September 2017
Volume 17, Issue 10
Open Access
Vision Sciences Society Annual Meeting Abstract  |   August 2017
Functional readout analysis reveals nonlinear representational transformation from early visual to category-selective regions
Author Affiliations
  • Marieke Mur
    MRC Cognition and Brain Sciences United, Cambridge, UK
  • Judith Borowski
    MRC Cognition and Brain Sciences United, Cambridge, UK
  • Nikolaus Kriegeskorte
    MRC Cognition and Brain Sciences United, Cambridge, UK
Journal of Vision August 2017, Vol.17, 1230. doi:10.1167/17.10.1230
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to Subscribers Only
      Sign In or Create an Account ×
    • Get Citation

      Marieke Mur, Judith Borowski, Nikolaus Kriegeskorte; Functional readout analysis reveals nonlinear representational transformation from early visual to category-selective regions. Journal of Vision 2017;17(10):1230. doi: 10.1167/17.10.1230.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Category-selective regions in human visual cortex respond strongly to objects from their preferred category and weakly to objects from non-preferred categories, with a step-like drop in activation at the category boundary (Mur et al. 2012). This categorical response profile across object images is absent in early visual cortex (EVC). How does the brain transform the low-level visual image representation into a high-level categorical object representation? The appearance of a category step suggests a nonlinear transformation. However, objects from different categories also differ markedly in low-level visual properties. If images clustered by category in the EVC representation, a linear transformation might suffice to explain the step in the activation profile of category-selective regions. Can linear readout of the EVC representation explain the activation profile of category-selective regions? fMRI data were acquired in four subjects viewing 96 colored images from multiple categories, including faces and places. We computed the activation profile across images for the fusiform face area (FFA) and the parahippocampal place area (PPA), and for each EVC voxel (Fig. 1A). We used regularized (L1, L2) linear regression to weight the EVC voxel responses, so as to best predict the activation profiles of FFA and PPA. Model performance was evaluated on a hold-out set of images not used for fitting. Our results indicate that linear readout of the EVC representation does not fully explain responses in category-selective regions (Fig. 1B). Adding a category-step predictor explained significant additional variance, especially in PPA. Correlated fluctuations between EVC and category-selective regions that are unrelated to the stimuli contributed significantly to the explanatory power of EVC, especially for FFA as target. These results indicate that functional readout of EVC by category-selective regions is nonlinear, and that differences in lower-level visual properties between categories are insufficient to explain FFA and PPA responses.

Meeting abstract presented at VSS 2017

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×