we are using XCP-D vs 0.3.2 with Singularity with default smoothing and everything works as expected, but I am uncertain on how to interpret the parcellated timeseries output. The documentation at General Workflow — xcp_d 0.4.0rc1+7.g636cb62 documentation states:
“Third, XCP-D outputs processed BOLD data, including denoised unsmoothed and smoothed timeseries in MNI2009 and fsLR32k spaces, parcellated time series, functional connectivity matrices, and ALFF and ReHo (smoothed and unsmoothed).”
I do have both denoised_bold.nii.gz and denoisedSmoothed_bold.nii.gz in my outputs, but I can find only one tipe of timeseries.tsv for each parcellation (e.g. sub-…-_atlas-Schaefer1017_timeseries.tsv). Are these timeserieses calculated from the smoothed or from the unsmoothed images?
The parcellated time series come from the unsmoothed images. We wouldn’t recommend parcellating the smoothed data.
On a related note, at the moment, both the denoised data and the parcellated time series contain interpolated data. Starting with the next release, these files will only contain the censored (i.e., low-motion) volumes.
Thank you for your quick answer. May I ask you why you would not recommend parcellating the smoothed data? Isn’t it common practice to smooth the bold images before extracting the regional (parcellation) timeseries?
Thanks a lot.
Best,
Riccardo
Smoothing improves SNR in a given voxel by blurring the signal across voxels, but when you are averaging the signal in an ROI, there’s no reason to do that, since you’re not interested in individual voxels. Smoothing before parcellating could, at worst, blur signals from voxels outside the region into the region.
I also have a question about the parcellation. Although the time series files are all labeled (e.g.,sub-…_atlas-Schaefer117_timeseries.tsv), I’m not well understanding the name. Does atlas schaefer100-17networks parcellation are applied? From the information of the repo CBIG/stable_projects/brain_parcellation/Schaefer2018_LocalGlobal/Parcellations/MNI/fsleyes_lut at master · ThomasYeoLab/CBIG · GitHub , there are two parcellation for schaefer100-17networks ( Schaefer2018_100Parcels_17Networks_order.lut & Schaefer2018_100Parcels_Kong2022_17Networks_order.lut). Which one is the atlas that XCP-D used?
It is the "Schaefer2018_100Parcels_17Networks_order.lut. You can tell by looking at the order of the names in the connectome TSV output. In addition, the citation does not include the Kong 2022 update.
Thank you so much for the quick response! I checked the connectome TSV file (i.e., sub-…_atlas-Schaefer117_measure-pearsoncorrelation_conmat.tsv), but it is a pure 100*100 digital matrix without any ROI information.
Is this the connectome file you refer to?
Where I can find the atlas XCD-P that applied to the imaging data?
Hmm, my version of the file does have ROI information. I ran on version 0.4.0rc1 though, so maybe that feature was added later? Can you confirm you do not have column/row headers like this?
I am sorry to bring this up again. Given the parcellation time series are not smoothed, I am wondering if the correlation matrices are also calculated based on the unsmoothed data or not.
In other words, are there any smoothing procedures done on the parcellated time series? Is it possible?
Correlations are based off of unsmoothed data averaged within parcels. There’s no need to smooth parcellated data, since the spatial averaging within parcels serves the same purpose.
Thanks for the quick reply. Does it mean smooth parameter does not have effect on the functional connectivity matrices? If so, I should have misunderstood the pipeline, although it is more reasonable. And if I just want to calculate the connectivity matrices, I could just turn off the smoothing?
Smoothing does not have an effect on connectivity,. No matter what smoothing value you specify, connectivity matrices are calculated on unsmoothed data.
