nipreps / fmripost-rapidtide

A BIDS App for running rapidtide on a preprocessed fMRI dataset.
https://fmripost-rapidtide.readthedocs.io
Apache License 2.0
1 stars 0 forks source link

Use cranial bone and/or dural sinus signal as global mean mask? #4

Open tsalo opened 2 weeks ago

tsalo commented 2 weeks ago

I was checking out Huber et al. (2024) and I wondered if rapidtide would benefit from directly using cranial bone or dural sinus signals. If so, maybe we need to use a more intensive segmentation method (whether in fMRIPost-rapidtide or by requesting it in sMRIPrep).

bbfrederick commented 2 weeks ago

Historically we've used the superior sagittal sinus to good effect - FSL BET keeps it in the brain mask, although the fmriprep brain mask tends to remove it. This is fine for healthy circulation, but if you have vascular territories with large delays (due to pathology, or even, it seems, normal aging), then you can get multiple copies of the sLFO regressor showing up in the SSS. I refer to this as getting a "poison regressor". This is a problem because I've never found a way to remove these "echo" signals - they tend to make it impossible to disambiguate delay. And the refinement process can't split the signal into components.

tsalo commented 2 weeks ago

Do you think the cranial bone signal would be better than the superior sagittal sinus?

Also, how do you (whether within rapidtide or as a user) isolate the SSS from the brain mask?

bbfrederick commented 1 week ago

I'm not familiar with the cranial bone signal - where would I get that from?

tsalo commented 1 week ago

In Huber et al. they use a "multispectral segmentation" that I'd never heard of. I was hoping you might have heard of it.

AFAICT it's not just the "skull" part of a standard segmentation, as they separate bone from blood vessels. I might ask on Neurostars to see if anyone in the broader community knows how to get that kind of segmentation.