search

DCAN-Labs / abcd-hcp-pipeline

bids application for processing functional MRI data, robust to scanner, acquisition and age variability.
https://hub.docker.com/r/dcanumn/abcd-hcp-pipeline
BSD 3-Clause "New" or "Revised" License
49 stars 19 forks source link

Time-series ready for correlation? #34

Closed mriedel56 closed 3 years ago

mriedel56 commented 3 years ago

We are interested in calculating a dense connectome similar to the HCP. I know those files are quite large, so I imagine thats why they were not shared, but I was wondering if I can calculate it directly from the following file type: sub-#/ses-#/func/sub-#_ses-#_task-(MID|nback|SST|rest)_bold_desc-filtered_timeseries.dtseries.nii

That is, have these been sufficiently processed to the point that they are ready for correlation? Do I need to incorporate motion censoring using the associated files? I dont recall the HCP stance on that.

arueter1 commented 3 years ago

Thanks for reaching out! We will get back to you about this question soon.

ericfeczko commented 3 years ago

Hi @mriedel56 ! You are correct. The sub-#/ses-#/func/sub-#_ses-#_task-(MID|nback|SST|rest)_bold_desc-filtered_timeseries.dtseries.nii is the file type we would recommend for calculating dense connectomes (https://collection3165.readthedocs.io/en/stable/derivatives/). The dense timeseries file does have connectivity BOLD preprocessing applied (https://collection3165.readthedocs.io/en/stable/pipeline/#stage-6-dcanboldprocessing-dbp), however, the data have not been filtered for motion (in case users are interested in examining effects of motion-contaminated data).

You will need to filter/QC the data by motion. The motion_mask.mat contains a series of lists/cells (depending on what software open it, MATLAB vs. R) that contain the temporal masks for a given total framewise displacement (total FD) threshold. Personally, I prefer to use an FD of 0.2, I also recommend using a signal outlier detection to remove additional contaminated frames (which may help reduce the need to go below FD 0.2).

We have tools for calculating dense connectomes if interested, I've linked them below. Feel free to use them :)

https://github.com/DCAN-Labs/cifti-connectivity/ https://github.com/DCAN-Labs/biceps/

Hope this helps! -Eric Feczko

mriedel56 commented 3 years ago

Thanks for the information. Any recommendation on algorithm for signal outlier detection? I think the only one Im familiar with is based in AFNI, and I dont know if it will work on vertex data.

Thanks for the references to calculate dense connectomes! I'll check those out, maybe it will resolve my question above.

ericfeczko commented 3 years ago

Both the cifti-connectivity and biceps incorporate signal outlier detection by default. We recommend using a DVARS or SD based approach to filter excessive frames.

I have a paper coming out shortly on biorxiv with more detail on the method used. Sorry for the delay on this! -Eric Feczko