I am motion scrubbing fMRI data. I was wondering how conservative are most people with motion scrubbing?
Do most people mark any volumes with striping or only volumes with significant striping?
Do most people additionally mark the volume before and the volume after the volume with motion, or only the specific volume with motion? (Or do they only mark the volume before and/or after only if it shows some motion artefact, no matter how small?)
Are there pros or cons to being more or less conservative from a conceptual/analytical standpoint? Thank you!
Hey Katie-
If you want to be very conservative then you could use the same scrubbing rules as resting state fMRI data (e.g., ~.2mm FDD). Obviously this would decrease your power to detect significant task/group effects, but it would be much harder to criticize your findings as being the result of motion artifact. Also, I might consider using Ted Satterthwaite's 36 degree covariate model (plus additional covariates for spiking head movements):
Satterthwaite, T. D., Elliott, M. A., Gerraty, R. T., Ruparel, K., Loughead, J., Calkins, M. E., … Wolf, D. H. (2013). An improved framework for confound regression and filtering for control of motion artifact in the preprocessing of resting-state functional connectivity data. NeuroImage, 64, 240–256. https://doi.org/10.1016/j.neuroimage.2012.08.052
I hope this helps!
Ben
Hi Ben,
Thanks for the reply! I really appreciate it! :) I will check these out- thank you!
Do you have any specific suggestions for fMRI task-based analyses, particularly for youth with ASD who might have more motion than NT peers? (I know that standards might be different within this context compared to rsfMRI of NT adults.) Originally, we were thinking of using a max percentage of volumes as our cut-off. Our lab has done about 15-20% in the past. We also considered if we should use a more lenient threshold, since we have an adolescent ASD sample, but I am not sure if a significantly higher threshold might be too lenient (e.g., 30% volumes with striping/artefact is 60 volumes). Do you know if there is a standard in the field commonly used with youth samples with developmental disabilities that I can reference? (I know you have a lot of experience imaging youth with ASD and ADHD!)
Relatedly, I have seen that some studies completely exclude (not just scrub) participants if they have too high translations or rotations (e.g., anyone who moves more than Xmm or Xdegrees is excluded). Is this commonly used in the autism developmental literature, and if so, what cut-offs have you seen?
Finally, do you recommend only scrubbing really eggregious volumes or any volumes with hints of striping? (As a side note, I tested out two programs for automated motion detection, one of which was FSL DVARS with default values- I felt like it did a decent job for picking up motion for participants with minor motion, but unfortunately, for those with significant motion, it gravely underestimated the number of effected volumes).
Thanks so much!
Katie
Hi Katie
There are several camps of thought on this issue. Firstly, I would strongly advise not using visual analysis for motion scrubbing as you will not easily detect a large number of problematic volumes. You will need to choose what you use as a marker for scrubbing, commonly DVARS or framewise displacement. (See JD Power, Neuroimage, 2012). Typically, FD will be more sensitive to small motion, but very temporally focused as opposed to DVARS. If using FD, I would definitely consider removing at least 1 volume after, if not 2. DVARS probably 1 is sufficient. The one back removal is typically due to temporal smoothing inherent in rs-fMRI analysis and will be an aggressive approach so you have to consider if you have the data to spare.
Motion "cut offs" are problematic because overall bulk motion is far less of a concern than intra-volume motion or motion between two volumes. (i.e. 1 mm movement from 1 volume to next is going to be a bigger problem than a 1 mm movement as a slow drift over the course of the scan). This is why FD is a more reliable marker.
Also, be weary of "regressing motion parameters." This can be good for resting state data, but can introduce serious problems into task-based data if you have task-correlated motion.
Having said all of this...have you considered other methods, such as ICA denoising and MGTR? These tend to produce better outcomes than does motion scrubbing (see Middlebrooks, et al. 2016, among others). This is true of both resting state and task-based.
Erik
- Badges
- Science topic