Different FDs in MRIQC vs. FMRIPREP


I plan to use FD thresholds to identify outliers in functional runs, and
to eliminate runs entirely when there are too many frames above a given
FD threshold.

While comparing the FD calculations in an MRIQC individual report to those
reported in the fmriprep confound file, we get differing results, but
not in a systematic way. The two FD calculations are related, but the
fmriprep FD value for a frame will be slightly higher and other times
slightly lower than MRIQC’s FD value.

(Comparisons were made between FD column in the fmriprep confounds file and the FD
vector that is plotted in the MRIQC report, with values found in the
fd_power_2012.txt file inside the MRIQC working directory).

Does anyone know why this occurs? Is this a bug in one program or the other? Is one report more accurate?


1 Like

MRIQC uses AFNI 3dVolReg to estimate motion (which in theory is slightly more accurate according to this paper from 2005 http://ece-research.unm.edu/vcalhoun/courses/fMRI_Spring10/Oakes_motion_correction_NeuroImage.pdf) while FMRIPREP uses FSL MCFLIRT (which allows us to combine motion correction affine matrices with other coregistration in a single shot interpolation).

In practice I don’t expect the discrepancies to be significant.

Whilst here, we’ve found that in a sample of ~300 kids with generally high motion that the FD statistic from mriqc is 10% higher than the FD statistic from fmriprep