We are analyzing a task-based experiment at individual level. The task contains 7 runs. When we look at the movement parameters in each run, we have noticed that in some of these runs the dP parameter contains spikes followed by squared and flat waves (see the attached figures). For some unknown reason, this component of motion is not modeled correctly. Which could be the cause? How it can influence subsequent analyses if we introduce these parameters in the 3dDetrend?
After performing the subject level analysis (linear regression), we noticed that the results in those runs were displaced. For instance, activations that normally occurred in fusiform areas now appeared in the cerebellum.
Would it be possible to discard this parameter when doing motion correction?
What do you see in the data along these time intervals, and particularly at the jumps? Please check both the unregistered time series as well as the registered one.
If I look at the time series of one of those runs without any preprocessing, when I pick one voxel within the brain sometimes it looks relatively fine (without spikes). But, if I pick a voxel near to de edge of the the signal it looks very similar to that motion parameter (see image as an example). I also attached a gif showing a video of the time series in which you will notice an abrupt and fast movement on axial view.
As you can see in the example there are some spikes and there is noticeable movement/displacement when we jump between these time points. Then, for some reason at the last abrupt movement the signal stays in the higher displacement without returning to the original position.
This situation happened in a few subjects in different experiment, but acquired with the same MRI machine.
Is it possible that the cause could be an issue in the MRI machine? or could it be an issue at the moment of estimating the motion parameters?
I would like to clarify that this occasionally happen, not always.
This is what I was asking to look for:
“there is noticeable movement/displacement when we jump between these time points”
The motion plots and time series graph plots are just pieces of information. The real picture comes from looking at the data, preferably in all 3 views at once (though the sagittal one is usually the most useful for motion), and see what is happening at those time points.
So looking more closely, is this simply subject motion, or might there be some issue with the scanner? What seemed odd to me is that it looked there was only motion parameter that was highly affected. That smells of an issue with the scanner, as it is almost impossible for a subject to bounce along only one of 6 parameter axes, without influencing the others. And not just once, but repeatedly.
You will have to examine the data very closely to determine such things.
Is this data from a local scanner? Is it still being acquired?
Thank so much for the answer, it makes a lot of sense. Everything indicates that there is a high probability to be a scanner issue.
The scanner is a local one (the one in our University; Phillips Ingenia 3.0) and the data is still being acquired. In your opinion, what could be the best solution to handle this situation? Should we ask for technical support? Or is there a special way to control this motion parameter since this issue is happening in about 3 of every 10 subject.
I would not worry about how to handle it yet. First, try to figure out what is actually happening to the data. For example, for some reason, there is a 1 voxel I/S shift once in a while, possibly corresponding with some subject motion. If you are fairly convinced it is coming from the scanner, or maybe even if you are not quite positive, it might be good to discuss this with a scanner tech.
Note that there are related possibilities, such as the DICOM files getting messed up, or other informational issues.
Thanks for the recommendations. I think there is not a problem with DICOM files in this case. Indeed, the information is the same if I compare the runs which did not have the motion problem with the ones with the motion problem. What can I do to handle this issue? I know that motion in general is a real problem and evidence shows that it can be mitigated until some point (doing alignment and regressing the motion parameters). But, this case is quite special since just one parameter is importantly higher than the orders. Are there any options to solve this?
Karel
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.
