Bayesian Analysis of fMRI Data with Spatial Priors

• Activity Number: 5
• Type: Invited
• Date/Time: Sunday, August 7, 2005 : 2:00 PM to 3:50 PM
• Sponsor: Institute of Statistical Science, Academia Sinica
• Author(s): William D. Penny*+
• Companies: University College London
• Keywords: fMRI ; Bayesian ; spatiotemporal ; variational

Abstract

In this paper, we describe a Bayesian estimation and inference procedure for the analysis of functional Magnetic Resonance Imaging (fMRI) data. Time series at each voxel are characterized using General Linear Models (GLMs), and a spatial prior is used to regularize estimation of the regression coefficients. This embodies our prior knowledge that evoked responses are spatially contiguous and locally homogeneous. We assume an arbitrary order Auto-Regressive (AR) model for the errors, and these parameters also are spatially regularized. Inference takes place using the computationally efficient Variational Bayes framework. This also provides an approximation to the model evidence that can be used to optimize parts of the model, such as the design matrix and order of the AR model. This allows us to make inferences about spatially extended signal and noise processes thought to generate fMRI. Our model generalizes earlier work on voxel-wise estimation of GLM-AR models and inference in GLMs using Posterior Probability Maps (PPMs). Results are shown on simulated data and on data from an event-related fMRI experiment.