Group analysis

1

I would like to know how to compute group analysis with structural connectomes. I have processed many subjects individually up to the level of their matrix of connectivity. What I don’t know how is to do the statistical analysis?

2

I have been suggested to compute a cell with the desired regions and then perform statistical tests but I don’t know where in the documentation it explains this issue. It would be helpful to know which command or code to use.

3

Hi @andraderenew,

How did you create the connectome? If you are interested in using the SIFT2-weighted connectome for group analysis, then you need to use the same response function to generate the FODs for all your subjects and run mtnormalise after that. If you already did that, then you can use connectomestats to perform your comparison.

Best regards,

Manuel

4

For the connectome I used tcksift2 and also tck2connectome I have applied mtnormalise previously. Then after these steps I will do connectomestats thanks you for the suggestion.

5

Hi.

It is also important to use the same response function for all the subjects.

Best regards,

Manuel

6

for connectomestats the design file and contrast file are in which format? because I have just seen in another thread of this forum what to write in the file although I don’t know what is the format though.

7

I have seen that the contrast and design are text format files. The issue I still have is how to actually write the design and contrast text files.

Let’s say I have one group with 3 subjects.
design
1 0
1 0
1 0
contrast
1 0

would this be correct?

8

Hi,

I’m not an expert on this topic, you would like to double check my answer with somebody else, but I give it a try.

The design you are proposing basically is testing if the mean of the group is greater than 0, and it will always be the case, but this is not informative at all.

Let’s say that you have one group of three subjects with a covariate of age:
design
1 33
1 45
1 27

and you want to test if you metric of interest (e.g. FD) is correlated positively with age, then your contrast will be:
0 1

or negative
0 -1

Does this make sense? I hope this helps.

Best regards,

Manuel

9

thanks so much I will see if it works.

10

I have tried but it only did 6 permutations I wonder if it is due to small sample size N=3 in one sample t test. My next question is how to understand the output. I mean which is the matrix that I can use to evaluate structural connectome of the “group”.

This are the files I have:

connectomebeta0
connectomebeta1
connectomeenchanced
connectomefwe1mpvalue
connectomenullcontributions
connectomenulldist
connectometvalue
connectomecorrectedpvalue
connectomezstat

11

Hi @mblesac, thank you for your previous answers in this thread. I am new to the field and there is not a lot of documentation.

I have a question about the statement “it is also important to use the same response function for all the subjects”. From what I can gather, response functions are estimated using the dwi2response tool, but this tool estimates response functions given an individual dwi dataset. So when you say that it is important to use the “same response function”, do you mean the same approach used to compute the response function? Otherwise, I am unclear as to what input to supply to dwi2response to get a response function that is not subject-specific.

Thanks,
Avanti

12

Hi @Avanti_Shrikumar,

You are right, dwi2response calculates the response for one dwi dataset. What you need to do is calculate the response function for different subjects (this can be all the population or a subset, depending of your population or the number of subjects for example) and then calculate the mean of the response functions using responsemean, then you use the result of this as a input for dwi2fod for all your population.

Technically, it doesn’t have to be the mean, it has to be the same for all subjects, but it makes sense to use the mean. I hope this helps.

Best regards,

Manuel

2 Likes
13

Hi @mblesac ,
I just started learning dMRI and Mrtrix.
I’ve followed Andy’s brain book mrtrix tutorial 8 and your suggestion that use the same response function to generate FODs then create all the subjects’ connectome. (BTW, I use WM group response.)

Can you tell me more about how to compare the connectome between the two group? I mean visualization, or using MATLAB to compare the difference.

Thank you.

Best wish,
Vivian

14

Hi,

It is also good practise to apply mtnormalise to your FODs, in case you didn’t do it. In addition, you might want to use SIFT or SIFT2 before constructing your connectomes.

Once you have your connectomes constructed, there are several approaches you could take, for example use network based statistics, also implemented in MRtrix, so you wouldn’t need matlab for this. Another quite common approach is to compute different global metrics for each connectome, and then compare them between grous (or do any analysis you want), one of the most popular tools for doing this is Brain Connectivity Toolbox in matlab. I hope this helps.

Best regards,

Manuel

15

Thanks for your explination!! I use connectomestats to analyze them and I also use mtnormalise ans SIFT2 before I generate all subjects’ connectomes.
I will try to use mrview to show my connectomestats results.

Best wish,
Vivian

16

Hi Manuel,

Thanks for your guidance on this thread. The connectomestats tool (with the default algorithm) highlighted some edges that show group differences, and now I am curious about how we can visualize these group differences in the streamlines.

I know that connectome2tck can be used to filter the tracks for those that connect specific regions, given that I generated the assignments file at the tck2connectome stage. My thinking is that I could generate filtered tracks for each subject using connectome2tck, transform the tracks for each subject to a common coordinate space using tcktransform, concatenate the transformed filtered tracks across all subjects within each group, and then visualize the two concatenated tracks files in the common coordinate space. Let me know if you see any issues with this approach.

Based on my understanding, though, this approach wouldn’t take advantage of the track weights generated by the tcksift2 algorithm (which I used in order to get my connectome matrices). Do you have suggestions for how to incorporate the tcksift2 information into the visualization?

17

Hi,

I’m not sure why you want to visualize the individual connections (with streamlines), the usual approach to visualize this is using a connectome viewer. But you can of course visualize the individual streamlines. If I had to do this just for visualization, I would use one specific subject.

Your approach seems fine. I don’t know if there is a way in to incorporating the sift2 weights into the visualization.

Best regards,

Manuel

18

Thanks @mblesac. The reason to visualize the individual streamlines would be to get an intuitive sense of “effect size”, and give intuitive meaning to the scale of the values in the connectome matrix.