Impact of FD and DVARS on preprocessing in fMRIPrep

1

Hello,

I have a question regarding the --fd-spike-threshold and --dvars-spike-threshold options in fMRIPrep. By default, fMRIPrep uses 1.5 for dvars-spike-threshold and 0.5 for fd-spike-threshold to identify motion outliers (spike volumes). I initially thought these thresholds only affected the outlier detection in the confounds.tsv file. However, I noticed that adjusting these values also resulted in different preprocessed T1w and BOLD images in the template space.

Are these thresholds involved in other steps such as head motion correction, co-registration, or normalization? If not, does this imply that the registration results differ every time preprocessing is run? Is there any seed or reference point I could set to ensure the registration results are reproducible? Thank you!!

This is the command I used:

nipreps/fmriprep:23.2.1 /data /out -w /scratch participant \
--participant-label 01 --task-id rest --nprocs 15 --omp-nthreads 12 \
--stop-on-first-crash --ignore slicetiming \
--output-spaces MNI152NLin2009cAsym:res-02 \
--fd-spike-threshold 1.9 --dvars-spike-threshold 3
2

Hi @MaryW,

Motion detection threshold should not impact registration. You can use the --random-seed argument and use single cpus for more consistency in registration between runs on the same data.

Best,
Steven

3

Hi Steven,

Thank you for the clarification. I will rerun the analysis on the same data using a single CPU and the --random-seed argument. I’ll update the thread with the results once they are ready.

Best regards,
Mary