Fixel based analysis with single shell

Dear all,

I am interested in performing Fixel based analysis with single shell data.
I have noticed that there are four pipelines mentioned in mrtrix/readthedocs and mrtrix community for data pre-processing before FOD estimation.

1. First pipeline can be found here
http://mrtrix.readthedocs.io/en/latest/workflows/DWI_preprocessing_for_quantitative_analysis.html and brings the title “DWI Pre-processing for Quantitative Analysis”. The steps outlined in this pipeline are the following:

  1. DWI denoising
  2. DWI preproc
  3. Brain mask estimation
  4. Bias field correction
  5. Global intensity normalisation across subjects
  6. Computing a group average response function

2. Second pipeline can be found here
http://mrtrix.readthedocs.io/en/latest/workflows/fixel_based_analysis.html and brings the title “Fixel-Based Analysis (FBA)”. The steps outlined in this pipeline are the following:

  1. Upsampling DW images
  2. Compute upsampled brain mask images
  3. Fibre Orientation Distribution estimation

3. Third pipeline can be found here
http://mrtrix.readthedocs.io/en/latest/tutorials/basic_dwi_processing.html and brings the title “Basic DWI processing”. The steps outlined in this pipeline are the following:

  1. DWI preproc
  2. Brain mask estimation
  3. Response function estimation
  4. Fibre Orientation Distribution estimation

4. Fourth pipeline can be found here Single-shell data. The steps outlined in this pipeline are the following:

-Upsample your dMRI to the same isotropic resolition of your sMRI
-Preprocessing your diffusion data (eddy correction, bias, etc…)
-Co-register your strucural data to the diffusion
-Correct for distortion using restricted registration (this option is implemented in ANTs, with the option -g). Take a look to this paper4 and this one3 for more details.
-Now booth images are aligned, if do you think that the alignement is good, you can perform ACT.

In my effort to unify the above I ended up with the following pipeline:

  1. Denoising
  2. Upsampling
  3. Preproc (for eddy only, no topup)
  4. Brain mask extraction
  5. Bias field correction
  6. Global intensity normalisation across subjects
  7. Group average response function estimation
  8. Structural and diffusion data co-registration
  9. Distortion correction using restricted registration (this option is implemented in ANTs, with the option -g).
  10. Fibre Orientation Distribution estimation
  11. Study-specific unbiased FOD template generation
  12. Registration of all subject FOD images to the FOD template
  13. Computation of intersection of all subject masks in template space … next steps as listed in the FBA page

Could you please let me know if you would rearrange/amend/delete any of the above processes or whether you would add any?

Do you also might have a specific command-line/ link for the following step: Correct for distortion using restricted registration (this option is implemented in ANTs, with the option -g)?

Cheers,
Vasiliki

Hi Vasiliki
To perform FBA you could just follow pipeline 1 and then 2. The reason why I spilt them is because you only need to perform pipeline 1 if you are wanting to investigate Apparent Fibre Density. As noted in the fixel-based analysis workflow you can modify just 2 steps to perform comparisons of any other fixel-based metrics (e.g. CHARMED fibre volume fractions). Depending on the other models, they may not require bias field correction and intensity normalisation if they assume it’s safe to normalise by the b=0 signal in each voxel.

I’m currently improving the documentation that will be released with the next MRtrix release. So hopefully it will become more clear soon.

With respect to using ANTS -g to correct for distortions without any reverse phase encode pair. Technically you don’t need to do this for FBA (only if you want to use ACT for tractography and connectivity studies), however if it works well you might as well. I’d recommend doing this after Preproc. Assuming you also correct the hypo and hyper-intense signal in the susceptibility distorted regions (by modulating by the warp Jacobian determinant), then the subsequent bias field correction and tensity normalisation may benefit.

As per pipeline 1 an 2, I’d recommend up-sampling at the last minute before computing FODs, and not so early on as in your steps. This is mainly to save space and computation time.

Cheers,
Dave

Hi Dave,
Thank you very much for your response. It is very helpful.
Cheers,
Vasiliki