Dear Alessio,
I've lunched the invwarp process approximately 120 hours ago.. and it hasn't finished yet.. I read on the fsl-fnirt page that this process could be "quite computationally intensive, and may take quite long to run".. but I didn't think it could take this much time.. Since FNIRT finished in not more than 30 hours, I was thinking something went wrong..
Any suggestion?
the most important factor deciding the time it takes to run invwarp is the FOV of your structural scan. It will spend a vast majority of its time trying to work out the inverse for the parts were it isn't properly defined, i.e. the parts of the FOV in the structural that has no correspondence in the template (where the FOV is a rather snug fit around the brain).
So, by cropping your structural to a reasonably snug fit around the brain (see the template for an example of "snug") _before_ any other processing you will save a lot of time.
N.B. that running bet is _not_ cropping. That just changes the intensity values while leaving the FOV unchanged. Use fslroi to crop.
Some time in the not so far future there will be a new version of invwarp that is a little better at ignoring the unimportant parts of the FOV, but until then cropping makes a BIG difference.
Good luck Jesper