running eddy on old data

[hanayik]

I would like to run eddy on a quite old (2015) DTI dataset as a preprocessing step. The measurement was acquired only in AP with a resoluition of 2x2x2 mm3, 1 b0 and 30 directions spanning just a half-sphere. The whole sequence (1b0 + 30 directions) was repeated three-times as separate measurements right after each other to increase SNR. I merged these three repetitions and this is my command:

time eddy_cuda7.0 --imain=dti_merged.nii.gz --mask=dti_merged_brain_mask.nii.gz --acqp=acq_param --index=index --bvecs=bvecs --bvals=bvals --out=dti_eddy --ol_nstd=3 --slm=linear --fwhm=10,0,0,0,0 --nvoxhp=5000 --repol

Because the data is suboptimal for eddy some default parameters were changed: second level modelling was turned on (--slm=linear), --ol_nstd was decreased to 3, fwhm=10 was added to the first iteration, --nvoxhp was increased and --repol was used. Could you please verify that running eddy with the above parameters has a good chance to work on my old data merged from 3 repetitions. If it is not appropriate, could you please advise parameters that I should modify or additionally insert?

[hanayik]

I don’t see any problems with using eddy on this data set. You are right that --slm=linear is a good idea for data sampled on the half sphere.

I don’t think that you should need to reduce the ol_nstd to 3, I would have thought that the three repetitions would give you enough power to use the default.

A fwhm=10 will make it more robust to any large movements. Though I think I would personally add it rather than replace the first fwhm=0 iteration (i.e. --niter=6 --fwhm=10,0,0,0,0,0).

There is never any harm in increasing --nvoxhp, apart from execution time.

--repol is always a good idea.

On the whole I think it looks like a good set of parameters given your data, and assuming that this is not a group that moves “unusually much”.