I’m trying to extract whole brain connectivity matrices from a single-shell dataset (2 acquisitions of 33 directions, bvalue=1000) which unfortunately I cannot correct for EPI distortion. I was wondering what is the recommended processing with mrtrix in this case, and what would be the most ‘meaningful’ measure of connectivity that I can aspire to use.
I thought about generating the tracts between each pair of nodes and averaging mean diffusivity or fractional anisotropy within each of them, as it has been done in many studies, but perhaps there are better options with mrtrix with this dataset.
I understand I should use lmax =6 and FOD determination with the tournier algorithm. I suppose I cannot use ACT because I cannot correct for geometric distortion.
Is there anything else I should think about or a recommended protocol or other suggestions?
Thanks, Manuel, I do have structural data.
I thought that this was not advisable due to a post I read a while ago, but I have not really tried.
Is there a specific registration protocol you suggest?
Best, Benjamin
All depends of yoour data, but you could try something like this:
-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 paper and this one for more details.
-Now booth images are aligned, if do you think that the alignement is good, you can perform ACT.
Dear Manuel,
Your post is quite interesting because I am facing the same problem.
Could you please let me know how do you upsample the diffusion data with the same resolution as the structural data?
According to this pipeline (http://mrtrix.readthedocs.io/en/latest/workflows/fixel_based_analysis.html) the command line is
Yes, that’s the command, you can take a look to the options of mrresize and change the interpolation method if you want, I suggest you to use spline interpolation. If your diffusion is 2x2x2 and the structural is 1x1x1, if not first of all you have to resample your structural to 1x1x1.
To align the structural and the diffusion, and at the same time perform for EPI distortion correction, you can do the following (supposing that both modalities are in different space):
Rigid register your masked T1w to your masked B0 (I think is important use the masked version because if not the skull will drive the registration)
Perform the non rigid registration (constraining the direction of you registration to the phase encoding direction, the -g option in ANTs) of the masked B0 to the previously registered T1w. Apply this transformation to all the volumes.
Finally, to get an accurate alignment, non rigid registration between the T1w (the previously registered) to the corrected B0.
I use ANTs for all the process, note that you have to play with the parameters to get an accurate registration of your data.
This process works for my data. There are some papers in the literature explaineng the EPI distortion correction using registration, is not the best method but if you don’t have different encoding directions for the B0…
I wonder if it would make sense to register the T1w to the B0 instead, so that the DWI data is untouched and it can be analyzed in the original space. Perhaps in some applications this would not be a good idea (e.g. if I want to measure tract length because this would be affected by geometric distortions), but I just want to compute a connectivity matrix at the end, so I could non-linearly register the parcellation from the T1w to the DWI and 5TT-segment the Warped T1w to use it for ACT.
The benefit of un-warping the DWI distortions is that you can also modulate the DWI intensity by the Jacobian determinant at each voxel of the warp to correct for signal pile-up/stretch.
Also, by correcting the DWI it should fix any issues where the fibre orientations are inconsistent with the surrounding anatomy (caused by differing degrees of A-P distortions with respect to L-R and I-S) . However, in practise I’m not sure how much it really improves tractography afterwards, because unless you have reverse phase encode pairs for each DW direction, there is no way get back the information lost when the signal is averaged in regions with signal pile up. A quick google returned this article with some relevant citations on the effects of distortion correction on tractography.