AFNI version info (`23.0.07):
Dear AFNI message board:
I am currently analyzing a dataset and experiencing issues with EPI variance line warnings, which I hope you could provide me with some guidance!
The dataset I am working on is an old dataset that was collected about a decade (10~12 years) ago. Both the anatomical and functional data were collected with obliquity. After pre-processing the data through my afni.proc.py script, I am seeing the EPI variance line warnings coming up for almost the majority of the study participants, and across all levels of motion (even among those participants with really low motion). I have attached a few screenshots of the different representative patterns of EPI variance line warnings from five of the participants (attached below). Would you please give me some insights on what does the EPI variance line warning really mean, what factors could have caused these EPI variance line warnings, and provide me with some guidance on what would be the best way to deal with them?
Thank you so much!
code text # or delete if not needed
Variance lines are high variance voxels that span the brain mask from top to bottom. One possible cause of them is a device in the scanner room that is emitting "noise" at a specific frequency. That frequency might translate to a single voxel (or more), per slice. We try to detect this by looking for such high noise at single voxels across all slices, subject to a brain mask.
Only the first of those 5 images shows clearly what what we view as these high variance lines, the solid vertical bars through the brain. The other images are hard to say, and so would not imply any concern.
Does that seem reasonable?
Thank you so much for taking a look at my images and explain what might have caused this EPI variance line warnings! It makes sense!
Since my sample size is small and I am trying to save any data that I can, so I want to ask if you have any guidance on how to deal with this warning? I am doing a whole-brain as well as an ROI analysis, and I wonder if there will be different ways of dealing with this EPI variance issue depending on the analysis?
(For the ROI analysis, should I check the voxels [that cause the EPI variance warnings] based on the coordinates provided by the AFNI output, and make sure that they do not intersect with my selected ROIs? Just a guess!)
Thank you so much!
It is difficult. Note that this is detected in the original space, before even volume registration. Noise at a frequency may correspond to a voxel location and have nothing to do with brain location. While this report helps to detect issues in already-acquired data, it is perhaps more useful in correcting problems at the scanner before too much data has been acquired. For example, this warning might suggest trying to remove an offending device from the scanner room.
Registration transformations would not only blur these lines (as would the blurring step itself :), but they would warp them, usually in a non-linear manner. Dealing with that would be a mess in final space, even for one subject, and it would vary across subjects. Even if the lines are in the same locations in orig space, they would be warped differently due to motion (every time point is warped differently) and all of the other alignment steps.
One can picture ways to (try to) mask this out, or possibly even mitigate the effects, but each would seem to come with its own set of issues to fight with.
So perhaps the conclusion is that if the problem does not seem too bad, just live with it, since these are old scans rather than ongoing ones. If there are just a few bad subjects out of many, perhaps you can afford to drop the ones that might dilute real results. If most of the subjects have this (which would be common if they were all acquired at a single scanner in a small time window), you will have to decide whether the scans are useful.
I know this is not a very useful reply, but such is life. :)
This is actually very helpful for me to learn your perspectives on this warning!
I took some time to go back and recheck the EPI variance warnings. It looks like the first image is the only one so far that I am seeing to have this clear pattern that may imply concern. This image also came from the actual first participant in the study, which I have been struggling with other aspects of the processing as well… I am going to check the rest of my processing outputs and re-do the analysis excluding any participants necessary.
Thank you so much for always helping me understand these processing issues at a conceptual level as well as practical considerations :)