Dwipreproc: What's the difference between -rpe_pair and -rpe_all? Which one should be used for HCP data?

1

Hi Community,
In the command dwipreproc, what’s the difference between -rpe_pair and -rpe_all? Now I have a HCP data, which one should I use? And, what’s format of using -rpe_all?

Thanks!

Best Wishes,
Jian

2

Hi Jian,

  • -rpe_pair should be used when you have acquired one (or more) pair(s) of b=0 volumes that are to be used purely for estimating the inhomogeneity field; these volumes do not form a part of the output image, but the estimated field is used to correct for geometric distortions present in the input DWIs.

  • -rpe_all should be used if you have acquired all of your DWI volumes (both b=0 and diffusion-weighted volumes) with opposing phase encoding. In this scenario, a volume recombination is performed, that combines the two volumes acquired for each diffusion b-vector into a single volume, which helps to reconstruct signal in areas of signal pileup / stretching.

In your case, since you are almost certainly working with minimally processed HCP data, the steps performed by the dwipreproc script have already been applied prior to your downloading the data, and hence you do not need to use this script for those data.

And, what’s format of using -rpe_all?

I’ll leave this not only since I don’t believe you need to use it, but also the usage of that option is just about to change, and there will be corresponding updates in the documentation when that change comes.

Cheers
Rob

3

Hi Robert,
Thank you for your answers.
I got it.

Cheers
Jian

4

Hi Robert,

I want to use the SIFT for the HCP data. Now I’m a little confused with the processing of Bias field correction.
In the tutorial of HCP, it doesn’t mention the Bias field correction.
http://mrtrix.readthedocs.io/en/latest/quantitative_structural_connectivity/ismrm_hcp_tutorial.html

But in the document of SIFT,you have emphasized the Bias field correction.
http://mrtrix.readthedocs.io/en/latest/quantitative_structural_connectivity/sift.html

So for the HCP data,Is it necessary to process the Bias field correction ? Or if I don’t have the processing,the result of SIFT is acceptable ?

Thanks
Best Wishes
Liyuan

5

Hi Liyuan,

The HCP tutorial was written purely for users to be able to recreate the analysis pipeline that was demonstrated at ISMRM in 2015. So it shouldn’t be interpreted as the “recommended” pipeline for using HCP data, and probably won’t be kept up-to-date with new and future developments.

With regards to B1 bias field specifically, the bias field is in fact quite strong in HCP data and so in general I would in fact recommend using it. Without it, SIFT will still be beneficial; but the strength of the bias present in the B1 field will propagate through to the resulting connection densities.

We additionally have a novel method that performs both bias field correction and intensity normalisation in a single step. It was included in yesterday’s release candidate update as the command mtbin (whoops, it didn’t get included in the list of new executables! Fixed.), and is listed in the instructions for fixel-based analysis of multi-shell data. For anyone here at ISMRM, this method will be presented today at 2:45 at computer 95 (abstract 3541) (look for @Thijs).

Cheers
Rob

6

Hi Robert,

I feel a great pity that I am not attending the ISMRM. Fortunately I always pay attention to the new methods of Mrtrix.

About the AFD,in the single-shell ,I found the bias field correction and intensity normalisation (dwibiascorrect+dwiintensitynorm) are done **after the DWI general pre-processing.**In the multi-shell,the new method mtbin is done after Fibre Orientation Distribution estimation.
The difference of single and multi shell, maybe the mtbin isn’t updated timely in the single shell?

So for the SIFT,the bias field correction(dwibiascorrect) should be performed after DWI general pre-processing(dwipreproc)?

Cheers
Liyuan

7

About the AFD, in the single-shell, I found the bias field correction and intensity normalisation (dwibiascorrect + dwiintensitynorm) are done after the DWI general pre-processing. In the multi-shell,the new method mtbin is done after Fibre Orientation Distribution estimation.

The difference of single and multi shell, maybe the mtbin isn’t updated timely in the single shell?

It’s not currently possible to use the mtbin command with single-shell data. The method is intrinsically dependent on being able to perform a multi-tissue decomposition of the diffusion data, which is currently only possible from a multi-shell acquisition (for three ‘tissues’, at least). So the single-shell pipeline documentation still describes the earlier bias field / intensity normalisation process.

So for the SIFT, the bias field correction (dwibiascorrect) should be performed after DWI general pre-processing (dwipreproc)?

We’ve generally done bias field correction after motion / eddy / distortion correction, since it estimates the bias field from the mean b=0 image, and it’s therefore beneficial to have had any inter-volume motion corrected before combining those volumes.

Cheers
Rob