I’m running a single-shell fixel based analysis pipeline (following the relevant steps on the mrtrix documentation) on a dataset of around 300 or so images with 42 directions (b=1250).
Upon inspecting the FODs created by dwi2fod, there seems to be unusually bright GM voxels in the cortices, whilst white matter in the splenium of the CC is darker when viewed in mrview; see Figure 1 as an example.
By overlaying the FOD lobes onto the image you can see that although the white matter is distinctly anisotropic, the gray matter voxels seem to be isotropic but with relatively large FOD amplitudes. The csf seems to be normal; see Figure 2.
We’ve checked the response files for the WM, which look fine; see Figure 3.
Have you encountered this kind of problem before, or do you have any insight as to what might cause this?
Thanks in advance,
This is expected: at this low b-value the signal intensity in GM exceeds that of WM, resulting in larger apparent fiber densities in GM than in WM
Just to add to this:
You’ll find that the AFD in GM will be lower than in WM using higher b-values (nearer 3,000s/mm²), and these noisy lobes in the GM would be attenuated – another good reason to go for high b
However, you might consider the use of a 2-tissue multi-tissue CSD approach here – we’re starting to recommend that for all single-shell analyses, definitely worth a try. General instructions can be found here – only change is that in the
dwi2fod call, you should only pass the WM and CSF arguments (one for the input response, one for the output ODF image, for each tissue type), and skip the GM ones. That should already attenuate the problem to some extent (although not completely).
going to high b is something we have difficulty to accept … the old habit …
Just a precision about multi-tissue CSD for single shell (which is 2 shell with the B0)
The advise is to run dwi2fod with WM and CSF. But the dwi2response have to run with the 3 argument (wm gm csf).
Talking with Thijs at ismrm I understood there was an other script (still not release) to deal with muti-tissue estimation of single shell acquisition.
Personally I used to run dwi2fod with the 3 response function even though I can see for the tissue map that only 2 tissue has been detected. It may not make a big difference on the CSD_wm … ?
It’s not per se a “have to” for
dwi2response, it’s just what those algorithms give you automatically (WM, GM and CSF response functions). But from an “interface” point of view, yes, you’ll always get all 3 response functions returned. For
dwi2fod with the current
msmt_csd algorithm, yes, 2 b-values (i.e. single-shell data with b=0 images) will only give you 2 tissue types. In that scenario, the natural thing to do is 2-tissue CSD with WM and CSF; to at least cover a wider range of diffusivities (compared to e.g. using a WM-GM model without CSF; which I definitely wouldn’t recommend).
Yep, exactly. That would be “single-shell 3-tissue CSD”, or SS3T-CSD for short. We’re using it over here on all our single-shell data (which is, mostly, all our data) and in several collaborations with people in the same scenario because of a range of reasons. At the moment it’s only an internal prototype; but a full version will be released at some point in MRtrix3, and along with it, or eventually after it, a further range of tools to deal with many “3-tissue” things. I’ve had successful results using lower b-values as well, and in a range of applications. See some links to ResearchGate stuff in this post as well (at the end of it): Low b-values with increasing and decreasing response function magnitude - #6 by ThijsDhollander .
Yep, depending on your response functions, you’ll almost certainly get just a WM-CSF fit, with the GM inherently not being used in the fit. So you’d probably describe that just as well as 2-tissue CSD. Be careful though: if you just keep on using the “3” outputs from this (including the GM which is, apart from numerical inaccuracies, essentially zero), you can easily run into other problems. That would for instance be the case if you perform
mtnormalise on this output: the close-to-zero GM image may/will cause issues there.
mtnormalise works reasonably well on 2-tissue results, but you’ll have to explicitly only feed it the 2 tissue types (WM and CSF). Hence, to avoid these kinds of issues even getting an (accidental) chance of happening, I’d recommend to simply also perform the
dwi2fod with only 2 tissue types; for the sake of clarity and self-documentation of your pipeline.