Bayesian source separation for reference function determination in fMRI.

In analyzing fMRI results, identification of significant activation in voxels is a crucial task. A standard method selects a "known" reference function and performs a regression of the time courses on it and a linear trend. Once the linear trend is found, the correlation between the assumed to be known reference function and the detrended observed time-course in each voxel is computed. But the most important question is: How does one choose the reference function? Here, a Bayesian source separation approach to determining the underlying reference function is described and applied to real fMRI data. This underlying reference function is the unobserved response due to the presentation of the experimental stimulus.