Hi Anderson,
Thank you very much for your help !
Best regards,
David Romascano
________________________________________
From: FSL - FMRIB's Software Library [[log in to unmask]] on behalf of Anderson M. Winkler [[log in to unmask]]
Sent: Saturday, November 29, 2014 8:05 PM
To: [log in to unmask]
Subject: Re: [FSL] Randomise : Two-sample unpaired t-test with covariates
Hi David,
The design looks fine (both). The message isn't an error, but just a warning. It appears more often when there's a very large effect, and common causes are contrasts testing whether the amount of GM or FA is different than zero (which is almost always the case in the areas of interest). In these cases, the warning raises suspicion that maybe some contrasts are testing things that aren't really useful.
If your data looks correct (i.e., you did check them carefully before starting randomise, to make sure there were no errors in the preprocessing), and the outputs also look sensible, then I'd say it's fine.
All the best,
Anderson
On 28 November 2014 at 16:48, Romascano David <[log in to unmask]<mailto:[log in to unmask]>> wrote:
Hi Anderson,
Thanks for your reply. I indeed only want to test for group differences. You can find the contrast and design files attached. I call randomise with the following statement :
randomise -i merged_data.nii -o TFCE -d ./Glm/design.mat -t ./Glm/design.con -m master_mask.nii -T -n 10000
I get the following messages in the terminal :
Loading Data:
Data loaded
4.2672e+13 permutations required for exhaustive test of t-test 1
Doing 10000 random permutations
Starting permutation 1 (Unpermuted data)
Starting permutation 2
Starting permutation 3
...
Starting permutation 44
Warning: tfce has detected a large number of integral steps. This operation may require a great deal of time to complete.
Starting permutation 45
...
Starting permutation 76
Warning: tfce has detected a large number of integral steps. This operation may require a great deal of time to complete.
Starting permutation 77
...
Starting permutation 80
Warning: tfce has detected a large number of integral steps. This operation may require a great deal of time to complete.
Starting permutation 81
...
Starting permutation 93
Warning: tfce has detected a large number of integral steps. This operation may require a great deal of time to complete.
Starting permutation 94
...
Starting permutation 110
Warning: tfce has detected a large number of integral steps. This operation may require a great deal of time to complete.
Starting permutation 111
...
I thought the warnings didn't show when running the non-demeaned design matrix (see attached design_no_demean.mat), but actually they do... Sorry for misleading you.
The design and contrast matrices are good then ?
Thanks a lot,
David
________________________________________
From: FSL - FMRIB's Software Library [[log in to unmask]<mailto:[log in to unmask]>] on behalf of Anderson M. Winkler [[log in to unmask]<mailto:[log in to unmask]>]
Sent: Friday, November 28, 2014 3:29 PM
To: [log in to unmask]<mailto:[log in to unmask]>
Subject: Re: [FSL] Randomise : Two-sample unpaired t-test with covariates
Hi David,
As shown, your desing is correct, although it's not testing any interaction, just the group difference after adjusting for age and sex. For the two contrasts shown, that simply compare groups, mean-centering age and sex doesn't make any difference, unless there's some other contrast (not shown) that is testing whether the mean of either group would be zero (this would explain the warning message and, for these other contrasts, mean centering impacts the result).
Could you please paste here the exact call to randomise that you are using? And could you also attach the very same design files that are used in this call (that is, design.mat and design.con)?
All the best,
Anderson
On 27 November 2014 at 11:19, Romascano David <[log in to unmask]<mailto:[log in to unmask]><mailto:[log in to unmask]<mailto:[log in to unmask]>>> wrote:
Dear FSL Users,
I'm wondering on the proper way to generate the design and contrast matrix to perform a Two-sample unpaired t-test with covariates using randomise.
I have 52 subjects (18 controls and 34 patients), and I want to test for differences between groups, while controling for age and gender.
I merged the 52 volumes in a single file (the first 18 volumes are the controls and the rest is the patients).
I've read here (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Two_Groups_with_continuous_covariate_interaction) that the covariates should be demeaned. I've created the following design.mat file, where the 1st column describes the group effect, the second the demeaned age and the 3rd the demeaned gender. The demeaned age is the subject's age minus the mean age of all subjects. The gender is 1 for males, 0 for females. As I have 20 males, the mean gender is 0.384615384615, which I subtract to each subject's value:
/NumWaves 4
/NumPoints 52
/PPheights 1.000000e+00 1.000000e+00
/Matrix
1 0 -5.51923076923 0.615384615385
1 0 -13.5192307692 0.615384615385
...
1 0 -9.51923076923 -0.384615384615
1 0 0.480769230769 -0.384615384615
0 1 -1.51923076923 -0.384615384615
0 1 2.48076923077 -0.384615384615
...
0 1 14.4807692308 -0.384615384615
0 1 -12.5192307692 0.615384615385
And the following design.con file to test for differences between groups:
/ContrastName1 controls > patients
/ContrastName2 controls < patients
/NumWaves 4
/NumContrasts 2
/PPheights 1.000000e+00 1.000000e+00
/Matrix
1 -1 0 0
-1 1 0 0
Is this correct ? Am I missing something with the files or the volumes ? All volumes were registered to the MNI space before merging all subjects.
If I run TFCE, I get a warning because "tfce has detected a large number of integral steps. This operation may require a great deal of time to complete". I've read that this might be due to an error in the design matrix.
TFCE runs without warning if I ***don't*** demean the age and gender columns (ie I put the subject's real age and 0 or 1 for the gender). Would this be correct even if the GLM guide states that covariates should be demeaned ?
There is also the -D option in randomise. Should I use it if I use a design matrix without demeaning the covariates ?
Thanks for your help !
David
|
