Purchase this article with an account.
Edward Vul, Danial Lashkari, Polina Golland, Po-Jang Hsieh, Nancy Kanwisher; Discovering the structure of object representation through fMRI clustering. Journal of Vision 2009;9(8):797. doi: https://doi.org/10.1167/9.8.797.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
A striking and robust property of the ventral visual pathway in humans is the existence of regions selective for faces, places, and bodies. However, efforts to understand the functional organization of the rest of the pathway have encountered several nontrivial methodological challenges. First, the space of possible visual categories that might produce selective response profiles is large and we lack a theory of how to sensibly sample this space. Second, responses may be selective not just to a single object type, but rather to some set of object classes, a set that might fit our intuitions (e.g. all living things), or might not. Thus, the number of hypotheses to consider is not simply equal to the number of possible object types N, but is instead xN (where x is the number of resolvable response magnitudes). Third, standard brain-mapping methods that search for “blobs” of selective responses in the brain, that is, spatially contiguous regions with a similar functional profile, would miss functionally robust response patterns that are not spatially clustered at the grain of multiple adjacent voxels. Here we present a clustering method that overcomes all three of these problems by searching for functional structure in the ventral visual pathway without assuming either selectivity for single object types or spatial clustering of functional profiles beyond the scale of individual voxels. In positive controls, the method has robustly discovered selectivity for face, places, and bodies, as expected. Importantly, with this method the result was discovered without making any assumptions about spatial clustering of voxels with the same response profile, and without presuming that response profiles should reflect selectivity for a single object class. More ambitious efforts to apply the method to the discovery of new functional structure in the visual pathway are ongoing and will be reported.
This PDF is available to Subscribers Only