next up previous
Next: Results Up: Null Data - Pseudo Previous: Null Data - Pseudo

Methods

The dataset used is the null/rest dataset described in section 7, although, here we do not add artificial activation to the data. We also need to decide on an arbitrary assumed response for modelling the signal, for this we use the random ISI design also used in section 7. We can then use the full model (including the best noise model and full HRF modelling with ARD on the undershoots) on the dataset. For each voxel we obtain the marginal posterior probability over the activation height, $ p(a_i\vert y)$. We can then compute the probability, $ p(a_i>0\vert y)$, for each voxel. We then compute the number of ``positive'' voxels $ N_H$ with $ p(a_i>0\vert y)>1-H$, and plot $ N_H/N$ against $ H$. This is repeated for a range of $ H$.