[Fmriprep] - Issue with using initial brain mask for T1w --> Standard transform for SyN-SDC?


Hi Neurostars community and fmriprep devs,

I’ve been trying to understand the workflow implemented by fmriprep as we’ve been running into some issues relating to signal from the dura and coil-induced inhomogeneity causing bad initial masks. As a result, I’ve noticed a small detail that I wish to understand better, hoping you can help me out with this:

In Fmriprep (v1.4.0 and below) the T1w --> Standard transform is derived using the antsBrainExtraction workflow which as noted under https://fmriprep.readthedocs.io/en/stable/workflows.html#refinement-of-the-brain-mask may contain inaccuracies due to external MR signal. I’ve noticed that in the data I’ve worked with that the “Moving Image” used in the antsRegistration phase sometimes excludes brain or includes external regions. As a result the T1w --> Standard transform may be sub-optimal.

From my understanding this T1w --> Standard transform is subsequently used for SyN-SDC (please correct me if I’m wrong here). Is there a reason why the T1w --> Standard transform isn’t re-computed after Freesurfer segmentation and mask refinement? Wouldn’t that produce a better nonlinear warp to use in the SyN-SDC workflow? As mentioned above, the initial mask from antsBrainExtraction sometimes includes non-brain matter or clips parts of the cortex, I can imagine that in these cases using the Freesurfer mask for T1w --> Standard would be desirable.

Would love to hear your thoughts,
Jerry Jeyachandra