Figure 12.

A posterior probability map (PPM) from a Bayesian first level analysis of one OpenfMRI dataset. Subjects were presented with pairs of either words or pseudowords in a block based design, and made rhyming judgments for each pair. The first level analysis here includes motion correction, segmentation of the fMRI data, smoothing, and statistical analysis. The PPM represents the probability of the first GLM beta weight being larger than zero, and has been arbitrarly thresholded at a probability of 0.99. Note that the PPM has been calculated by using a Gibbs sampler, and not by using techniques based on variational Bayes. Also note that the frequentist approach uses a voxel-wise AR(4) model of the GLM residuals, while the Bayesian currently uses a voxel-wise AR(1) model (due to hardware limitations).