-----------------------------------------------------------
Statistics Seminar Series
Thursday, Nov. 8, 2012, 4:00 PM
Cullimore, Room 611
New Jersey Institute of Technology
-----------------------------------------------------------
Bayesian General Linear Model for fMRI Data
Yu Ryan Yue
Baruch College, The City University of New York
Abstract
The general linear model (GLM) (Worsley and Friston, 1995) has arguably become
the dominant statistical method to analyze functional Magnetic Resonance Imaging
(fMRI) data. It models the time series of response as a linear combination of
different signal components and tests whether activity in a brain region is
systematically related to any input functions. In this work we propose a
Bayesian GLM method, which has following advantages compared to the existing
methods. First, the proposed method accounts for the spatial-temporal structure
in fMRI data by using a class of sophisticated spatial priors to model baseline
and amplitude parameters, and assuming autoregressive errors on the temporal
domain. Our spatial priors feature adaptive smoothing, edge correction and easy
computation. Second, we propose a method to estimate activation regions so that
the probability for exceeding the level in the entire set is equal to some
predefined value. It corrects posterior probability map (PPM) approach for
multiple comparison. Third, we employ an approximate Bayesian inference tool to
relieve the computational burden imposed by the huge size of fMRI data.