So I reran my analysis with my new contrasts added and I ran into some additional issues with one of the newer contrasts. All my previous contrasts and cluster outputs were the same and appear to be correct, however, for one of the newer contrasts (new2-essentially the only one that has significant results at uncorrected p=.001), the cluster command does not appear to be giving the correct output.

Some of the peak t-statistic coordinates seem correct but others point to places outside the brain or do not point to a cluster at all.  It might also be worth mentioning that on the same contrast there is a substantial cluster outside the brain at more liberal uncorrected thresholds (see attached image), which seems strange since this is supposed to masked? Even weirder that the other contrasts seem to be fine.

I've listed the contrasts and the randomise and cluster commands I used below:

1 -1 0 (old1)
1 0 -1 (old2)
0 1 -1 (old3)

-1 1 0 (new1)

-1 0 1 (new2)

0 -1 1 (new3)

randomise -i GM_mod_merg_s4.nii.gz -o GM_mod_merg_s4 -m GM_mask.nii.gz -d design.mat -t design.con -n 5000 -T -V

cluster -i GM_mod_merg_s4_tfce_p_tstat8.nii.gz -t .999 --scalarname="1-p" --mm > T8_clust_s4_.999UCORR.txt

On Thu, Jan 26, 2017 at 9:26 AM, Amanda Rodrigue <[log in to unmask]> wrote:

Thank you very much!

On Thu, Jan 26, 2017 at 4:47 AM, Anderson M. Winkler <[log in to unmask]> wrote:

Hi Amanda,

Please see below:

On 25 January 2017 at 15:34, Amanda Rodrigue <[log in to unmask]> wrote:

Thank you so much for the response. So just to clarify, does that mean by using only the contrast provided say in the first line I am not testing if there are any regions that are larger in group B  than group A. And if I am interested in both directions I would have to add additional contrasts to the originals like below:

1 -1 0
1 0 -1
0 1 -1

-1 1 0

-1 0 1

0 -1 1

Exactly.

 

Also on an unrelated issue, if I wanted to dump the volume values from individual subjects, which file would I get that from? Is it the *GM_to_template_mod.nii.gz?

To give some context I have genotype information for some subjects and want to see if the regions that are significant in my group comparisons differ by genotype within and across groups. I don't have enough power to put the genotypes in the initial analysis so this is more of a pilot thing.

You'd use the same input image that is given to randomise. If using the FSL-VBM pipeline, that would be the file GM_mod_merg_s3.

All the best,

Anderson

 

Thank you!

amanda

On Sat, Jan 21, 2017 at 8:54 AM, Anderson M. Winkler <[log in to unmask]> wrote:

Hi Amanda,

It depends on how the regressors were coded in the design. Take your first contrast for instance: [1 -1 0]. It tests if the regression coefficient for the first regressor is larger than the second, and ignores the third.

Say the coefficients are for group A, group B, and group C. Then another way of saying is that the first contrast is the same as testing:

+1*groupA -1*groupB + 0*groupC > 0

that is the same as:

groupA > groupB.

The same principle goes for the other contrasts.

All the best,

Anderson

On 20 January 2017 at 15:51, Amanda Rodrigue <[log in to unmask]> wrote:

Beginner question here. I've run TFCE based analysis on my VBM and have clusters that differ between my groups.  Can someone confirm how you figure out what the direction of the difference is since the final output is a p-map? For example how do you know whether group 1 >  group2 or the other way around.  My design.con file looks like this:

1 -1 0
1 0 -1
0 1 -1

 I can see the values on the t map but still not sure which group is first when comparing the means (group1-group2 or group2-group1).  Is whatever group that is denoted by the positive 1 the first? This seems to make the most logical sense but just want to check if this is correct. Thanks!
Amanda

--

Amanda Rodrigue

Doctoral Candidate

Franklin Foundation Neuroimaging Fellow

Psychology: Brain and Behavior Sciences

University of Georgia

--

Amanda Rodrigue

Doctoral Candidate

Franklin Foundation Neuroimaging Fellow

Psychology: Brain and Behavior Sciences

University of Georgia

--

Amanda Rodrigue

Doctoral Candidate

Franklin Foundation Neuroimaging Fellow

Psychology: Brain and Behavior Sciences

University of Georgia