Combining two DTI scans for tractography

Dear MRtric community,

We have identified a limitation in our scanner regarding the utilization of different directions for different b values. Specifically, we are currently unable to independently acquire images with b=1000 and b=2000 while using varying directions and gradient tables.

To overcome this limitation, we propose a solution: acquiring the b=1000 images with 30 directions and the b=2000 images with 45 directions separately, all on the same subject, while keeping all parameters constant except for the gradient directions. We are wondering if MRtrix offers any commands that could facilitate the process of combining these separate scans. Additionally, we have concerns about the potential impact on tractography performance when combining these datasets. I would appreciate your input


1 Like

Hi @pabbasian,

This is not uncommon, and the solution you’ve come up with is how many centres will also be doing it. There is a command in MRtrix called dwicat, which is designed to merge these kinds of datasets and avoid the most common issues you might encounter in the process.

All the best,

1 Like

Thank you, Donald.

Dear Donald,

I have two relevant questions with the same kind of dataset. I’m using dwicat as dwicat dcmdir1/ dcmdir2/ dcmdir3/ dwi.mif:

  1. Is there a way to extract the JSON file using this method?
  2. Regarding the mask, the subject may have moved between the acquisitions and masks derived from each may have slightly mismatch. So, which mask should be provided to it?


Hi @AmirHussein,

Yes, you should be able to extract the JSON from the resulting DWI.mif using mrinfo – there are options to handle this. Failing that, you can convert the dwi.mif to e.g. NIfTI usingmrconvert using its -json_export option.

Regarding the mask, I would suggest you treat your combined dwi.mif as you would a regular single input, things should broadly work as expected. Yes, motion may be a problem, but it’s a problem for single series too. There’s no masking required until the later stages anyway, so provided you’ve managed to get past the motion correction with dwifslpreproc, you should be in a position to derive a single mask from the output of that command – which is what we’d recommend anyway.

Hope that all makes sense!

Great. Thanks!