Question regarding to the parcellated timeseries in XCP-D for rs-fmri

Software Support
1

Summary of what happened:

Hi all, I am very new to fmri analysis. So I apologise in advance if any of my question sounds silly.

I would like to analyse rs-fmri using Hidden Markov Model, so I used fmirprep + XCP-D pipeline to process my resting fmri data. The desired input for my follow on HMM analysis is parcellated time series. Although I have run everything successfully, but to check I didn’t accidentally screw something up, I compared my parcellated time series with another one of the references that has opened their code and data. I found out the value in my parcellated time series is quite small compared to their parcellated time series input (often ranges from -100 to +100), does that mean I made some error somewhere? I have attached the screenshot of the fmriprep, XCP-D, and parcellated time series. Does this make sense to you?

In addition, what would you normally look for in the qc file to ensure the processing does a good job?

Command used (and if a helper script was used, a link to the helper script or the command generated):

fmriprep code

fmriprep-docker ${PROJECT_DIR}/Nifti ${PROJECT_DIR}/fmriprep-deriv \
 --participant-label 101 \
 --fs-license-file ${PROJECT_DIR}/license.txt \
 --fs-no-reconall
 --stop-on-first-crash

XCP-D code

docker run -it \
 -v ${PROJECT_DIR}/fmriprep-deriv:/fmriprep:ro \
 -v  ${PROJECT_DIR}/tmp:/work:rw \
 -v  ${PROJECT_DIR}/XCP-D-output \
pennlinc/xcp_d:latest \
/fmriprep \
/out \
participant \
--mode linc \
--participant-label 101 \
--input-type fmriprep \
--file-format nifti \
--smoothing 6 \
--despike y \
-p 36P \
--lower-bpf 0.01 \
--upper-bpf 0.15 \
--atlases 4S156Parcels

Environment (Docker)

Screenshots / relevant information:

Rest fmri parcellated time series (My data):

image
image1670×713 447 KB

Reference parcellated time series from published paper, a larger parcellation ROI:

image
image1114×1057 94.6 KB

Report from XCP-D: (Happy to provide more if this is needed)
image

Report from fmriprep: (Happy to provide more if this is needed)

image
image779×443 111 KB

Thank you in advance for the help!

Kind regards
Eric

2

Hi @PSYC_Eric and welcome to neurostars!

The xcpd time series might just be standardized / z-scored. You can always look at the correlation between the two time series, though you shouldn’t expect a perfect correlation due to differences in denoising and nuances introduced by different softwares.

Best,
Steven

3

Hi Steven

Thank you for the reply. I thought xcp-d output the unstandardized time series. Did they mention the output is standardized anywhere in the documentation?

In addition just to clarify, the other time series I provided does not come from the same dataset. It’s just that I have never seen what a processed time series should look like (their given time series is unstandardized), so I just downloaded the processed data provided by one of the papers I refer to (and found this problem). So I think I cannot calculate the correlation between theirs and mine.

And sorry if this sounds silly. But is there a reasonable range for the bold time series? (e.g. it should be 100-ish or something).

Thank you once again!
Eric

4

Hi @PSYC_Eric

No, units are arbitrary, thee is not a necessarily reasonable range.

Best,
Steven

5

Hi @Steven

Great to know that. So should I just treat the output as an already standardized time series and continue with my analysis? How do I know if the output has been standardized or not?

Kind regards
Eric

6

Hi @PSYC_Eric,

The timeseries are mean-centered but not standardized. Standardization is not necessarily required.

Best,
Steven