Dynamic Multiscale Analysis of Functional MRI Data

• Activity Number: 52
• Type: Invited
• Date/Time: Sunday, August 1, 2010 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistical Computing
• Author(s): Marco Ferreira*+ and Nilotpal Sanyal
• Companies: University of Missouri and University of Missouri
• Keywords: fMRI data ; state-space models ; wavelets

Abstract

We develop statistical models and methodology to analyze functional magnetic resonance imaging (fMRI) data. There are two main difficulties related to the analysis of fMRI datasets. First, these datasets are usually massive; for example, data for four sessions from one individual are of the order of 0.25GB. Second, these data are usually very noisy and currently used spatial smoothing techniques usually lead to loss of information on shape and extent of the activation area. Here we attack both problems by using fast wavelet decompositions and Bayesian dynamic models. The wavelet decompositions break down the massive dataset into many smaller components that can be efficiently analyzed with the Bayesian dynamic models. Finally, wavelets respect function discontinuities; this characteristic of wavelets leads to sharper estimated shape and extent of the activation region.