search

3Tissue / MRtrix3Tissue

MRtrix3Tissue adds capabilities for 3-Tissue CSD modelling and analysis to a complete version of MRtrix3.
https://3Tissue.github.io
Mozilla Public License 2.0
10 stars 6 forks source link

SS3T-CSD on b=2000 single shell data #2

Open archithrajan711 opened 4 years ago

archithrajan711 commented 4 years ago

SS3T-CSD was applied on a single shell b=2000s/mm2 data with a single b0 and no reverse phase encoded data. The screenshots for FOD images of axial slices are attached below, and compared to White matter and CSF FODs from MSMT-CSD applied on the same data:

b0 b0

SS3T-CSD Grey Matter GM_SS3T-CSD

White Matter WM_SS3T-CSD

CSF CSF_SS3T-CSD

MSMT-CSD White Matter WM_MSMT-CSD

CSF CSF_MSMT-CSD

Regards, Archith

thijsdhollander commented 4 years ago

Hi Archith,

Those results look absolutely excellent! SS3T-CSD was indeed able to filter out the GM from WM/CSF, mostly impacting the WM (and thus FOD) estimation for the better. The b=0 data plays an important role here, so the fact that there's only a single b=0 image is the risk / challenge in this case, I reckon. Your b=0 image does look nice and clean though, no obvious problematic artefacts (of the kind relevant to SS3T-CSD). The preprocessing must've also been able to align it (e.g. motion correction) successfully to the other data as well; that's sometimes also challenging, but it clearly worked well here.

In your data, at b=2000, it looks like the signal in the GM is still slightly larger than that of the WM in the diffusion weighted images; this is apparent from the 2-tissue MSMT-CSD result of the WM, where the cortex is in fact slightly brighter than the WM. This is an extra incentive to want to go for SS3T-CSD of course: you do want to get rid of all that "false positive" WM signal to get a cleaned up WM (FOD) image. SS3T-CSD succeeds well at this here; as apparent from looking at the WM (FOD) images from both techniques side-by-side:

MSMT-CSD SS3T-CSD
WM_MSMT-CSD WM_SS3T-CSD

You can also notice a similar difference between both techniques' CSF ("free water") images, where again 2-tissue MSMT-CSD slightly overestimates it in the cortex, and SS3T-CSD corrects this:

MSMT-CSD SS3T-CSD
CSF_MSMT-CSD CSF_SS3T-CSD

The difference here is visually of course slightly harder to spot, since the CSF image has far larger magnitudes in... the actual CSF, in any case. Indeed, 2-tissue MSMT-CSD ends up modelling the cortex as "a lot of" WM and "a bit of" CSF. 3-tissue SS3T-CSD instead puts most of these parts of the signal in the GM compartment; benefiting both WM and CSF.

Excellent results for b=2000 data with only a single b=0 image! This should come in handy for any subsequent processing, e.g. tractography, registration or template construction, and even fixel segmentation (either on the individual subject or a template in a group analysis).

Thanks again for the feedback!

Cheers, Thijs