Gradient artifacts, high motion pediatric data 3dDespike

Hi all,

We are preprocessing fMRI data from young children (ages 3–5 years old) and are facing two compounding data quality issues:

  1. High motion — as expected in this age group

  2. Gradient artifacts — some scans were affected by scanner gradient issues, introducing large transient signal spikes in a subset of volumes

We are considering using AFNI’s 3dDespike to handle the spike artifacts prior to or after running fMRIPrep.

Our specific questions are:

  • Should 3dDespike be applied before fMRIPrep (i.e., on raw BIDS data) or after fMRIPrep (i.e., on the preprocessed output)?

  • Are there known interactions between despiking and fMRIPrep’s internal steps (e.g., slice timing correction, motion correction, susceptibility distortion correction) that would make the order of operations important?

  • Given that our spikes are likely gradient-induced rather than purely motion-induced, does 3dDespike adequately handle this type of artifact.

  • If we do fMRIprep preprocessing after running 3dDespike on all the data - we will get .tsv files with stdDVARS and FD output; should we consider scrubbing/censoring affected volumes (with threshold of FD>1.5MM and stdDVARS>1.5mm?

  • If I run 3ddespike after pre-processing, then how do I decide on scrubbing?

  • Is there a risk that desdespiking before fMRIPrep could interfere with fMRIPrep’s motion estimates derived from the raw data?

  • Any guidance on the recommended order of operations, or experience with similar pediatric/high-motion datasets affected by gradient artifacts, would be greatly appreciated.

Hi @Avantika_Mathur,

3dDespike should be applied after fMRIPrep.

3dDespike is agnostic as to what the source of artifact is, it cares more about the shape of the artifact. I think it will more target motion though.

That is a complicated question that is dependent on how much data you collect, the nature of motion, and what kind of analyses you want to run. Emerging evidence suggests that censoring is not recommended: [2603.07380] Excessive data censoring in fMRI undermines individual precision and weakens brain-behavior associations

Best,
Steven

Hi Steven, thanks for the helpful answers. This is task based fMRI data being collected on two tasks - rhyming and meaning judgement. Each task has two runs and the paradigm is block design paradigm (with block being 54.95 sec long). We are mostly interested in hypothesis driven ROI - based analysis or performing brain - behavior correlation (within ROIs and whole brain). In the past on different projects when we used SPM pre-processing, followed by art-repair (we have set thresholds to 1.5 mm for scan to scan movement and 4 % to Global signal for interpolation and not more than 10% volumes interpolated for a run to be included in the analysis). Question: If I am running 3ddespike after preprocessing, how do I know how much data is interpolated in a run.

Hi @Avantika_Mathur,

You can get the “spikiness” map from the -ssave argument. By default, values with spikiness > 2.5 are interpolated.

Best,

Steven