|
Abstract:
|
The identification of a specific brain region with a specific function is a central problem in fMRI data analysis. Conventionally, a classical, mass-univariate approach is taken, and the resulting statistic image is searched for significance. Our work develops a hierarchical Bayesian model that improves the standard methods in several ways. First, we pose an explicit spatial model for activations. Second, while some authors have proposed spatial models, they have only considered single subject data; our method models multi-subject data, accounting for intersubject heterogeneity in activation location about a population location. Lastly, we use a fully Bayesian framework, so all sources of uncertainty can be considered and quantified. We estimate the quantities of interest using reversible jump MCMC methods and demonstrate the method on simulated and real data.
|
