| maybe this has been discussed before.
| We just found, that the activated area is much larger with the same
| statistic on normalized data than in the same data without normalization.
| Smoothing (8*8*8) has been performed in both cases as the last step before
| the statistic.
| Could it be, that the normalization step introduces further smoothing and
| if so, how can I determine the resulting smoothing kernel.
Some smoothing is introduced by the resampling step in the spatial
normalisation. For example, if tri-linear interpolation was used to sample
between the centres of 8 neighbouring voxels, then the resulting resampled
value is the average of those eight neighbours. However, if the resampled
point is exactly on the centre of an existing voxel, then no such smoothing
will occur. Off the top of my head, I don't know of an elegant way of
formulating the new smoothness (since the introduced smoothness is not
spatially invariant or Gaussian). Generally, when smoothing an image twice,
the FWHMs combine by Pythagorous' Theorem, so if an image has a smoothness
of 2mm and you smooth by 3mm, then the resulting smoothness is 3.6056mm.
Another reason for the increased smoothness may be because images are zoomed
slightly during spatial normalisation. The template images are about 10%
larger than the average brain. Different voxel sizes may also have some
effect on smoothness estimates.
I hope this helps,
-John
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|