A couple of months ago singularity/apptainer had an issue with some NVIDIA libraries. Here is the PR already merged. It has been part of apptainer since version 1.2.5.
Can you make sure you using apptainer/singularity >= 1.2.5 ?
I am a collegue of Arnaud. I will try to help you out while he’s out of office.
What Arnaud’s referring to is not the version of the scilus singularity (1.6.0), but the version of the singularity/apptainer software you are using. If you are using a software version prior to 1.2.5, the local tractography step will fail.
You need to upgrade the version you are using to at least 1.2.5, to have OpenCL work on Nvidia graphics cards, which in turns makes the tractography step work.
I realized the version 1.2.5 I gave you is for apptainer only, it aligns to about version 3.8.5 of Singularity. Thus, your version of Singularity is outdated. You need to upgrade it to at least 3.8.0 to get full GPU capabilities.
We updated our singularity version and the tracking worked. The output does not look good, however. There are spurious and non-anatomical fibers, particularly in the cerebellum, and above the motor and parietal cortices. The reconstructed DWI data looks fine, could this be an issue with the tracking algorithm itself?
The BET step seems not to be perfect but it’s not a big deal. If you use fa tracking especially if you use a low FA threshold you can easily end up with streamlines crossing gyris. We suggest at least a FA threshold of 0.15 if you know what you are looking for otherwise you should be more conservative and go for a WM mask instead.
Which software do you use to visualize your data ?
You can use MI-Brain to dig into your data by creating a small box and explore your data.
If you end up having just spurious streamlines even with more conservatives parameters, I’ll be able to help you if you can send me the bvec that are inside of Eddy_Topup folder and DWI and bval inside Extract_FODF_Shell.