Dear Steve,
>That's correct - however from my previous reply - I think that the
>standardisation that is done is a reasonable one. However, you might
>argue that a possibly better way of scaling the spatial regressors is
>to set the _maximum_ voxel to (say) 1 (in each spatial regressor),
>whilst keeping the best zero-centering. This might give you what you
>want - though might be a 'noisy' kind of standardisation as it would
>depend on the value in just one voxel........it's something that we're
>currently looking into the practicalities of.
Good point, I hadn't thought about that: amplitude- instead of
variance-normalization of the spatial maps. This would mirror the usual
temporal GLM framework, where the regressors corresponding to different
stimulus types have the same amplitude for the elemental HRF response to a
single event. I understand, though, the vulnerability to noise of such
procedure so the method already implemented in FSL -- normalization of the
central fitted gaussian -- may be the safest approach for now.
Thanks again for the very helpful comments,
best
giuseppe
|