[log in to unmask] type="cite">

Hi Anderson,

Thanks very much Anderson.

What an idiot I am!

I see. As I put the number "1" in the contrast, so the significant result will be the positive not the neagtive. If I want to get the negtive correlation, I should put the "-1" in the proper site in the contrast.

Am I right?

All the best.

Rujing Zha

 

University of Science and Technology of China

[log in to unmask]

[log in to unmask]

[log in to unmask]

 

2014-01-14

---

charujing123

---

发件人："Anderson M. Winkler" <[log in to unmask]>

发送时间：2014-01-14 20:09

主题：Re: [FSL] glm_output by randomise

收件人："FSL"<[log in to unmask]>

抄送：

 

Hi Rujing,

You don't need to open the beta images to see what is the sign. With the contrasts you are showing, any significant result has a positive regression coefficient. If you wanted to test the negative side, you'd use contrasts with -1 instead of 1.

If, however, you had to open the images with the betas, yes, it would be the 5th volume.

All the best,

Anderson

On 14/01/2014 06:36, charujing123 wrote:

[log in to unmask] type="cite">

Hi Anderson and other FSL experts,

Here is my design matrix, contrast and design.fsf in the attachment.

In my case, I completed the correlation. In the design.mat, there are 5 regressors:sex,age,edu,score_week, and score_year(ordered from left to right).Also I have 5 contrasts,and I found the second contrast(i.e. score_year) has a corrected significant result, so I want to know whether the coefficient of correlation is negtive or positive. So I run randomise --glm_output option, and it generates 5 stats images(i.e. *_glm_pe_tstat1,*_glm_pe_tstat2,*_glm_pe_tstat3,*_glm_pe_tstat4,*_glm_pe_tstat5).

Now I want to know which volume in *_glm_pe_tstat2 I can get the positive or negtive beta value? Maybe the 5th volume in *_glm_pe_tstat2, as the score_year is the 5th regressor in design.mat?

Thanks.

All the best.

Rujing Zha

 

2014-01-14

---

charujing123

---

发件人："Anderson M. Winkler" <[log in to unmask]>

发送时间：2013-12-14 00:10

主题：Re: [FSL] glm_output by randomise

收件人："FSL"<[log in to unmask]>

抄送：

 

Hi Rujing,

Say you have a contrast C = [1 -1 0]', and for a certain voxel, you have beta = [1 4 2]'. The COPE is simply C'*beta = [1 -1 0] * [1 4 2] = 1*1 -1*4 + 0*2 = -3. This is what goes in the numerator of the t statistic.

A suggestion of a book to begin studying is http://www.amazon.co.uk/Handbook-Functional-MRI-Data-Analysis/dp/0521517664

All the best,

Anderson

On 13 December 2013 15:53, Rujing Zha <[log in to unmask]> wrote:

Hi Anderson,

Thanks for your precious help.

I knew the *pe* is what I need to get the beta of regressors.

Most of those you told me I can understand except the *cope*. Would you please introduce me a website or manu describing the "contrasts of parameter estimates" detaily?

Thanks in advance.

All the best.

 

2013-12-13

---

Rujing Zha

---

发件人："Anderson M. Winkler" <[log in to unmask]>

发送时间：2013-12-13 19:18

主题：Re: [FSL] glm_output by randomise

收件人："FSL"<[log in to unmask]>

抄送：

 

Hi Rujing,

Please, see below:

On 13 December 2013 02:42, Rujing Zha <[log in to unmask]> wrote:

Dear all,

I just type "randomise -i GM -o test -m GM_mask -d design.mat -t design.con -n 5000 -T -D --glm_output=GLM_test", but it just generate test_glm_cope_tstat*, test_glm_pe_tstat*, test_glm_sigmasqr_tstat*, test_glm_varcope_tstat* for glm_output.

I have two questions about this:

first: Did GLM_test cannot be used to name the glm output by "--glm_output=GLM_test" or just type "--glm_output" not "--glm_output=GLM_test"?

Use just --glm_output. The output filenames will be prefixed by the string you gave with the option -o (in this case "test"), plus "glm", to indicate that these refer to GLM terms.

 

second: What are meaning of *cope*, *pe*, *sigmasqr*, *varcope*?

These are all the 4 outputs of the --glm_output, and I suspect that if you aren't sure they mean, you probably don't need them. In any case:

- The *pe* are the parameter estimates, also known as betas (this image has as many volumes as regressors in the design);
- The *cope* are the contrasts of parameter estimates, one image per contrast. It's the inner product between the contrast vector and the betas;

- Because the *pe* are estimates, so are the *cope*. These estimates have an uncertainty surrounding them, and the variance around these estimates is the *varcope*;

- The *sigmasqr* is the variance of the residuals.

Hope this helps.

All the best,

Anderson

 

Thanks.

All the best.

 

2013-12-13

---

Rujing Zha