Hello Anderson,

Thanks a lot for that explanation. I understand it much better now.

Yes, sorry I meant the DMN.

Also, could you please correct me if I am wrong in the following :

Previously, I was using the fslsplit on the dr_stage2_subject_Z images, and then using my randomise result mask with this as input in fslstats.

Now, from what I understood from your explanation, I should use my randomise result as the mask but with dr_stage2_ic in fslstats? How will I get a value per subject then in this case?

What is the difference in these two methods based on the input? Does it make any difference to my connectivity values that I obtain?

I have seen in multiple papers, figures and analysis of barplots with the x-axis-->RSNs, y-axis-->connectivity(z-values) and the barplot showing differences in patients and controls. This is what I am trying to do with my data.

Regards,

Shruti

On 13 June 2016 at 08:10, Anderson M. Winkler <[log in to unmask]> wrote:

Hi Shruti,

On 12 June 2016 at 22:39, Shruti Narasimham <[log in to unmask]> wrote:

Dear Anderson,

Thank you for your help and reply.

Regarding the z-score connectivity values for each subject, how do I solve for example the following :

I have an IC 30 corresponding to the midbrain region and I need to find the z-score connectivity value

Everytime we think of a "z-score" we need to know of what parameter and over what variability. There are multiple ways to compute z-scores that will tell different things. Here, with patients vs. controls, using the dual-regression, perhaps z-scores aren't even needed. More important than ask for z-scores is to ask what is the hypothesis being tested, i.e., what is the research question. Please, see more below:

 

for patients and controls for particularly the precuneus.

Perhaps for the precuneus you are really interested in one of the regions involved in the DMN?

 

How do I do this?

The best way to compare patients vs. controls is probably via dual_regression, which from the first email it seems is already being used, so it's fine. The dual regression will show parts of the brain where patients and controls differ for a given network (IC).

 

Could you please explain me what you mean by using the cluster as a mask?

The results from randomise include various "blobs" (a.k.a., "clusters"), and from the original question it seems you are interested in these. It's possible to take the image that contains the results from randomise and use that as a mask if it contains just one blob. If more than one, then it's possible to use the command "cluster" to label them with different indices, and later use fslmaths to split these apart (options -thr and -uthr used together) to separate these apart, thus producing multiple masks, one fro each region where patients differ from controls.

However, there is no guarantee that, even in the DMN, eventual differences between groups will be within the precuneus or any other part of the DMN. Thus, if the interes in on these regions, perhaps use the precuneus region from the Harvard-Oxford atlas as the mask.

Once a mask has been defined, use the command fslstats to compute the average value for each subject, for the ICs of interest (such as the DMN) from the image that was used in randomise. These results will not be z-scores, but they are the data used for the comparison between groups, and are something that might be interesting to see in barplots (or other types of plots).

All the best,

Anderson

 

How do I obtain this?

Thank you,

Regards,

Shruti

On 11 June 2016 at 07:56, Anderson M. Winkler <[log in to unmask]> wrote:

Hi Shruti,

On 10 June 2016 at 16:01, Shruti Narasimham <[log in to unmask]> wrote:

Hello,

I have performed a Melodic temporal concat on my 3 cohorts. I am investigating rs-fMRI data and probing connectivity changes across my 3 cohorts of participants.

I have two important doubts that I need help with please : 

1. The tcfe-corrp results from randomize output are my results/answers for - regions that show significant differences in FC between groups?
   

Do you mean after dual regression? Yes.

 

1. I need to make bar graphs of group averaged resting state connectivity in the significant clusters to compare across my cohorts. How do I obtain these z-score connectivity values for each subject's region of interest?

It's possible to use the cluster as a binary mask and run fslstats to obtain the average value of that region in the corresponding IC for each subject, then use these to make a bar plot in, e.g., a spreadsheet software, such as LibreOffice or Excel. That said, consider not doing bar plots at all -- use instead boxplots, histograms, violin plots (Hintze & Nelson, 1998), or other methods (see Allen et al, Neuron, 2012).

See this page for an example of how barplots, even with some dispersion measure, can be misleading: https://pagepiccinini.com/2016/02/23/boxplots-vs-barplots/

All the best,

Anderson

 

I'd be grateful for any help regarding this.

Regards,

Shruti