Dear list,
I'm putting together a DTI preprocessing pipeline with some modification to what FSL's FDT page has recommended, and I'd appreciate it very much your feedback on some of the details:
1. Run dcm2nii on the DTI data, extract original bvecs and bvals
2. Run eddy_correct using the first b0 as reference, with bvecs rotation
[Questions: If the data was acquired with multiple b0 images, would you recommend 1) averaging all the b0s, and use the averaged image for reference, and then discard it; 2) averaging all b0s, split the original 4D file into a b0 images and a DWI images, use the averaged image as reference to correct them separately, then combine them into a single 4D file and discard the averaged image; 3) not averaging the b0s, and just use the 1st one. If not all of the b0s are acquired at the beginning of the acquisition, i.e., some at the beginning while some at the end, how would that change the pipeline?]
3. Use Feat b0 unwarping to apply field map correction and register to native T1 images.
[Questions: 1) Should MCFLIRT be applied if some basic motion correction has been done in Step 2; 2) if motion correction is done in this step, with the registration to T1/Standard, should bvecs be rotated accordingly?]
4. Run BET on the unwarped DTI data, obtain the brain mask
5. Run dtifit with the rotated bvecs to construct the tensor model.
I'm also wondering if an extra T2 image (slightly higher res than DTI) help in registering DTI data to structural. I'm informed that this is sometimes done to bridge the different spaces, but when I tried it out previously I did not see any major improvement. If this extra T2 is used, with the field map, I imagine that this T2 should also be undistorted before being used in registration?
Thank you very much for comments and feedback!
Best,
Yifan
|