Hello all,
I am using Halfpipe, which is a tool that combines multiple tools under the hood, a key tool of these is fmriprep.
I use it to do preprocessing and extract atlas-based connectivity from rs-fMRI data. the outputs of it include the confounds matrix, the ROI time series and the correlation matrix of these ROIs for each subject.
I want to do Volume censoring, and my question is since I already have the confounds matrix and the ROIs timeseries, can I do volume censoring at this step? e.g., simply by checking the FD values timeseries in the confounds matrix, and simply dropping the indices that correspond to a high FD from the ROIs timeseries.
a relevant question to this is at what step can I include bandpass filtering of the data (since it is commonly used with rs-fMRI), or is the order flexible?
Having both the censored and uncensored ROI timeseries is beneficial for me, since I plan to use the censored ones to compute functional connectivity and use it with ML methods, and use the uncensored (or compare both) with DL methods.
Hello Ammar,
Although I am also a beginner of HALFpipe, I have used fmriprep and Xcp_d to (pre)process rs-fMRI for a long time. As for me, I would censor the dummy scans at the very first step with fslroi command before I run fmriprep or SPM12 (usually 5 volumes) and indicate --dummy-scans 0 to generate confounds and Nifti images. If you calculate the slicetiming and CompCor with all volumes, you will get a wrong confound file polluted by non-stable-scans, and consequently your time series results would be polluted because wrong covs have been controlled in post-processing. However, if the HALFpipe has marked the non-stable-scans, it might be OK. Please refer to these posts the benefits of censoring dummy scans; researchgate for more details and explanation. They also discussed the calculation of confounds in fmriprep after detecting dummy scans, maybe the same in HALFpipe?
Emmmm… Nevertheless, if you want to discard the volumes not at the.beginning, I have no idea. I used to use CONN and it did not allow the user manually discard the volumes FD>0.5, CONN usually regresses it out (scrubbed regressors). I think it is a wise way to regress it rather than directly censor it. On the other hand, in Xcp_d, you can choose to discard the volumes or interpolate, both seem to be reasonable.
When I want to have dummy-scan-censored time series, I would firstly run fslroi to censor the first 5 volumes, and then run fmriprep, and then xcp_d -f 0.5 command (0.01-0.08Hz bandpass filtering should be done here, after preprocessing but before 1st level analysis, which is similar to software CONN based on MATLAB). Bandpass filtering is NOT a flexible but a critical step, please refer to Methods of this article the reason. If you are interested in ReHo and ALFF, this step is particularly essential because rs-fMRI signals out of 0.01~0.1Hz are almost noises.
I can only provide my experience and theoretical framework for dealing with the rs-fMRI data, as for how to manipulate these options on HALFpipe, maybe the developers could help.
Best wishes!
Weissley
