined independently in each participant, and those regions are then probed further to determine their precise function. Despite the many advantages of this approach (Saxe et al., 2006a), a key limitation is the subjective nature of the choice of which activation cluster should count as the fROI in question in each subject. Such decisions are often made by human data coders guided by macroanatomical landmarks (e.g., gyri and sulci) and stereotaxic coordinates from published studies. However, because of variability across individuals in fROI locations and the lack of a clear mapping between function and cortical structure, these constraints do not always provide clear and unique solutions (Nieto-Castañon et al., 2003). For example, even for well-characterized functional regions like the Fusiform Face Area (FFA), expert data coders may sometimes disagree about whether a given cluster of face-selective voxels constitutes the FFA or the more posterior occipital face area (OFA), or whether the FFA should include two nearby but not contiguous clusters in a given individual (Weiner and Grill-Spector, 2010) or just one of these (and if the former, which one?). Of course, procedures are sometimes put in place to eliminate these judgment calls, such as choosing only activated voxels that land within a sphere of a given radius around a published activation peak. Any such algorithmic procedure will eliminate experimenter biases in fROI selection, and adoption of a common method across labs will enable replication and direct comparison of results from different labs. But ideally the convention so adopted would be a principled one. Here we propose a particular algorithmic solution for defining fROIs in the ventral pathway that is based on not only the peaks or centroids of activation across subjects for each fROI, but their shape, spatial extent, and anatomical variability across subjects. Importantly, this method does not require strict voxelwise anatomical overlap of fROIs across subjects.

The Group-Constrained Subject-Specific (GSS) method was originally developed for identifying functional regions of interest engaged in high-level language processing (Fedorenko, et al., 2010). This method was designed to discover regions that are activated most systematically across subjects and – crucially – to define the borders around and between each of these regions. Guided by the spatial distribution of individual activations in a set of subjects, this method identifies key “parcels” within which most subjects show activation for the contrast of interest. The selection of individual subject fROIs is then accomplished by intersecting each individual subject's localizer activation map with each of the parcels, thus defining fROIs in each individual subject in a fully algorithmic fashion. We test here how well this method identifies well-established fROIs in the ventral visual pathway. Specifically, we use the GSS method to define face, scene, body, and object selective fROIs in visual cortex, and we compare these fROIs to handpicked regions of interest defined by expert human data coders on the same data. We show that the GSS method is able to identify known category-selective fROIs in visual cortex, and that such fROIs are spatially and functionally similar to those defined using the traditional handpicked approach. Thus, the GSS method retains the benefit of defining fROIs within individual subjects while avoiding the subjectivity common in the traditional individual-subjects fROI methodology. The major category-selective group-level parcels resulting from the GSS analyses on a set of 30 subjects are available online (http://web.mit.edu/bcs/nklab/GSS.shtml) along with instructions and software, so that other labs can easily use these parcels to define individual-subject fROIs in the same fashion.

Section snippets

Participants

Thirty-five participants (15 males, mean age 23, range 18–36) were recruited from the Boston area for this experiment. All participants had good visual acuity, and were free of ophthalmic, neurologic, and general health problems. Participants provided informed consent in accordance with the Internal Review Board at the Massachusetts Institute of Technology.

Design

A blocked fMRI design was used in which participants viewed three-second movie clips of faces, bodies, scenes, objects and scrambled objects

Using the GSS method to identify known category-selective fROIs

Table 1 shows the percent of subjects used to define the parcels in whom each of the best-established ventral pathway fROIs was identified in each hemisphere. First, from the faces > objects contrast, the GSS method successfully identified the main face-selective regions in the right hemisphere: the Fusiform Face Area (FFA) in 93% of subjects, Occipital Face Area (OFA) in 75% of subjects, and posterior Superior Temporal Sulcus (pSTS) in 93% of subjects. The GSS method also identified other fROIs

Discussion

The data presented show that the algorithmic GSS method described here is highly effective in quickly and reliably identifying in individual subjects each of the main face, scene, body, and object perception fROIs in the ventral visual pathway. This method avoids the subjectivity inherent in choosing fROIs by hand, yet identifies regions that closely match the intuitions of human data coders. The major category-selective group-level parcels resulting from the GSS analyses discussed here are

Acknowledgments

We would like to thank the Athinoula A. Martinos Imaging Center at the McGovern Institute for Brain Research, MIT, Cambridge, MA. We would also like to thank David Pitcher and Daniel Dilks for data collection and defining fROIs, and Jia Liu, Chris Baker, Kami Koldewyn, and Russell Epstein for defining fROIs. This work was supported by a grant from the Ellison Medical Foundation to NK.

References (24)

  • R. Epstein et al.

    A cortical representation of the local visual environment

    Nature

    (1998)
  • E. Fedorenko et al.

    New method for fMRI investigations of language: defining ROIs functionally in individual subjects

    J. Neurophysiol.

    (2010)
  • Cited by (208)

    View all citing articles on Scopus
    View full text