Gaussian Spatial Modeling on Signal Detection of fMRI Data

• Activity Number: 79
• Type: Contributed
• Date/Time: Sunday, July 29, 2012 : 4:00 PM to 5:50 PM
• Sponsor: ENAR
• Author(s): Shuzhen Li*+ and Lynn E. Eberly
• Companies: Medtronic and University of Minnesota
• Keywords: fMRI ; FDR ; SPM

Abstract

In this work, we apply a Gaussian Spatial model to statistical parametric map (SPM) data in the third stage of fMRI analysis (Marchini, 2004). Since functional effects in the human brain are assumed to be clustered, not randomly scattered, the effect of neural and/or spatial correlation between neighboring voxels needs to be accounted for in an analysis to identify brain regions activated by the experimental stimuli or neural disease (e.g., Alzheimer's disease or Schizophrenia). A sliding window technique is used to fit the Gaussian spatial model to each 3D window (e.g., 5 × 5× 5 voxels) and then de-correlate all voxels in that window. Only the center de-correlated voxel value is kept. The `sliding window' is then moved around the whole brain volume to de-correlate every voxel and thus a de-correlated SPM forsignal detection can be produced. We applied the widely used Benjamini-Hochberg False Discovery Rate (FDR) thresholding approach to both the original and the de-correlated SPM (Benjamini, Y. and Hochberg, Y., 1995) under various simulations. We compare sensitivity and achieved FDR, and apply the approach to fMRI data from an auditory word-pair-associates experiment.