Next: Inference
Up: Signal Modelling
Previous: HRF priors
Clearly, a good deal of consideration could be given to activation
height modelling. Typically, within the Bayesian framework people
have looked to include spatial models that incorporate the idea of
finding activation (or indeed non-activation) in
clusters (8,26), and/or to model the
classification of voxels or spatial areas into activation or
non-activation (29). Indeed there are no difficulties
in implementing a continuous MRF on the activation height, for
example. However, in this work for the purpose of signal
modelling we choose to treat each of the voxels as independent.
This is so that we can assess the information that we can extract,
voxel-wise, about the shape of the HRF. We want to be able to
assess how much the HRF varies between voxels in the absence of
spatial regularisation of the signal.