Spatio-Spectral Approach in fMRI Data Analysis: An Approach to Dealing with High Dimensionality of the Data

• Activity Number: 15
• Type: Topic Contributed
• Date/Time: Sunday, August 1, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Hakmook Kang*+
• Companies: Brown University
• Keywords: spatio-temporal ; fMRI ; spatio-spectral ; Neuroimaging ; mixed model

Abstract

One goal of brain imaging (fMRI) is to identify brain networks and local regions involved in processing information. The standard analysis uses voxel-based linear regression models and often does not take into account the spatial correlation in the fMRI data. Ignoring spatial correlation can lead to misleading results. Due to high spatio-temporal dimensionality of the fMRI data, it is challenging to fit a full spatio-temporal model. Here, we develop a new approach by modeling the spatio-temporal correlation in the frequency domain. This significantly reduces the complexity of the problem and the computational burden because, under mild assumptions on the stochastic process, the spatio-temporal covariance matrix is approximately block diagonal. We estimate model parameters via maximum likelihood. We apply the method to a fMRI data set collected in a working memory experiment.