Dear everyone,
one comment on that issue: If you like to manually reorient the data onto AC-PC to have an equally good/bad starting point for all your subjects (as segmentation might fail due to large displacements relative to TPMs), and/or if you want to avoid a skull-stripped version generated with BET, which might conflict with the 6 TPM algorithm, you can also go this way:
- Reorient data series (functional + structural) based on structural volume
- If necessary (e.g. subject was repositioned): Additionally reorient the functional series to correct for displacement relative to structure
(- Slice timing)
- Realignment
- Segmentation of the structure (SPM12 Segment or SPM8 New segment to make use of the 6 TPMs)
- Create a skull-stripped version with Imcalc, e.g. structure * (c1+c2+c3)>.75 seems to make quite a good job on my data
- Coregistration: reference image = skull-stripped structure, source image = mean, other = the functional series
- Normalisation: Write: apply the normalisation parameters from segmentation step onto the functional series
- Smooth
In that case it might be better not to generate a skull-stripped version of the mean, as segmentation seems to take away too many voxels around the boundary (but this might well depend on your functional data/resolution). It could also be useful to use bias-corrected versions, or maybe a resampled mean with a higher resolution. In general, the coreg issue would certainly be worth a further in-depth investigation.
Best,
Helmut
|