Tid: 18 januari 2017, kl 13-14
Plats: B705

Abstract

Spatial whole-brain Bayesian modeling of task-related functional magnetic resonance imaging (fMRI) is a serious computational challenge, due to the huge amount of data in the brain images. Most of the currently proposed methods therefore do inference in subregions of the brain separately or do approximate inference without comparison to the true posterior distribution. A popular such method, introduced in Penny et al. (2005), processes the data slice-by-slice and uses an approximate variational Bayes (VB) estimation algorithm that enforces posterior independence between brain activity coefficients in different voxels. This talk will present a fast and practical Markov chain Monte Carlo (MCMC) scheme for exact inference in the same model, both slice-wise and for the whole brain using a 3D prior on the brain activity coefficients. The algorithm exploits sparsity and uses modern techniques for efficient sampling from high-dimensional Gaussian distributions, leading to speed-ups without which MCMC would not be a practical option. Using MCMC, we are able to evaluate the approximate VB posterior against the exact MCMC posterior, and show that VB can lead to spurious results. In addition, an improved VB method is presented, which drops the assumption of independent voxels a posteriori. This algorithm is shown to be much faster than both MCMC and the original VB for large datasets, with negligible error compared to the MCMC posterior.

Reference: Penny, W.D., et al. (2005). Bayesian fMRI time series analysis with spatial priors. Neuroimage, 24, 350 – 362.