August 2014
Volume 14, Issue 10
Vision Sciences Society Annual Meeting Abstract  |   August 2014
Supervoxel parcellation of visual cortex connectivity
Author Affiliations
  • Christopher Baldassano
    Computer Science, Stanford University
  • Diane M. Beck
    Psychology, University of Illinois at Urbana-Champaign
  • Li Fei-Fei
    Computer Science, Stanford University
Journal of Vision August 2014, Vol.14, 1080. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Christopher Baldassano, Diane M. Beck, Li Fei-Fei; Supervoxel parcellation of visual cortex connectivity. Journal of Vision 2014;14(10):1080. doi:

      Download citation file:

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

  • Supplements

New large-scale studies using fMRI and dMRI have begun to reveal the fine-scale functional and anatomical connectome of the human brain. We have developed a new approach for understanding these massive datasets, allowing us to discover and visualize how connectivity changes over the entire cortical surface. Given a matrix describing the functional or anatomical connectivity strength between each pair of voxels, we apply a nonparametric clustering algorithm based on the distance-dependent Chinese Restaurant Process (ddCRP) in order to group voxels with similar connectivity properties into spatially contiguous "supervoxels." Our method is hypothesis-free, requires no specification of seed voxels, and produces a true parcellation of the brain into spatially-connected subregions rather than ignoring location information. We first validate the clustering method by dividing the Parahippocampal Place Area (PPA) into two subregions based on functional connectivity properties, matching previous work. We then cluster the 59,412-voxel whole-brain group functional and anatomical dataset from the Human Connectome Project, producing a ~200 supervoxel parcellation which provides a compact summary of the full connectivity matrix. For example, resting-state connectivity clustering in early visual cortex divides peripheral V1 from the foveal confluence of V1-V4, revealing that these regions have different functional connectivity patterns with other occipital regions (which match anatomical connectivity differences from probabilistic tractography). In addition to aiding in the discovery of more fine-grained connectivity patterns (allowing us to move beyond a localizer approach to region discovery), the learned parcellation is a general-purpose atlas that can be used to aid in other experiments such as whole-brain decoding. We plan to publicly release the connectivity atlas using an interactive 3D browser-based visualization tool, which will allow anyone to explore the rich connectivity structure of the brain.

Meeting abstract presented at VSS 2014


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.