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 with different number of iterations (3, 5 and 7) on low b value, low angular resolution data #4

Open archithrajan711 opened 5 years ago

archithrajan711 commented 5 years ago

Dataset Description: 30 direction, single b0 data with b=1000s/mm2, and no reverse phase encoding.

Screenshots of FOD images are attached, with the corresponding b0 slice.

b0 b0_30dir CSF_3_Iterations CSF_3_Iterations

CSF_5_Iterations

CSF_5_Iterations

CSF_7_Iterations

CSF_7_Iterations

GM_3_Iterations

GM_3_Iterations

GM_5_Iterations

GM_5_Iterations

GM_7_Iterations

GM_7_Iterations

WM_3_Iterations

WM_3_Iterations

WM_5_Iterations

WM_5_Iterations

WM_7_Iterations

WM_7_Iterations

Regards, Archith

thijsdhollander commented 5 years ago

Dataset Description: 30 direction, single b0 data with b=1000s/mm2, and no reverse phase encoding.

Nice! Thanks, Archith. This is about as challenging as things (should) get, I reckon: a typical low b-value low angular resolution acquisition with only a single b=0 and nothing else.

Also very interesting to show us the output for 3 (default), 5 and 7 iterations. It should be noted that this isn't a case of "more iterations is always better". There's some additional documentation about it over here for those interested; but the gist is that more iterations will "exchange" more GM-like signal in favour of more WM-like and CSF-like signal. But the results at 3 iterations look very good indeed, considering the data quality. Based on this, I'd very happily advise to stick with the default 3 iterations.

The GM-like signal output looks very much like your typical MSMT-CSD output even, or SS3T-CSD output at a higher b-value. Similar for the CSF-like output, which can e.g. even be turned into a free-water fraction map as Benjamin showed us in #3 . Finally, the WM-like signal map looks very good as well at this b-value. This should help greatly for tractography, registration, template construction and/or (population) fixel segmentation; given what this would otherwise look like at this low b-value (with otherwise high signal in the cortex).

It should be noted that, in terms for fixel-based analysis, it's still generally recommended to have a high b-value for other reasons too: the apparent fibre density metric (amount of WM-like signal in case of 3-tissue CSD) gets more specifically sensitised to intra-axonal signal for increasing b-values. It doesn't mean though that you can't perform FBA at a low b-value: the quality here shows that this is very much feasible indeed. The interpretation of any effect in the result should just be interpreted very cautiously (as does any result of course).

Thanks again, very reassuring results!

Cheers, Thijs

archithrajan711 commented 5 years ago

Dr. Dhollander,

Thanks for the explaination. I hope I can carry forward with 3 iterations for this dataset.

Regards, Archith

thijsdhollander commented 5 years ago

Thanks for the explaination.

No worries at all; thanks for the feedback!

I hope I can carry forward with 3 iterations for this dataset.

Yes, absolutely. At b=1000, it's always going to be a bit of a balanced trade-off in terms of final quality, but the defaults still produce a surprisingly good result. The defaults generally seem to hold up well so far. This is very useful input that'll allow me to fine tune the SS3T-CSD documentation page better in the future.