Using skull-stripped T1 image from SPM or CAT12 as a brain mask for DWI


I am constantly struggling to get a very nice brain mask for the DWI datasets using tools from FSL or MRTRIX. Recently, I have been using the P1 P2 P3 images from CAT12 segmentations to create a brain mask, and then using ANTS to warp the brain mask into DWI space. The masks are much nicer!

Can I just check if this is a reasonable approach?

Thank you!

This makes sense! We do the same thing in functional studies to get a clean brain mask for fMRI analysis.

Otherwise, what can help if you want to get a clean brain mask directly from your diffusion data with FSL BET or MRTRIX’s dwi2mask : be sure to correct the diffusion data properly for intensity bias with the mrtrix tool for instance. With bet you can tune the parameters: adjusting the center of mass with -c option helps a lot, you can then tune it with -f (extent of the brain you want to include).
But once you get your segmented T1w in the DWI space, it is a good idea indeed to use the sum of P1 P2 and P3, (perhaps with an extra step of holes filling, there is an option for that with fslmaths).

You can also check for different tools that allow to segment DWI images directly into tissues. You can go for tissue-segmentation (as the one implemented in Mrtrix) or use other deep-learning segmentation tools (such as the approach recently proposed by Freesurfer’s team an Eugenio Iglesias - GitHub - BBillot/SynthSeg: Contrast-agnostic segmentation of MRI scans ).

If you wanna go with the registration of T1-mask to DWI space, I do strongly suggest to eddy-correct the images before the co-registration, or 6 DOF registration might fail miserably…

1 Like

Thanks for the tip. It definitely looks like a promising method – nice weekend reading material :slight_smile:

I do not have extra b0 for the TOPUP. Does it still matter that I redo the BET after Eddy?