Welcome to the new version of CaltechAUTHORS. Login is currently restricted to library staff. If you notice any issues, please email coda@library.caltech.edu
Published August 7, 2017 | Submitted
Report Open

A Bayesian Unobservable/Observable Source Separation Model and Activation Determination in fMRI

Abstract

In functional magnetic resonance imaging, the most important question to be answered is that of deciding which statistical method to use in analyzing the data. The statistical analysis is the most crucial task because it determines the statistical activations which are interpreted. In computing the activation level, a standard method is to select assumed to be known reference functions and perform a multiple regression of the time courses on them and a linear trend. In performing the multiple regression, t to F statistics are computed in each voxel then they are colored accordingly. But the most important question is: How do we choose the reference functions? Several different functions have been suggested. This paper, based on Bayesian source separation, determines the underlying source reference functions by instead assuming they are known, assessing a prior distribution for them along with the other parameters and determining them statistically. Both Gibbs sampling and iterated conditional modes algorithms are used to determine marginal posterior mean and joint maximum a posteriori values of the parameters along with statistical activation levels. It was found that the underlying response can be statistically determined and that the iterated conditional modes algorithm performed better than Gibbs sampling.

Additional Information

Revised version. Original version dated to May 2001.

Attached Files

Submitted - sswp1120_-_revised.pdf

Files

sswp1120_-_revised.pdf
Files (1.5 MB)
Name Size Download all
md5:0e7350a20b8de661a5e057dd0422b567
1.5 MB Preview Download

Additional details

Created:
August 19, 2023
Modified:
January 13, 2024