June 2004
Volume 4, Issue 8
Vision Sciences Society Annual Meeting Abstract  |   August 2004
Pattern recognition of orientation-selective fMRI signals in the human visual cortex
Author Affiliations
  • Yukiyasu Kamitani
    Princeton Unversity, USA University of Tokyo, Japan
  • Frank Tong
    Princeton Unversity, USA
Journal of Vision August 2004, Vol.4, 186. doi:https://doi.org/10.1167/4.8.186
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Yukiyasu Kamitani, Frank Tong; Pattern recognition of orientation-selective fMRI signals in the human visual cortex. Journal of Vision 2004;4(8):186. doi: https://doi.org/10.1167/4.8.186.

      Download citation file:

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

  • Supplements

Although the columnar organization of orientation-selective visual neurons has been well characterized in monkeys and cats, it has been elusive how the human visual cortex responds to orientation. Functional imaging techniques used for humans have been presumed to lack the resolution to probe into the sub-millimeter structure of orientation columns. Here we show that stimulus orientation can be decoded from the global pattern of fMRI signals in the human visual cortex by applying pattern classification algorithms that can combine information contributed by weakly orientation-selective voxels. We performed conventional fMRI scans (3T scanner, voxel size 3×3×3mm), while a square-wave annulus grating of one of 8 orientations (0–157.5 deg; 22.5 deg step) with a randomized phase was flashed twice per second in each 16s stimulus block. Pattern analysis was performed on the average activity in each stimulus block, or the single-volume activity of ∼500 individual voxels surrounding the calcarine sulcus. The activity patterns were labeled by the corresponding stimulus orientation, and were provided for the training and the test of pattern classifiers. We used a linear support vector machine (SVM) as a classifier, which determine the hyperplane in the voxel value space that separates two classes of the training patterns with the maximum margin. If pixel intensities are provided as raw input, the classes of gratings with different orientations are not linearly separable (thus not classifiable by a linear SVM). However, the fMRI activation patterns from two classes with 90 deg difference were nearly perfectly classified. In 8-class classification, the predicted orientation showed a bell-shaped distribution, peaking at the correct orientation. Our results demonstrate that fMRI activation patterns preserve orientation information, and that the information can be recovered by pattern recognition of signals from an ensemble of voxels in the visual cortex.

Kamitani, Y., Tong, F.(2004). Pattern recognition of orientation-selective fMRI signals in the human visual cortex [Abstract]. Journal of Vision, 4( 8): 186, 186a, http://journalofvision.org/4/8/186/, doi:10.1167/4.8.186. [CrossRef]
 Supported by a JSPS grant to YK and NIH grants R01-EY14202 and P50-MH62196 to FT

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.