Hello Niels,
Your observation is absolutely normal. Segmentation-normalization will indeed result in slightly "smaller" brains (which should actually be the "correct" size as far as I can see).
This is related to differences in smoothness between T1/EPI template and the tissue probability maps. Look at the images in tpm folder compared to the templates, the border between "outside" and GM is much sharper in the TPM file compared to the T1/EPI-template (which have been smoothed to a larger extent).
The normalization provided by the segmentation procedure should be more accurate than the normalization with EPI/T1 template because it uses more information of the original volume. Therefore it should lead to the best overlap between different subjects afterwards, at least in anatomical terms (which does NOT have to mean to get higher p-/t-values).
In any case, never mix different normalization procedures within a study. EPI template and T1 template will result in different results as well, even if your functional series has been coregistered perfectly well.
Best,
Gabor
|