Qasim Bukhari
Research Assistant and Doctoral Candidate
Institute for Biomedical Engineering
ETH and University Zurich
Wolfgang-Pauli-Strasse 27, HIT E22
webpages: MR-page
http://www.biomed.ee.ethz.ch/people/sybukhar
Sorry I wanted to add another point to explain the question properly. So I was referring to the stage2.1. So whether stage2 IC number 0005 is significant or not, provided that stage3 tstat1 is not significant and tstat2 is significant ?And again I m unable to interpret what does significance tells for stage2.2. Does it mean that the regions that are shown up in this particular IC are the significant difference between the two groups activation ??3. stage2 IC is a 4D image, some clusters have high values some have less. If this is my significant difference between the two groups I was testing, what does these high and low values within this image represent since this image itself had been classified as significant through the max value of tstat2.Thanks a lot for any guidancebest,QasimQasim Bukhari
Research Assistant and Doctoral Candidate
Institute for Biomedical Engineering
ETH and University Zurich
Wolfgang-Pauli-Strasse 27, HIT E22
webpages: MR-page
http://www.biomed.ee.ethz.ch/people/sybukhar
Date: Mon, 28 Apr 2014 22:27:31 +0000
From: [log in to unmask]
Subject: Re: [FSL] interpretation of randomize results
To: [log in to unmask]
Hi Anderson,Thanks again for the email.
Here is a real example of the question I actually wanted to ask
dr_stage3_ic0005_tfce_corrp_tstat1.nii.gz; I found the maximum value here as 0.84; thus p is greater than 0.05 that means insignificant.
However for the same IC, but tstat2, i.e.
dr_stage3_ic0005_tfce_corrp_tstat2.nii.gz; I found maximum value as 0.99, that is p is less than 0.05, that means significant.
Now my question is; do I consider this particular IC as a significant result or not ? I m having problems how to interpret these results.
Thanks again
best,
QasimQasim Bukhari
Research Assistant and Doctoral Candidate
Institute for Biomedical Engineering
ETH and University Zurich
Wolfgang-Pauli-Strasse 27, HIT E22
webpages: MR-page
http://www.biomed.ee.ethz.ch/people/sybukhar
Date: Mon, 28 Apr 2014 20:01:01 +0100
From: [log in to unmask]
Subject: Re: [FSL] interpretation of randomize results
To: [log in to unmask]Hi Qasim,2014-04-28 19:23 GMT+01:00 Qasim Bukhari <[log in to unmask]>:
Hi Anderson,Thanks a lot for your answer1. I had actually run randomise with two sample unpaired t-test as I mentioned in my last email too, however I hadnt run F test over it. So do I have to run the whole randomise command with -f <design.fts> option, or is there any other shorter way to directly input thedr_stage3_ic0001_tfce_corrp_tstat2.nii.gz and dr_stage3_ic0001_tfce_corrp_tstat2.nii.gz to compute the F test ???Nope, but there is the option --fonly in randomise that will run just the F-test, and not repeat what you've already run.
Also, note that, because you are using TFCE, the results of the F-test aren't expected to perfectly include what was found with each of the t-tests. Only a two-tailed t-test would work if implemented in a certain way, but I'm afraid this isn't yet available (sorry).
2. I m sorry I still couldnt properly understand the difference between tstat1 and tstat2 file. So each IC has in principle 2 images, why are these two images ? I understand that there are 2 tstats because i defined two contrasts, but I mean why do we come up with 2 values for each IC ?The answer is in the question itself: 2 contrasts = 2 t-stats per IC. The test is run for each IC separately, hence 2 for each.All the best,
Anderson
Thanks againbest,Qasim
Qasim Bukhari
Research Assistant and Doctoral Candidate
Institute for Biomedical Engineering
ETH and University Zurich
Wolfgang-Pauli-Strasse 27, HIT E22
webpages: MR-page
http://www.biomed.ee.ethz.ch/people/sybukhar
Date: Mon, 28 Apr 2014 17:32:53 +0100
From: [log in to unmask]
Subject: Re: [FSL] interpretation of randomize results
To: [log in to unmask]Hi Qasim,
Please see below:On 28 April 2014 12:29, Qasim Bukhari <[log in to unmask]> wrote:
Dear Experts,I successfully ran dual regression and randomize on 70 ICs. I have ran aNow I m trying to see which of the images survived the 0.95 threshold. I had run a two-group unpaired t-test case using design_ttest2 in my randomise. I m now using the following script to check the results quickly.for i in dr_stage3_ic0001_tfce_corrp_tstat2.nii.gz;doecho $i `fslstats $i -R`donehowever I m getting two values as result. One is usually 0 and the other varies in every image.dr_stage3_ic0001_tfce_corrp_tstat2.nii.gz 0.000000 0.8498001. Can someone explain why 1st value is zero and what does it represent ??
The option -R in fslstats shows the range (min and max) of the values in the image.2. Can someone also please tell the significance of the second value and what does it represent ?Randomise saves results as 1-p, so your p-value in this example is 1-0.8498 = 0.1502, so there is nothing significant in this image at the level 0.05.
3. Since I had two groups, and thus two contrasts, so I have tstat1 and tstat2; however the question that I want to answer is whether there are regions/clusters that have different activations in the two groups.. if only one of tstat, for example tstat1 is greater than 0.95 but tstat2 is less than 0.95; then can I say that this IC cluster was the one differentiating the two groups ?
I think you mean that you'd like to compare the two groups for any difference (either group A>B or A<B), is this right? If yes, create an F-test that includes the two contrasts that test the differences in each direction. To be more clear, look at this example in the manual: http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Two-Group_Difference_.28Two-Sample_Unpaired_T-Test.29, then define an F-test that uses both contrasts.
All the best,
Anderson
Thanks a lot for the suggestionsbest,QasimQasim Bukhari
Research Assistant and Doctoral Candidate
Institute for Biomedical Engineering
ETH and University Zurich
Wolfgang-Pauli-Strasse 27, HIT E22
webpages: MR-page
http://www.biomed.ee.ethz.ch/people/sybukhar