Dear Krish,
> 1) Isn't it necessary to smooth the data to a particular level so that
> the appropriate criteria are met for Gaussian Field Theory? If so using
> no smoothing or variable size kernels might not be such a good idea.
That is correct (GFT requires the residuals fields to be a good lattice
approximation to an underlying Gaussian field).
> 2) An oft mentioned rule of thumb is that the kernel should be "2-3
> times the voxel size". Should that be 2-3 times the original voxel
> size (e.g. 3.75 3.75 7) or 2-3 times the voxel size of the spatially
> normalised volume (which is 3mm cubic)? Presumably the normalisation
> itself introduces some degree of smoothing.
The smoothness should be 2-3 times the voxel size of the data that
enters the estimation procedure (i.e. after spatial normalization).
One can therefor subsample the data to smaller voxels and then smooth
with a kernel that approximates the original voxel size. It is
important to note that the smoothness is the post hoc smoothness of the
residual fields (given in the SPM table footnotes). This smoothness
may be much greater than the size of the smoothing kernel.
With best wishes - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|