I am a bit navie with regard to dti and mrtrix. I have an old data set from 2011 (Siemens DICOMs: 1 b0 +64 directions, anatomical scan, Gre field maps, resting state connectivity and task fMRI + behavioral learning outcome).
I wish to perform mrtrix tractography between my ROIs and correlate its outcome with the behavioral outcome.
What procedure is the best to go. Normal procedure or ACT?
I was reading about the ACT procedure in MRtrix3 documentation and I noticed that geometric distortion correction is very important. To do this can I use the Gre scans (2 mag and 1 phase) and how?
I also noticed that in some dwi volumes (<10 per some subjects) there is some partial signal drop out (Siemens famous bed artifact at the time I guess). Is there any solution for this problem?
Sorry, I did not answer your email yet, I will do it here.
If you want to use tractography to delineate specific fiber bundles and study diffusion parameters in the volume spanned by those bundles, then using ACT is not really that important, as you can rely on the placement of the include ROIs to obtain sensible fiber bundles. In this case iFOD2 tracking will probably do just fine.
ACT (and SIFT) is (are) important if you are using tractography to study “connectivity”. Studying connectivity using tractography is a bit like treading a minefield, but if you are going to do it anyway, you should use ACT and SIFT.
In order to use ACT, you need your structural scan and diffusion scan to be aligned spatially. This indeed includes correcting your EPI data (diffusion scans) for susceptibility induced distortions. In theory, you should be able to do that with the Gre scans. However, this is not something that is supported by MRtrix. I believe FSL has functionality to do this, but I have never had much succes with it myself.
About your signal dropouts: as long as they manifest as outliers in the model fit (i.e. they are not present in every volume), we can address the issue using a robust estimator, as Quinten has shown you. This has already markedly improved the estimates. However, if the problem is present in every single volume, that is a lot harder to tackle. I don’t know of any post-processing methods specifically addressing the bed vibration artefact.
Thanks Ben for the reply.
So in order to proceed with iFOD2, I don’t need to do the GRE correction? (since I could make it work in fsl!)
And a naive question: Where exactly in the pipeline should I use the Quinten’s script?
I was reading about the ACT procedure in MRtrix3 documentation and I noticed that geometric distortion correction is very important. To do this can I use the Gre scans (2 mag and 1 phase) and how?
I would look into SPM and FSL for doing EPI distortion correction using multiple-echo gradient echo images. Personally I’m not a fan of this approach, as a significant amount of spatial smoothing is required to condition the estimated inhomogeneity field, which restricts the method from identifying localised inhomogeneities.
I also noticed that in some dwi volumes (<10 per some subjects) there is some partial signal drop out (Siemens famous bed artifact at the time I guess). Is there any solution for this problem?
The only automated approach I know of tailored to this sort of artifact is this one. I don’t know if the method has been made available. But if you have as many as 10 volumes with dropout, the data may not be usable: rejecting such volumes will not be equivalent to just reducing your number of diffusion sensitisation directions, because the directions for which dropout occurs due to the Siemens vibration will tend to be clustered together along the x axis, so their removal would result in a large ‘hole’ in your coverage of the half-sphere.
An model-outlier-based detection may struggle here also if there are many volumes with dropout with similar sensitisation directions, as those values will have considerable influence on the model fit and they will therefore cease to be outliers (though I don’t know which specific method Ben is referring to).
Hi Rob
1- As far as I remember in that paper they suggested: a correction approach that refurbishes DWIs by EXCLUDING gradients most affected by vibrational artifacts using the entropy measurement.
2- I also think since they are mainly around X direction, removing them may not be a good idea.
I totally agree with @rsmith regarding the vibration artefact. When we investigated this (quite a few years ago now…), the DW directions affected were very clearly those with a significant X component, leading to obvious ‘red’ areas in the DEC FA map. Outlier rejection (or any other method to correct for this) may remove all of these corrupted values, but that will then leave a gaping hole in the orientations. It’s hard to see how that wouldn’t bias the data.
On top of that, the artefact is so systematic that I’d actually be surprised if outlier rejection methods were even able to pick up these affected values reliably. The artefact gets stronger the larger the X component, so the DW signal will smoothly transition from non-corrupt to totally destroyed as a function of orientation. There’d be nothing to suggest a given orientation is inconsistent with its neighbours.
So if you were to try something, I strongly recommend you inspect the results very thoroughly before using them any further…