Shift in PE direction after processing with dwifslpreproc

Software Support
,
1

Summary of what happened:

Hi everyone,

I’m trying to preprocess DWI phantom data for a test acquired at fairly high b-values (bmax=8000) using mrtrix dwifslpreproc tools for Topup/Eddy correction tool. The data is has a total of 6,6,15,15 shells for b=500, 1500, 4000 and 8000 with additional 6 b-zeros spaced across the acquisition.

A large shift in the PE direction appears for the higher b-shells that is not visible in the unprocessed images.

Command used (and if a helper script was used, a link to the helper script or the command generated):

The dwifslpreproc is ran after a basic denoising using dwidenoise with MP-PCA default settings. Mask fed to eddy is obtained using median-otsu from dipy before all preprocessing.

dwifslpreproc "${denoise_mif}" "${preproc_mif}" \
              -rpe_pair \
              -pe_dir "${PE_DIR}"\
              -se_epi "${se_epi_pair_mif}" \
              -eddy_option "${eddy_opts}" \ # --data_is_shelled --flm=quadratic --slm=linear --repol --dont_peas --json=${dwi_json}"
              "${readout_opt[@]}" \
              -eddy_mask "${mask}"\
              -nthreads "${NTHREADS}"\
              -nocleanup

Version:

MRtrix 3.0.8
FSL 6.0.7.19

Environment (Docker, Singularity / Apptainer, custom installation):

Installed through anaconda package manager

Screenshots / relevant information:

The same problem happens on another test that I performed, where the b=8000 was not acquired, leaving only b0,b500,b1500 and b4000. Another thing to note is that I’ve acquired the data across 5 days, and only on one of the days I get this problem.

I can send/share the data if possible, has it might be more easy to spot the shift on the data instead of including screenshots.

2

Could you check if the “PhaseEncodingDirection” specified in the json files are correct?

3

Hi, thanks for the suggestion. Yes, I did check if the “PhaseEncodingDirection” is correct in the json files.

The data was converted from dcm to *.nii.gz with dcm2bids. For the dwi data, the phase encoding direction is AP: “PhaseEncodingDirection”: “j-” and for the fmap (reversed PE b0): “PhaseEncodingDirection”: “j”

4

I’ve also included a GIF of the shift. We can see the displacement between the b1500 and b4000 shell here.

shift_pe_phantom

5

Do you mean that the shift appears between different shells?
This reminds me about the --dont_peas argument that can be used in FSL’s eddy. The default of eddy is “peas” (“Post Eddy Alignment of Shells”).

6

Yes exactly, the shift appears between the b4000 and b1500 in this exemple. I also tried with the --dont_peas argument but no luck.

I’ve augmented the number of iterations to --niter=8 and --fwhm=10,8,4,2,0,0as recommended for agitated “subjects”, it worked for some of the runs but not all of them…

I’ll try combining the --dont_peas option to the higher number of iterations to see if that can work.

7

Looking at your data, I wonder if FSL eddy is well suited for correcting this kind of data.You may want to ask the developers directly on the FSL forum.