Hi,

First you should segment your structural volume in three tissues (with FAST in FSL), then you get the mean temporal signal from the WM and CSF masks in the fMRI volume, and now you can include them in a GLM as a regressors.

Good luck :)

As Zeus said you can use a tissue segmentation algorithm such as FAST to get maps of CSF and white matter. I would recommend eroding the masks though to remove voxels on the edge of the mask that may not contain pure white/CSF tissue. You then can use these masks to extract the time series using fslmeants. I would suggest looking at the FSL message board (https://www.jiscmail.ac.uk/cgi-bin/webadmin?A0=FSL) as it is a much better place to ask FSL related questions than Research Gate.

The above answers are correct, but I think a crucial step is missing. Basic steps are:

- segment the structural scan into GM, WM, and CSF (using FAST)

- coregister the segmented images with the functional image (using ApplyXFM, this was missing in the answers above).

- Threshold the segmented images (using fslmaths)

- Extract values from the functional images using the the coregistered, thresholded, segmented image as mask (using fslmeants).

I have attached a script that you may find useful, but you'll have to make some modifications for it to apply to your situation. Also replace the .txt extension with .sh (RG doesn't support .sh files as attachments).

Best of luck!

Peter, thanks a lot for your comprehensive answer. I have fortunately  already managed, but I am sure others will find your comment very useful. have a great day!

Happy to hear you've managed. 

Dear Farzaneh,

unfortunately, I can't help directly, since I am no longer working on the same project and dont have access to the data to compare. Is there a way you can visualize it? may be regress them out and compare pre and post to see if the result makes sense?  Or upload a screenshot and ask the community if it looks all right :)

Best of luck!

Hi Peter, your script was very helpful for CSF and WM segmentation. What threshold would you suggest for GM to apply CSF, WM, and GM timeseries as regressors in a seed based resting state functional connectivity analysis?

Hi Sarah,

That's difficult to say just from my limited experience. I have not thresholded GM before as I only wanted to correct for WM and CSF signal. I guess you need to as GM is an independent variable in your analysis. I have tried different settings for WM and CSF and the higher thresholds (i.e. >.90, high probability of the signal coming from CSF/WM) tend to give better/cleaner results. I also believe this is generally advised as you don't want to explain the dependent signal variance by itself. Perhaps you need to run a few analyses with different thresholds and determine which gives the cleanest results. Although someone else might also have a better theoretically supported answer.

Best, Peter

Hi all, re-opening this discusses after a year! I am using SPM for my fMRI analyses. When extracted the mean signal, are you pulling from the soothed (or unsmoothed?), co-registered functional pre-processed

.nii files? Or modeling the pre-processed functional in their TRs and pulling from the betas?

Depending on your processing pipeline, you may want make sure to use the same filter (e.g. a high pass filter) on the WM and CSF signal that you use on the rest of the brain, to avoid re-introducing noise; I think in SPM this is not a problem given the way the HPF is implemented as hidden columns in the design matrix. Also, if its task-based fMRI, you may need to consider subtracting out the effect of your task design first (because of design correlated head motion)