Insight into motion correction

,
1

Hello Neurostars,
I was looking to exclude participants for an fmri study based on a threshold of 1.35 mm of mean frame-wise motion over a particular run.

Though I originally used the output from mriqc this seemed to disagree with a metric that I was getting from MCFLIRT.

Both MRIQC ( “fd_mean” )& fmriprep (“rmsd”) both cite that they implement the algorithm in Jenkinson, et al., 2002. also implemented in fsl’s MCFLIRT.

However, the MRIQC output was about 2 times as large as those from MCFLIRT and taking the mean of rmsd from fmriprep. I didn’t see anywhere in the MRIQC documentation where it explicitly states how it framewise displacement averaged or what units are output, but I assumed it took the mean in millimeters.

image
image806×577 40.7 KB

Any additional insight would be appreciated,

Best,
Jeff D

2

My two cents :wink:

MRIQC can use AFNI 3dvolReg or FSL mcflirt whereas FMRIPREP uses only FSL mcflirt.

The methodology of the processing stages is also quite different between MRIQC and FMRIPREP due to the different goals of the two programs.
Here are some observations from my side regarding the option possible within FMRIPREP of using the SBref image as target for the realignment:

3

Regarding your experiment, how did you choose your theshold in FD? I was wondering was would be the relationship of the FD threshold and the length of the TR for instance.
One might consider a DVARS threshold as an additional threshold for exclusion as it is proposed within FMRIPREP.

Also one technical note: The coil used may have an influence as coil with high receive heterogeneity may overestimate the motion parameters: see here:

Especially in this paragraph:

What if we can’t even attain the physiologic noise-limiting regime? It’s quite possible to be in a subject motion-limiting regime, as anyone who has run an fMRI experiment can attest. In that case, the use of a high dimensional array coil (of 32 channels, say) could actually impose a higher motion sensitivity on the time series than it would have had were it detected by a smaller array coil (of 12 channels, say), due to the greater receive field heterogeneity of the 32-channel coil. This was something a colleague and I considered last year, in an arXiv paper ([1210.3633] A Simulation of the Effects of Receive Field Contrast on Motion-Corrected EPI Time Series) and accompanying blog post. In an e-poster at this year’s ISMRM Annual Meeting (abstract #3352; a PDF of the slides is available via this Dropbox link) we simulated the effects of motion on temporal SNR (tSNR), as well as the potential for spurious correlations in resting-state fMRI, when using a 32-channel coil. In doing these simulations we assumed perfect motion correction yet there were still drastic effects, as the above figure illustrates.

4

Hey @jsein,
Thanks for the input and it’s good to know that mriqc can use either 3dvolReg or mcflirt. I was under the wrong assumption that both were using mcflirt. Also good to know about the single-band reference vs some other reference however I also just called mcflirt on just the _bold.nii.gz data through fsl and my values were more inline with what I got from fmriprep.

Sorry, I was actually I was mistaken about the threshold we chose. We were looking at excluding runs with mean framewise displacement greater than 0.5mm or greater than 1.35 (in our case about 1/2 a voxel width) in absolute head motion from the initial time point. Actually, despite the fact that the measure of mean framewise displacement from mriqc is about 2 times what comes out of fmriprep and mcflirt none are greater than 0.5 so this is more of a learning experience for me.

In our experiment the TR actually varies a bit across some acquisitions that we’re testing through but we have the same exclusion criteria for each run/acquisition.

We are using a headcoil with 64 channels at the moment but had done some piloting with a lower number. I can take a quick look and compare those sets as well to see if that’s a reasonable source of this problem.