Improve registration of AMU WM/GM mask and PAM50 template

[jcohenadad]

Following a discussion with the Marseille group, we decided to update the PAM50 template to fix some of the issues reported in #32.

• archive the AMU template somewhere
• Register AMU to PAM50 T2 #40
• Symmetrize objects after AMU to PAM50 registration #42
• Fix negative values of the T2* mask #44
• Consider scaling registered AMU GM mask to 1.0 #43
• Regenerate the lumbar and top cervical GM mask #45
• adapt edges of the cord – currently very smooth
  • unclear what approach to choose
• regularizing cord values inside the cord (eg: some values are 0.999) – that might be addressed by the previous point about the smoothness of the edge
• @joshuacwnewton : regenerate the WM and GM atlas → takes as input the “new” WM mask (ie newly registered to the PAM50)
  • Create_atlas.m
  • Assuming that GM/WM regeneration will naturally address the stray voxels (make sure that create_atlas.m does not introduce these stray voxels)
• @jcohenadad recreate the cord mask
• @joshuacwnewton @mguaypaq: Documentation + review
• @vcallot Validation of results
  • Batch_processing.sh results

Fixes #32 #40

[jcohenadad]

At this point of development, the GM mask is much better registered:

The T2* registration on the other hand looks quite similar than before:

[vcallot]

The T2* registration on the other hand looks quite similar than before:

Agree. When looking in the axial plane, it looks like the T2* image is still a bit dilated toward the anterior (top) direction. One or two pixels (+ bean shape for T2 versus circular shape for T2*)

[jcohenadad]

When looking in the axial plane, it looks like the T2* image is still a bit dilated toward the anterior (top) direction. One or two pixels (+ bean shape for T2 versus circular shape for T2*)

@vcallot That might stem from the pipeline used to generate the AMU15. Ideally, the AMU15 and PAM50 should follow a similar pipeline. One workaround, would be to regenerate the AMU15, which means I would need access to the original data and code that was used to generate the AMU15 template.

This would also help address one of my concern regarding the validity of the probabilistic values of the AMU15. Depending on the type of registration parameters that was used to generate the AMU15, it is possible that part of the reason the original AMU15 GM mask looks so "smooth" (with, overall, lower probability values than the current PAM50 GM mask), is that the inter-subject registration was imperfect, leading to this increased smoothness.