Reasonable thresholds for motion censoring in resting-state fMRI preprocessing

Hi all,

We have an in-house pipeline for preprocessing resting-state fMRI data following fMRIPrep, which includes regression of motion and other confounds. I’m looking for guidance on setting some motion-related thresholds in a way that aligns with best practices in the field.

Specifically, the pipeline does the following:

  • Censors volumes based on a user-defined framewise displacement threshold (FD).
  • Discards a run if more than a certain percentage of volumes are censored (percCensorThreshold).
  • Discards an entire session if it contains fewer than a certain number of minutes of uncensored data (minsThreshold).

Could anyone suggest reasonable or commonly used values for these thresholds? Any references or practical considerations would also be appreciated.

Thanks in advance!

Hi @swwalsh1,

There is no standard that the field uses for denoising data, and largely depends on the population you are studying (e.g., more lenient for children / clinical populations), your tolerance for throwing away data (e.g., based on sample size / number of runs per subject), how many temporal degrees of freedom you can afford to lose (based on your acquisition length), and biological construct you are approaching (especially for things that may or may not be related to global signal). I would start by identifying recent high-impact articles looking at similar populations to yours as a baseline.

I personally think that you would want at least 15min of usable resting state data per participant after motion censoring (thresholds based on population). Even better if you can further remove short non-contiguous segments (e.g., if a volume and another volume 6 indexes later are removed, just remove all the volumes inbetween to limit jump discontinuities).

Best,
Steven