I am running skullstripping on a large (>2000) T1w images from different sites and hence different image parameters. I started using the FSL bet function since we plan on using the FSL VMB function package later on and I found that for some datasets there is always a quite large chunk of tissue left form the area that is between the brain and spinal cord (left of the CSF that is left of the Pons).
I changed to AFNI’s 3dSkullstrip and I have the same problem. I played around with some of the tips in the 3dSkullstrip function description but those are mostly for when GM is missing.
Usually this is no problem for fMRI analysis since the T1 is warped and then masked but I want to use these T1w images in native space for some Deep learning applications.
Do you have any tips on how to improve the skullstrip for these datasets?
See attached image. Thanks in advance!
Have you run @SSwarper? That is our generally-recommended program for combining skullstripping (what you want here) with nonlinear warping to a template (bonus for you here!).
We do use the @SSwarper function in our normal fMRI pipeline (yes, it is amazing!). But this is for another project where I just want skull-stripped brains in native space (no non-linear warping) since we want to use them for classification in a 3D CNN. If we use non-linear warping we will potentially lose the differencies between the patient group and HC that we are investigating. This is why we are also using the VBM approach (normalize to template but compensate for the non-linear warp with Jacobians to keep individual differencies).
But step 1 is to use “raw” data, i.e. skull-stripped GM maps in native space. And so far all skull-stripping tools we have used leaves a lot of tissue in the areas that I described. Are we screwed on this? Perhaps without warping or manual segmentation we can’t avoid these regions?
Or can SSwarper skull-stripp without warping to template space?
EDIT: Stopped beeing lazy and tested. @SSwarper did indeed an excelent job skull-stripping without the need of warping. Thank you, it looks so nice =)