I am running a resting state analysis in an older clinical sample. For preprocessing, I applied SSWarper and the recon-all pipeline to the anatomical images and stuck closely to example 11 for proc.py. I used slightly higher censor (0.3mm) and outlier fraction (0.10) limits because I am working with a special population.
Rick mentioned in the “Start to Finish Hands-On” video from your YouTube channel that when viewing the radcor images in the HTML output you may see some effects of motion in the radcortcat image but that this motion is hopefully corrected in the radcorvolreg image. When viewing my data, I saw in some participants that the radcortcat image looks just as noisy as the radcortcat image, mostly with voxels rimming the edge of the brain suggesting effects of motion. I’ve attached an image here for reference. For context, the summary motion parameters in this participant are:
average motion per TR = 0.14
average censored motion = 0.13
max motion displacement = 0.59
Is there anything I can do in AFNI to adjust these particularly noisy datasets?
That is true that many of those large radcor magnitudes are outside the brain in both cases. And inside the brain, many of the high magnitudes are tightly located in GM (though also some in local WM and CSF, too).
How does your corr_brain map look? That might also provide a sense of how much global motion is left in, in an important way.
And are the seedbased correlation maps (under vstat) pretty localized?
I wonder if you processed this subject with lower enorm (0.2, say) and outlier frac (0.05, say), does this change things? I believe that you have a tricky population that you are studying, so I understand that you have constraints/tradeoffs possibly forcing the choice of the less strict motion censoring. It is a tricky balance.
Thanks for your reply. Attached are the corrbrain and Vis seed-based correlation maps. The seed-based correlation maps are pretty noisy, though the time series of regions that you would expect to be positively correlated with the seeds are popping up. You can see in the Vis map that the time series of the ventricles are negatively correlated with the seed.
I will rerun the participant using a lower threshold to see if this cleans up the output.
This subject indeed has a fair amount of motion, which is apparent in the volreg radcor images.
Re-running the analysis with more strict censor limits will change those seed-based correlation maps, but it will not alter the radcor ones (censoring is not applied (or even known) for them).
That average censored motion of 0.14 is pretty high. If working with typical motion subjects, we generally drop stubjects if their average censored motion is at least 0.1. If the subjecst are more prone to motion, then one must be more liberal in what is allowed.
Thanks for your reply. I am working with a sample of healthy older adults, MCI, and AD patients. All participants in my sample have less than 21% of total TRs censored and < 1 mm max censored displacement. However, it seems that about 44% of my sample have average censored motion values greater than 0.1. So it seems like there is consistent movement occurring throughout the scan in 44% of participants.
The results of an ANOVA suggest that my groups do not differ on average motion per TR or average censored motion values. Does this suggest that the influence of motion on functional connectivity parameters is equivalent across groups? Are there any other steps in preprocessing you would suggest to run to clean up some of the motion-related noise in the dataset?
Thanks,
Jenna
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.
