Dwi2mask error: empty mask

[ThijsDhollander]

@BennyD

Sorry for the slow reply.

No worries!

@BennyD

Yes you’re totally right, if I scroll all the way down to the bottom of the image I see parts of the cerebellum which survived somehow…

Yep, that makes sense! Alright, so now the question is what those parts look like in your DWI dataset that you provides to dwi2mask. They should be very bright (i.e. high in intensity). The other question is if this is due to an artefact, or otherwise. So step one would be to inspect those images (feel free to show a screenshot on here; best would be a sagittal or coronal view of a relevant slice).

@BennyD

Should I not perform the bias correction prior to masking? I followed one of the pipelines from the website but maybe it was outdated?

Not sure which parts of docs you were looking at. The most recent ones, I think, as well as other parts of documentation I have written elsewhere, clarify this somewhat. The gist is: no, you don’t have to perform bias correction at that stage in the pipeline generally, at least when you intend to perform MSMT-CSD or SS3T-CSD later on in your pipeline. This is because you can then apply mtnormalise right after MSMT-CSD or SS3T-CSD, which does the bias correction “instead”, and as a bonus adds the global intensity normalisation you’ll likely also require in a quantitative analysis.

However! If that’s the case, then why is dwibiascorrect even still mentioned at all in the docs, you might ask? Well, because there’s at least 1 step before mtnormalise in the pipeline where dwibiascorrect might help you, but alternatively might also ruin your day. And that step is dwi2mask indeed. Due to how dwi2mask works, it relies on at least having a “not too severe” bias field. In that regard dwibiascorrect might help you, if you do have a bit of a severe bias field. However! dwibiascorrect itself relies on a half-decent mask (which by default it estimates itself). Because of that, dwibiascorrect can sometimes make things far worse, rather than fix them. One of the ways in which it can make them worse, is possibly the exact situation you’re in. So the gist here is: choose to run dwibiascorrect “or not”, depending on how much it helps you to run dwi2mask decently (or not). There’s no other considerations mostly, as for the actual “final” bias field correction, mtnormalise will do the job in principle later on.

So questions for you:

1. Did you run dwibiascorrect before dwi2mask when you encountered this issue?

2. Regardless of question 1, what does the data look like at the point where you provide it to dwi2mask?

I hope this also gives you some insights into what’s going on, and what the relationship between these steps is!

Cheers,
Thijs