I'm running into some errors while converting 7T multi-shell diffusion data.

I am using the most recent version of dcm2niiX (v1.0.20180622). While converting I get several warnings:
Slices stacked despite varying acquisition numbers (if this is not desired please recompile)
Warning: Saving 69 DTI gradients. Validate vectors (image slice orientation not reported, e.g. 2001,100B).

The conversion finishes and produces the files, however, the diffusion data looks very grainy. I have tried several possible flag options (-m, -p, -i) of dcm2niix but it does not seem to change the resulting nifti or the warning messages. I have continued analysis anyway to see what happens and performed tractography using a previously successfully used pipeline using AFNI which tests several flip options for the gradients, but tractography is not successful.

I would appreciate your insights on the data and conversion. I have attached the .json file for information of the scan.

Many thanks,
Sandra 

Although I have been working with diffusion data for several years, I have not worked with multi-shell and 7T data, so apologies for any misunderstandings that might have arisen from that.

Individual DWI images typically have low SNR, so they appear grainy, in particular if you use higher B-values (which provide more directional specificity) of multi-shell sequences. The dcm2niix conversion should be lossless, so you can compare the output with your favorite DICOM viewer to convince yourself that all converted well. With MRI we typically want to merge acquisitions, so you can ignore that warning (the vendors disagree on what counts as an acquisition for CT scans, which explains the warning).


I suspect your images will be fine once you process the data. As your images were acquired with partial fourier, you can not use de-Gibbs, but the standard dwi denoise, TOPUP, Eddy should give you a nice idea of you data quality. Since you have a Siemens system, I suspect the vector conversion is fine, but I always encourage users to run through the dedicated document on the dcm2niix wiki.

I also note that your sequence uses very small voxels - there is about half as much hydrogen in each voxel than the 7T HCP sequence. Very small voxels necessarily trade off SNR for spatial resolution. You will probably need a lot of signal averages to get nice data. I am a huge fan of the HCP sequences, though in my opinion they already push the limits of spatial resolution to the ragged edge. You may want to use those sequences as a starting point. That would also allow you to compare your data with theirs.