QSIPrep doc suggests dipy recons support determ+prob tractography on DSI data. Is this true?

I have DSI data and I want to produce deterministic and probabilistic tractography.

In the reconstruction workflow page (Reconstruction — qsiprep 0.20.1.dev23+g5433518 documentation), it says:

Option MultiShell DSI DTI Tractography
dsi_studio_gqi Yes Yes Yes Deterministic
dsi_studio_autotrack Yes Yes Yes Deterministic
dipy_mapmri Yes Yes No Both
dipy_3dshore Yes Yes No Both
csdsi_3dshore Yes Yes No Both

Which I took to mean that the dipy and csdsi recon workflows can produce both deterministic and probabilistic tractography outputs. However, in the description of those workflows, it says tractography is all done by DSI studio, which seem to only perform deterministic tractography. Am I misunderstanding the workflow? Is there any way to perform probabilistic tractography on DSI data?

Hi @kjamison , and welcome to Neurostars!

This thead may answer your question: QSIrecon: both: probabalistic and deterministic? - #4 by mattcieslak

Thank you for the reply!

That does help explain what “Both” means, but I’m still not clear on whether it’s possible to do probabilistic tractography on DSI data.

Hi @kjamison, qsiprep definitely supports probabilistic tractography on DSI data. The csdsi workflow will extrapolate the DSI grid scheme to abcd or hcp shelled schemes and then run the mrtrix ifod probabilistic tractography method on your data. Would this work for you? That workflow also does the dsi studio tractography, which is why the docs say “both”

I’ll definitely give that a shot! Do the two dipy workflows do that as well?

They can be modified to do it. I really like visualizing reconstructions in DSI Studio, which also lets you do interactive tractography. All the dipy recon nodes have a "write_fib": true that you can use to get a DSI Studio file that you can look at, or send it to other nodes like autotrack. Or if you prefer to use mrview you can "write_mif": true.

Yes I do like that about DSI Studio, and the speed! But we definitely want both deterministic and probabilistic tractography, and using MRtrix helps us with harmonization with other datasets.

I have a related question: Is there a specific reason you use lmax=4,8,8 for wm,gm,csf in dwi2fod with the csdsi_3dshore pipeline, instead of the default 8,8,8? Is this something related to the 3dshore interpolation?