Necessary processing steps and their order? Registration before or after dwipreproc?

Dear Mrtrixler,
for my study I do different measurements but all with the same person. Due to the fact that the measurement is about 6 hours I record the data in 2 sessions.

Because I am not very familiar with all necessary processing steps and their order I would like to ask some questions.
First I was thinking to do the DWI denoising followed by the preproc script. After that I am not sure if I have to run a kind of registration, thus all my images are in the same space? When I do it (mrregister) do I have to register them all to a T1 image or to the first non diffusion image (b0)? I also read (Leemans, Jones Paper b-matrix rotation) so I was wondering if I have to reorientate my b files after registration?

So my questions are:

  1. Do I need to run a registration
  2. If yes…how will be the order of my pre-processing pipeline?
  3. On which image should the registration refer to and do I need to reorientate my b matrix?
  4. Do I have to do more preprocessing steps? (I want to estimate the directional information, so calculate fod’s and peaks, single shell and multi shell approach)

Thank you for any kind of help!

Hi Max,

Have a look at the dwipreproc* doc entry, in particular, section 4. The dwipreproc script does take care of the alignment and distortion correction using FSL’s eddy and topup commands.

Another pre-processing step is the bias field correction of the DW image via dwibiascorrect. I’d suggest you get started by reading this and clicking through the tutorials.


*replaced in version 3.0.0 by dwifslpreproc

Hi Max,
thank you for your help.
I have read both tutorials and already did the “Beginner DWI Tutorial”. But now I would like to do the Multi-tissue constrained spherical deconvolution approach. After reading the documentation I still have some questions:

1.The documentation says that I need to do a rigid-body registration between a T1 and the DWI images. Because there is no information how to do it I guess I just have to do the mrregister command with the rigid option?
2. Do I have to use the T1 image as template or as moving option? I guess as moving option?
3. Do I have to preprocess the DWI’s and the T1 before coregister?
4. I read that Eddy already includes the b-matrix reorientation but when I do a registration afterwards, do I need to reorientate again my b-matrix or is it just necessary for subject motion correction? I guess when I coregister just the T1 to the DWI’s I do not have to reorientate the b-matrix?
5. You suggested me the dwibiascorrect option. Is it also a necessary step cause I could not find it in the pre processing section on the documentation website?
6. I also try to find a solution by searching in the forum, but this confused me even more. In both i.) ii.) questions they were talking about freesurfer and flirt?

So I am just wondering if I have to do additional steps in my pipeline? Otherwise I would do it the way it is suggested on the documentation website?

  1. Do dwi denoising, dwipreproc, dwibiascorrect
  2. Registration of T1 with DWI (mrregister)
  3. Do 5ttgen, dwi2response
  4. Do dwi2fod msmt_csd

Sorry for all my questions but its my first time processing such data.
Thank you

I’d use a different software that allows using cross correlation or mutual information as cost function (for instance flirt).

Yes to all.

Assuming you want to process your data similar to the MSMT paper the steps are:

A different multi-shell approach that does not require the co-registered T1 is:

  1. dwidenoise, dwipreproc, dwibiascorrect, (dwi2mask)
  2. dwi2fod dhollander


Small typo: that should be dwi2response dhollander of course. :slight_smile:

@Max_Wichmann: dwi2response dhollander is an alternative, often much more accurate, way of obtaining multi-tissue (WM, GM, CSF) response functions from your data. It doesn’t need a co-registered T1w image, or 5tt segmentation thereof. For more information on the algorithm, see this abstract. The method only needs your (preprocessed) DWI data, and automates all other steps to obtain the response functions.