> I am a little confused as to the selection of the flags for spatial
> normalization in SPM2 ...
>
> For example, the help file for spm_normalisation specifies a DCT base
> default cutoff of 30 and a (light) regularisation value of 0.1. However,
> the actual spm_defaults settings are 25 and 1, respectively.
The help file was probably written at a different time than the actual code.
I suspect that I increased the default regularisation and decreased the
number of basis functions in order to increase the stability and reduce the
computation time. Selection of the optimum values is a matter of empirical
exploration. The default values are those that work best for data from the
FIL.
>
> I am not sure whether there is a reason for this discrepancy. Any
> clarification would be appreciated, especially since I'm not really sure
> how much (if at all) these setting would affect the result of the
> normalization process.
The settings are likely to influence the behaviour of the spatial
normalisation, but it is not always so straightforward to figure out the way
in which it would be affected. Changing the settings can improve
registration in some brain regions, but decrease it in others. The
evaluation work that has been done by Hellier et al seems to suggest that
more heavily regularised methods can give a smaller registration error,
whereas less regularised methods can make one image appear visually more like
another.
Best regards,
-John
|