Higher DVARS in XCP-D than fmriprep

Software Support
1

Hi,

I am running xcp-d (0.6.0) on fmriprep (20.2.6) output where ICA AROMA has been run. When checking the xcp-d html outputs, the DVARS values are much higher than those reported in the fmriprep html outputs. For example, one subject’s mean DVARS as calculated by fmriprep is 1.346 where as xcp-d pre-processing mean DVARS is 11.3 and post-processing mean DVARS is 4.3. Additionally, when looking at the carpet plots there is a large spike at the beginning of each plot in xcp-d but not in the fmriprep output. FD values are also different but I did use the --head-radius -auto feature in xcp-d.

I am curious if these discrepancies are problematic or are just due to different DVARS/FD calculations between fmriprep and xcp-d? Below is my fmriprep and xcp-d processing scripts:

xcp-d:

</path/to/xcpd> \
    /path/to/fmriprepdir \
    /path/to/xcpd/output participant   \
    --participant-label <participant> \
    -p aroma_gsr \
    --input-type fmriprep \
    --despike \
    --fd-thresh 0 \
    --head-radius auto \
    --motion-filter-type lp \
    --band-stop-min 6 \
    --task-id rest \
    --work_dir /path/to/workdir>

fmriprep:

</path/to/fmriprep> \
    /path/to/subjects/list \
    /path/to/fmriprep/output/ \
    participant \
    --dummy-scans 0 \
    --output-spaces MNI152NLin6Asym:res-native MNI152NLin2009cAsym:res-native fsaverage5 \
    --cifti-output \
    --use-syn-sdc \
    --ignore fieldmaps \
    --use-aroma \
    --aroma-melodic-dimensionality -200 \
    --fs-subjects-dir path/to/fs/dir \
    -w /path/to/work/dir

I appreciate any help!

Thanks,
Alex

2

There are some differences between how fMRIPrep and XCP-D calculate DVARS. I’m not sure there’s much justification for the XCP-D version.

The fMRIPrep plot uses the standardized DVARS, rather than the raw DVARS that XCP-D uses, though that is not the extent of the differences.

I can look into using fMRIPrep’s/Nipype’s DVARS formula in XCP-D.