Replace FSL FAST with ANTs-Atropos for brain tissue segmentation

[effigies]

Please see discussion in Neurostars #4859.

Summary:

1. ANTs N4 + fast -N results in poor segmentation
2. ANTs Atropos seems to work reasonably well
3. N4 + fast seems to work reasonably well, as FAST has built-in bias correction

Options:

1. Switch to Atropos - this has the advantage of removing an FSL, with its licensing issues
2. Use FAST, keeping bias correction in for safety, presumably at some cost
3. Give an option
4. Use FreeSurfer segmentation instead, when available

[oesteban]

I would go for 4 and fall back to 1 for --fs-no-reconall.

[oesteban]

Option 4 is now implemented via #180.

I leave this open, and edit the title to be explicit about option 1.

[lucsanara]

Hello!!

We are running aslprep, with freesurfer preexisting outputs. We have followed the node of the process and we thought that for tissue segmentation it was using fast thought, now having read you we are doubting. Indeed, we were wondering how to improve tissue segmentation. So, how can we make sure that the pipeline is using freesurfer preexisting outputs or ANTS atropos and not FSL fast?

[effigies]

The tissue segmentation is still FAST.