Apologiies for posting this again, I think it may have been missed the first time round.
I am working with multishell data with no reversed phase-encoding information, therefore cannot perform fsl topup. Is it still okay for me to use ACT in this instance? (I note there was some mention of this on the afdconnectivity page).
There’s a brief mention in the ACT documentation page, as well as the
afdconnectivity page as you mentioned; but the better reference would probably be the ACT paper itself.
The fundamental assumption underlying ACT is that for any position in space, one can access both orientation information from the DWIs / diffusion model, and tissue information from segmentation of an anatomical image, and that this information corresponds to the same anatomical location. If the DWI EPI distortions are not corrected, this assumption can be grossly incorrect. So no, it’s generally advised against using ACT in this instance. Although
tckgen won’t miraculously crash outright if it detects that your data have not been corrected by
topup, using ACT may give a false perception of the accuracy of the resulting tracks, and could introduce errors that would not have been there if ACT were omitted.
In that case for the following commands would you be able to recommend alternatives to using 5tt and it’s outputs?
Generate response functions
dwi2response msmt_5tt DWI.mif 5tt.mif RF_WM.txt RF_GM.txt RF_CSF.txt -force
create fod - use if running whole brain tractography
dwi2fod msmt_csd DWI.mif RF_WM.txt WM_FODs.mif RF_GM.txt GM.mif RF_CSF.txt CSF.mif -mask nodif_brain_mask.nii -force
#Mask for seeding
5tt2gmwmi 5tt.mif gmwmi.mif -force
Run whole brain tractography - select to retain 5 million tracts sufficient if using sift2 (10M ideal but takes years - could consider for smaller studies
tckgen WM_FODs.mif tracks.tck -act 5tt.mif -backtrack -crop_at_gmwmi -seed_gmwmi gmwmi.mif -select 5M -force
Thanks so much for your help.
That’s an easy one: use
dwi2response dhollander DWI.mif RF_WM.txt RF_GM.txt RF_CSF.txt -force
instead. It’s already more accurate out of the box, even on highly corrected incredible quality HCP data (see, among others, this post), but if you’re facing a scenario where you’ve got no phase-encoding information, this becomes even more of a no-brainer. I’ve seen
msmt_5tt response function outcomes in such scenarios before, and it’s not pretty.
Thanks this works perfect!
Is this okay to use for single shell data as well?
I am also using mouse data for another experiment, is this okay for mouse dwi images?
Yep, no worries. But MSMT-CSD will be limited to 2 tissue types. To get 3 tissue types from such data, you’d need SS3T-CSD, which isn’t available publicly yet.
Sure, no worries again, but you may want to use the
-fa option to
dwi2response dhollander, and provide it with a slightly lower value than the default of 0.2. See this post, where success was shown on rat data: Multi-tissue CSD . The general advise for a mouse would be similar, but you’d probably have to be extra-careful about setting that parameter decently. Set it to a couple of values, and look at the
-voxels output to see what makes sense (or doesn’t).