Register atlas to DWI space

thanks all for your attention!
I generate an atlas from subject’s T1 using freesurfer and its resolution is 256256256, then I use the 5ttgen and ACT to process subject’s T1 and DWI to generate tracts.
I want to get tracts belong to brain regions in the atlas, considering that the T1 and DWI are not in the same space, so I register the atlas to DWI :

flirt -in /Volumes/Elements/mrtransform/brain.nii.gz -ref /Volumes/Elements/mrtransform/bet_B0.nii.gz -omat /Volumes/Elements/mrtransform/T12DWI.mat

transformconvert /Volumes/Elements/mrtransform/T12DWI.mat /Volumes/Elements/mrtransform/brain.nii.gz /Volumes/Elements/mrtransform/bet_B0.nii.gz flirt_import /Volumes/Elements/mrtransform/T12DWI.txt

mrtransform /Volumes/Elements/mrtransform/MRN_006_hemisphere_7network.nii.gz -linear /Volumes/Elements/mrtransform/T12DWI.txt /Volumes/Elements/mrtransform/MRN_006_hemisphere_7network_inDWIspace.nii.gz -force

this is the registered atlas, background is bet b0:

this is the original atlas:

it seems that the registered atlas is bigger and the original one fit better? what could be the reason?

1 Like

Hi @LiuYuchen,

There are two things here:

  • The diffusion data has been corrected for EPI geometric distortions? if this is not the case, you have to use a non rigid registration to map the T1w to the B0

  • By default flirt performes affine registration, you could optimize a little bit your flirt code by adding -dof 6 and maybe changing the cost function to mutual information to handle images with different contrasts.

Best regards,


Thanks for your reply,
the answer is no, I can’t perform the EPI distortion because of the lack of AP-PA data, I built the whole brain tractography using ACT, while in other answers, they don’t recommend to perform the ACT to uncorrected DWI, do you have any suggestions on tacking?


If you don’t perform EPI distortion correction, you shouldn’t use ACT and to register the T1w (freesurfer parcellation) to the B0, flirt won’t work, you should use other non-rigid registration tools.

Note that you don’t need AP-PA data for that correction, is always better if you have, but there are alternatives. Look here and here.

I hope this helps,

Best regards,


Hi, Manuel!

Your reply helps a lot, I think I need to use non-linear registration to my freesurfer parcellation, I know how to perform it with fnirt, but the fnirt result would resample my freesurfer parcellation from 1mm to 2mm(due to DWI’s resolution is 2mm), do you know how to do it with mrregister?


I don’t know exactly the steps to do it with mrregister, what I usually do in this case, is to upsample the B0 with mrresize to 1x1x1 and then I can use any registration I want.

Best regards,



Clever!I think that would solve my problem :stuck_out_tongue: