Processing Limited-FOV BOLD Images

, , ,
1

Hi all,
I’m looking for some advice regarding processing single-subject, non-whole brain functional images.
I ran an experiment to attempt to measure potential BOLD response to TMS at the site of stimulation. To optimize spatial resolution, we used an acquisition FOV limited to the region of interest, obliqued to cover just the scalp and a few inches into the brain underneath the TMS coil.
Because of this limited FOV, I wasn’t sure which standard processing steps would still apply or provide reasonably accurate information.
Here is a example of what I’m working with.

image
image1840×1084 234 KB

Ideally, we would like to perform a “task” analysis on the images to identify BOLD response within a sphere around the stim site (5mm red sphere).
We did one run of each of three stimulation paradigms, all with variable numbers of volumes and acquisition parameters.
In addition to the task images, we also acquired a reverse PE image to perform field map correction, as well as a long-TR whole brain image with the same rotational FOV for assistance in registration.
What is the best way to go about getting the cleanest signal before performing this analysis?
I have used fmriprep in the past, but after an initial pass on this data it seemed to struggle with aligning the slab images accurately to the T1 so I’m concerned about using any of the confound files to denoise the signal. I have also tried using FSL’s tools to manually realign/deoblique and fieldmap-correct the images before pushing them through the GLM, but again, I’m not super confident in the outputs based on the application of techniques and lack of denoising applied.
Any suggestions are greatly appreciated!
Thanks,
Jess

2

What options did you use for fMRIPrep? --bold2anat-init header should help with the BOLD registration by avoiding a bulk alignment before the bbregister refinement.

That said, it is a very small slice. I can imagine there are a lot of things that could go wrong. What does your fieldmap look like?

3

Sorry, I failed to absorb some of this text.

The reverse PE image should be usable with TOPUP, and the whole brain image should be usable to perform the bbregistration if you give it the sbref suffix and otherwise match its name to the BOLD images.

I don’t think this will work well together in fmriprep, because fmriprep will attempt to fieldmap correct your reference image, and there will be a lot of image outside the calculated field map, and extrapolation artifacts are likely to be a problem.

One thought would be to try to use bbregister directly on the whole brain image, and then pass the precomputed coregistration transform to fMRIPrep. (To do this, create an fMRIPrep-like output directory, putting your desired affine transform where fMRIPrep would put its from-boldref_to-anat transform. You’ll need to convert it to ITK using lta_convert.)