FCD as fitting target - question about signal quality

[Szymon Tyras]

 Dear colleagues, 

I am currently working on a simulation study and planning to use the FCD matrix as a fitting target. However, I am wondering about the minimal requirements for signal quality concerning motion. 

Specifically, I have tried using criteria such as >0.2 mm of mean FD, >20% of suprathreshold FDs, or any FDs >5 mm, as suggested by Parkes et al. (2017). However, for my dataset, these criteria seem too stringent. Thus, I am looking for less strict criteria that would still make sense. 

I would greatly appreciate any suggestions from the literature or your own experiences with this specific fitting target regarding what criteria I can use. Currently, I have the standard confounds file output from fMRIprep preprocessing.  

Parkes L, Fulcher B, Yücel M, Fornito A. An evaluation of the efficacy, reliability, and sensitivity of motion correction strategies for resting-state functional MRI. Neuroimage. 2018 May 1;171:415-436. doi: 10.1016/j.neuroimage.2017.12.073. Epub 2017 Dec 24. PMID: 29278773.  

Sincerely, 

Szymon Tyras

[Randy McIntosh]

Hi Szymon,

We usually use the voxel size as the criteria to assess motion – about ½ the acquisition voxel size works well.  What should also do is evaluate what the motion effects are. If its abrupt motion from someone shifting in the coil, you may be able to censor and/or interpolate.  If its gradual motion from slow drift in the person’s position, you could use a middle volume for the reference (many programs default to the first volume).

 

For fitting FCD, there isn’t a standard yet but one that seems to work is the variance of FCD in parameter space.  You can see this in Figure 6 of the paper from Petkoski, Ritter, and Jirsa.

--
You received this message because you are subscribed to the Google Groups "TVB Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tvb-users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/612c3bce-a721-4874-8ba9-e5aec60a9e9dn%40googlegroups.com.

[Szymon Tyras]

Dear Randy McIntosh,  

I wanted to express my appreciation for your response; it's been a significant help. 

Since I'm still relatively new to BOLD signal analysis, I've been searching for methods that can work well even with my limited experience. I eventually settled on using fMRIPrep files that are normalized to the MNI152NLin2009cAsym template. I've decided on following criteria for participant exclusion: for each timepoint, I exclude participants if their framewise displacement exceeds 3 mm or if there's a deviation of more than 3 degrees along any axis. It seems that these liberal criteria are commonly applied within my study group (participants diagnosed with schizophrenia). Next I performed rest of processing (parcellation, filtering, FCD calculation). I was wondering if you believe this preprocessing procedure covers the essential basics for conducting FCD-based simulations - it's clear to me that the specifics are highly dependent on a detailed inspection of the data. However, I'm more interested in a general feeling in this context. 

Once again, I'm grateful for your assistance.  

Best regards,

Szymon Tyras

[Randy McIntosh]

Hi Szymon,

 

I think the processing should work. I would suggest looking at both the “static” FC and FCD matrices if practical to see if you have any suggestions of residual motion (e.g., high positive correlations, “banding” in the FC matrices (zero correlations for an entire row&column). You can also correlate the motion parameters with FCD metrics do make sure FCD is not affected.

To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/fe594154-2fdd-4ce7-b7a1-1910a9745493n%40googlegroups.com.

[Szymon Tyras]

Dear Randy McIntosh,  

Thank you once more for your assistance; it has been incredibly beneficial to me. As you recommended, I will manually examine both the static and dynamic FC matrices. Finally, I have processed MNI152 files as follows: I excluded participants with framewise displacement >3 mm or >3 degrees along any axis. Additionally, I regressed out six motion parameters (x, y, z, rot_x, rot_y, rot_z), derived from the confounds file from fMRIPrep, and applied a filter within the 0.01-0.1 range. Following this, I conducted parcellation and FCD calculation. The results, with and without regressing for motion, appear quite similar, but for caution, I chose to retain the regression. I will also incorporate the manual inspection you advised. I hope everything seems appropriate.  

Once again, I am deeply grateful for all support. It is wonderful to have a community that helps entering the field.  

Best regards, 

Szymon Tyras