GCA, order

Submitted by Yoshinari Abe on Sat, 07/27/2013 - 07:42

 Dear REST experts,

 
 Thank you for your excellent software.
 I have a quetion about REST GCA.
 I set my images (TR = 2000 ms, 194 points) into Input Parameters, set the ROI(ventral striatum) in the voxel-wise manner, select "coefficient-based", and set the order of 1. The x2y maps show some regions, but the y2x maps show nothing else. I wonder if "order of 1" is incorrect. Maybe, should I set the order of 194?
 
Thank you in advance,
 
Yours sincerely
 
Yoshinari Abe

Submitted by jigongjun on Mon, 07/29/2013 - 13:43

Hi, 

I don't think the "order of 1" is incorrect. 
But I am not sure what do you refer to ( “the y2x maps show nothing else” )
Uploading the 'x2y' and 'y2x' one-sample t test map, may be helpful to show your question.

Gong-Jun

Submitted by zangzx on Tue, 07/30/2013 - 17:18

It is hard to define a wrong order. Basically, a 2s-TR is not recommended for voxel-wise GCA. There must be many brain regions that communicate more or less than 2s with vStriatum.
Since there is no way to deal with this problem in fMRI, I suggest you to pre-pick some ROIs and run a ROI-wise GCA, which somehow will reduce the problem.

Submitted by Yoshinari Abe on Fri, 08/09/2013 - 17:11

 Dear REST experts

 
Thank you for your comment.
I uploaded the 'x2y' and 'y2x' one-sample t test maps(figure 1 and 2, respectively). 
In the sentence of ("the y2x maps show nothing else"), I mean that there was no significant brain regions even at relatively liberal threshold such as p < 0.005 uncorrected in the case of y2x analysis.


 
And in addition, I show how I created these maps.
 
<Data acquisition>
3T, TR = 2000 ms, Voxel size = 3*3*3 mm3, 194 points.
<Preprocessing>
First, using SPM 8, I reoriented the structual images into T1 template, and the functional images into T1 template. Then, using your software of DPARSFA, I did following processes; Slice timing correction > Realign to mean > Coregister > Nuisance covariates regression (head motion (Friston 24)、Head motion scrubbing regressors, White mater signal,　CSF signal) >Normalize by DAEREL > Smoothing, FWHM [6 6 6] > Bandpass-filtering (0.01-0.08 Hz).
<GCA>
I computed using REST-GCA. I set the seed ROI in the superior ventral striatum (x=-10, y=15, z=0, radius=3.5 mm) in the voxel-wise manner, select "coefficient-based", and set the order of 1.
<One sample T-test>
I used "ZGCA" files and computed them using SPM 8. The contrast is shown in the figure 3.
 
Which process do you think is incorrect? Please tell me.
 
And, in fact, 2s-TR is too long, but you described a voxel-wise analysis with 2s-TR functional images in your article ("Granger causality analysis implementation on MATLAB: a graphic user interface toolkit for fMRI data processing"). And there is another voxel-wise GCA study with 2s-TR functional images ("Altered Effective Connectivity Network of the Amygdala in Social Anxiety Disorder: A Resting-State fMRI Study / Wei Liao et al"), and in this study they set the seed ROIs in bilateral amygdara.
 
So I have 3 questions now.
First, how long TR do you recommend when making a voxel-wise GCA? About 300 ms as you decribed in your article? 
Second, if the striatum is not suitable for a voxel-wise GCA with 2s-TR functional images, which region is suitable?
Third, up to how many ROIs can we deal when making ROI-wise multivariative GCA. In DPARSFA we can define ROIs (eg; Dosenbach's 160 fuctional ROIs), but can we deal 160 ROIs at the same time?
 
Yours sincerely
 
Yoshinari Abe

 

Submitted by zangzx on Fri, 08/09/2013 - 19:42

Hi, Abe,
I think it is maybe SPM that leads to no results in y2x results. As you know, the SPM t tests is only designed for one-tailed. As far as I know, you will always get a negtive-coefficient map with y2x. If you set the contrast to 1, instead of -1, the SPM will output the significant positive results which obviously show no results for negtive-y2x map. So, I suggest you to set a -1 contrast for y2x. Or, you can use one sample t test in REST (this is 2 tailed condition).
In my paper I did a voxel wise analysis but what I actually think is not perfect. I am still fine with it if I were the reviewer.
You can do a voxel wise GCA with striatum, definately. There is no wrong thing at all, but just not that perfect.
I don't quite say it is good to perform a multivariate GCA with 160 ROIs. I mean to select some regions that you are interested in, based some pre knowledge (papers that report related with striatum like your study). Four or five ROIs is definately better than 160.
Hope you like it.

Submitted by Yoshinari Abe on Wed, 08/14/2013 - 13:01

 Dear Zang

 
Thank you for your comment.
 
Immediately, I again did one sample t-test, using SPM 8 and setting the contrast of -1. But, the result didn't change.

 

Then, I also did it using REST and I don't think the result nii file shows any meaningful outcomes.

 

And I set 3 ROIs, tried  ROI-wise multivariative GCA, then acquired a 3*3 matrix.
 
Could you tell me how to interpret this matrix?
 
Yours sincerely
 
Yoshinari Abe
 

Submitted by zangzx on Thu, 08/15/2013 - 16:02

 Hi Abe,
That might be no results for y2x map.
For ROI multivariate GCA, the output are n*n matrixs, corresponding to each order. Since you set the order at 1, the output is one 3*3 matrix. In the matrix, the (i,j) element is the coefficient GCA results that ROI j effects ROI i. For example, your 7.17730 e-02 means that your 2nd ROI has that many effects on your first ROI. The diagonal value are auto-regression coefficients.