Recommended input and temporal masking for rsHRF

Software Support
,
1

We are setting up an rsHRF workflow and would appreciate guidance on the preferred input strategy when motion censoring is required.
We are considering two options:

  1. fMRIPrep desc-preproc_bold.nii.gz plus --temporal-mask
    The temporal mask would be generated from the fMRIPrep confounds and would exclude volumes with FD > 0.5 mm. rsHRF would then perform its internal filtering and voxel-wise standardization.
  2. XCP-D desc-interpolated_bold.nii.gz plus the same temporal mask
    This signal has already undergone despiking, 36P nuisance regression, interpolation of censored volumes, and 0.01-0.08 Hz filtering. Unlike the fMRIPrep output, it is therefore postprocessed and denoised, but it contains interpolated values at high-motion time points.

The first option preserves the original acquired BOLD signal and avoids interpolation, whereas the second provides more extensive denoising before rsHRF.

Which input would you recommend for HRF estimation? Our current preference is to use fMRIPrep output plus a temporal mask for the primary analysis and potentially use the XCP-D output as a sensitivity/validation analysis. Is this methodologically appropriate?