How to perform Seed-based FC with the output of xcp_d and the minimal required files from fMRIPrep for xcp-d calculations

Software Support
1

Dear all,

I am currently working on running seed-based functional connectivity (FC) analyses using output files from xcp-d (version 0.7.4) in CIFTI format and would appreciate your guidance.

I’ve reviewed an earlier post and the recommendations in the Nilearn documentation, but they don’t fully address my specific situation. The posts suggest that the seed should be one of the labels from a parcellation scheme. However, in my case, I have created my own seed based on MRS voxels, which I’ve transformed into MNI2009c space using the H5 file from fMRIPrep.

From what I understand, the whole-brain time series should be saved as sub-xx_task-rest_space-fsLR_den-91k_desc-denoised_bold.dtseries.nii in xcp-d. My question is: To run whole-brain FC, should I convert my ROI.nii into a dlabel file to extract its time course? If so, could you provide guidance on how to perform this conversion, especially into this space-fsLR_den-91k space? It seems the mid-thickness files for both hemispheres are needed, but I couldn’t find them in the xcp-d GitHub repository.

If xcp-d or the CIFTI format does not support calculating FC in this manner, I assume I would need to rerun xcp-d without the -cifti option. In that case, could you clarify the minimal required files from fMRIPrep for xcp-d calculations per subject? Given the large sample size I’m working with, the data output from fMRIPrep would be substantial, and I am concerned about the feasibility and time required, which is why I seek your advice.

Lastly, I’ve noticed that, unlike the previous xcpEngine, xcp-d does not appear to have an option for ROI definition. Is this correct?

Thank you for your assistance.

Best regards,

2

Hi @Meng_Li and welcome to neurostars!

I would not recommend using the XCP outputs for seed-based connectivity, since they are already denoised. Typically, denoising and connectivity should be calculated in the same GLM.

You can use any region / ROI definition that fits your hypothesis. As long as your ROI is a binary mask in the same space as your BOLD image, you should be able to fit a nilearn masker without converting into a dlabel. That’s assuming volumetric space, surface would be a bit more complicated.

Correct.

Best,
Steven

3

Hi @Steven,

Thank you for your clarification. I’ll proceed with rerunning xcp_d without the CIFTI option, using the outputs from fMRIPrep and FreeSurfer.

If you happen to know the minimal required files for xcp-d calculations from fMRIPrep, I would appreciate any guidance on that.

Thanks again for your comments.

Best regards,
Meng