BOLD fMRI is, by definition, measurement of vascular reactivity. If you want to assess the changes from baseline, you would have to somehow assess the baseline state of the vascular system in the brain.

Hi Sumit,

In response to your questions (reordered):

1. Do I have to use any other software other than FSL?

Probably not, unless you want to. FAST is FSL's tissue segmentation tool. In the study you cited, the authors applied FAST to their T1 anatomical image to obtain tissue probability maps of grey matter, white matter, and CSF. They also likely used FLIRT to register their rs-fMRI EPI images to their anatomical and FNIRT to register the anatomical image to MNI space. The FLIRT and FNIRT transformations can be concatenated to transform the EPIs to MNI space. I'd recommend you go through the FSL tutorials to learn how to do this if you are unfamiliar with these techniques. See https://fsl.fmrib.ox.ac.uk/fslcourse/.

3. It is not clear from the paper how the CSF signal was obtained... Does it mean that the CSF signal was obtained after masking all the slices individually that was in the volume and then calculating the average value of the voxels within a CSF mask of each individual slices separately?

Given the tissue probability maps output by FAST, a CSF mask was created of all voxels whose probability of being CSF was > 0.8 (in MNI space). The mask is a 3D binary volume equal to 1 where the CSF is.

4. And how did they obtain the BOLD signal of all the voxels from the CSF mask?

After performing the preprocessing on the fMRI data (i.e., registration, motion correction, detrending, high-pass filtering), the fMRI signals from all CSF voxels from the mask were spatially averaged in 3D on a repetition-by-repetition basis (not slice-by-slice). This can be done with the command fslmeants. This will output a text file where each line is the average fMRI signal intensity from the CSF mask for each repetition in the time series.

Nearly all of these steps can be performed from within the FEAT gui of FSL.

2. The details of how the GLM is designed is not clear and how it can be designed using FSL.

This is assuming you've already done the high-pass filtering, drift detrending, and regressed out the motion parameters. You would have just a single regressor that is the output file from the fslmeants command above. You could load the file into FEAT and run it from there. This will give you the beta outputs (called "cope" file in FSL).

Hope this helps. You could also contact the authors for further clarification.

Marcie Zinn: Typically in fMRI physiology, vascular reactivity is a quantitative parameter describing the signal change to a known stimulus and ideally a stimulus that does not induce neural (de)activation. Since the neural activity is unknown in resting-state studies, it's not really possible to quantify reactivity. The authors have assumed that the CSF signal reflects some systemic physiology that is not neuronal in origin but has a vascular component to it. By using this signal as a regressor, it is possible to quantify the vascular reactivity to some "known" physiological stimulus. In the paper that Sumit has cited, the authors were comparing their proposed CSF regression technique to a hypercapnia challenge, which is a more conventional way of eliciting a vascular response throughout the entire brain with minimal change in neural activity.

Just to clarify how the CSF mask would be generated (Question 3). After running FAST, you would have three "partial volume estimate" outputs (GM, WM, CSF), and the CSF one could be called something like anat_pve_3. To convert this to the binary mask, use fslmaths:

fslmaths anat_pve_3 -thr 0.8 -bin csf_mask

The mask has the name csf_mask.nii(.gz)

Thanks a lot, Avery Berman.

That was really very much helpful indeed. You have almost cleared all my confusion and thanks for giving your time to respond with detailed answers. I will be responding back as soon as I have my results after following the steps provided. I didn't know about fslmeants and I will look at it. However, I also have some doubts and would really appreciate it if you can clarify the following.

1. If the average value of CSF mask is used as the baseline neuronal activity to measure the beta value?

2. I can have two CSF masks, one from the T1 anatomical image and the second from the transformed rs-fMRI to MNI space image. I am confused here which CSF mask to be used to process the data.

Hi Sumit,

I'm not exactly sure what your question is in 1. Could you clarify a bit?

2. The CSF mask should be calculated from the T1 image. That mask needs to get transformed into MNI space (it could alternatively be calculated on the T1 image after being transformed into MNI space). Then you will average the rs-fMRI data in MNI space using the transformed T1 mask.

Avery

Hi Avery Berman

With the details that you provided, I finally mapped the vascular reactivity from the resting-state fMRI data. All thanks to you. In my last question, I was confused about how the T1 CSF signal was used to measure the beta value because the dimensions of my T1 weighted images were very different than my rs-fMRI data. Then I got your point of transforming the T1 weighted CSF signal to MNI space. After getting the CSF signal from the T1 weighted image, I used flirt together with the highres2example_func.mat transformation file in order to transform the mask to rs-fMRI space.

Thanks a lot for the help Avery. Your knowledge was of really great help. I would soon update you with my results.

Ok great, Sumit Swargiary! I'm glad you were able to get things to work. Good luck with the rest of your project!

Avery