Print

Print
Dear Arnaud,

If you use a design that includes two volumes per subject then you must have a large design matrix with one EV to represent the difference and one EV per subject (to represent the mean of each subject).  It is possible to include covariates in this kind of model, but a lot easier in the simpler case where you create the post-pre images beforehand and have a smaller, simpler design matrix.  If you really want to add covariates to the large matrix then they need to occur in positive/negative pairs for each subject, so that they are independent of the subject means and represent values that would explain the post-pre differences.

All the best,
Mark

On 13 Feb 2014, at 13:12, Arnaud Boré <[log in to unmask]> wrote:

Thank you Mark.

If I choose  to work with the first design proposed by Jared (with 2 volumes one for pre and one for post) I think I can't just have one post-pre difference covariable column (EV). Am I right ? If so, how can I put each value (pre and post depending of the subject) ?

Thank you very much for your help

Arnaud


2014-02-13 3:04 GMT-05:00 Mark Jenkinson <[log in to unmask]>:
Hi,

If you calculate the post-pre differences using fslmaths, as Gwen suggested and is on the wiki, then you can look for relationships with a covariate by either (a) having a single column (EV) of the covariate values and using the -D option in randomise to remove the mean from the data and the covariate values, or (b) have two columns (EVs) where one is a column of all ones and the other contains the demeaned covariate values.  In the latter case you must *not* use the -D option, and you are then able to have separate contrasts that look at the pre-post difference (e.g. contrast of [1 0] or [-1 0]) or a contrast that looks at the relationship with the covariate (e.g. [0 1] or [0 -1]).  Both designs will give completely equivalent results for the covariate tests.

All the best,
Mark


On 12 Feb 2014, at 20:21, Arnaud Boré <[log in to unmask]> wrote:

Hi,

Just wanted to know if I try to see the differences pre and post like Jared asked but looking at regression with a covariable. Do I put this covariable on just one column and do I need to demeaned this variable ?

Thank you for your help

Arnaud


2014-02-11 7:17 GMT-05:00 Gwenaëlle DOUAUD <[log in to unmask]>:
Hi Jared,

just a quick tip: while your design is correct, one very simple design to do the same thing would be to take the TBSS (or VBM for that matter) difference between pre and post for each subject and then simply set up a two-tailed t-test (then no need for a design.grp and complex design.mat/con).

This is also what we recommend doing in randomise when you have more than one group and 2 timepoints.

Cheers,
Gwenaëlle
 
------------------------------------------------------------------------------------------------------------------------
Gwenaëlle Douaud, PhD
University Research Lecturer & MRC Career Development Fellow
FMRIB Centre, University of Oxford
John Radcliffe Hospital, Headington
OX3 9DU Oxford UK
Switchboard: +44 (0) 1865 222 493
Fax: +44 (0) 1865 222 717
www.fmrib.ox.ac.uk/team/principal-investigators/gwenaelle-douaud
www.fmrib.ox.ac.uk/research/fmrib-interface-analysis-clinical-neuroscience-group
-----------------------------------------------------------------------------------------------------------------------

De : Jared Tanner <[log in to unmask]>
À : [log in to unmask]
Envoyé le : Lundi 10 février 2014 20h35
Objet : [FSL] Longitudinal MRI TBSS matrix design

Hi FSL team,

I wanted to verify that I had correctly set up matrices and contrasts for a 2 time point longitudinal MRI (10 subjects, repeated once). I used glm_gui to create the design, hopefully as a paired t-test.

design.mat

/NumWaves 11
/NumPoints 20
/PPheights 2.000000e+00 1.000000e+00 1.000000e+00 1.000000e+00 1.000000e+00 1.000000e+00 1.000000e+00 1.000000e+00 1.000000e+00 1.000000e+00 1.000000e+00

/Matrix
1.000000e+00 1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00
1.000000e+00 0.000000e+00 1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00
1.000000e+00 0.000000e+00 0.000000e+00 1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00
1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00
1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00
1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00
1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00
1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 1.000000e+00 0.000000e+00 0.000000e+00
1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 1.000000e+00 0.000000e+00
1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 1.000000e+00
-1.000000e+00 1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00
-1.000000e+00 0.000000e+00 1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00
-1.000000e+00 0.000000e+00 0.000000e+00 1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00
-1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00
-1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00
-1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00
-1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00
-1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 1.000000e+00 0.000000e+00 0.000000e+00
-1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 1.000000e+00 0.000000e+00
-1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 1.000000e+00


design.con

/ContrastName1 condition Pre > Post 
/ContrastName2 condition Post > Pre
/NumWaves 11
/NumContrasts 2
/PPheights 2.000000e+00 2.000000e+00
/RequiredEffect 4.414 4.414

/Matrix
1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 
-1.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00 

design.grp

/NumWaves 1
/NumPoints 20

/Matrix
1
2
3
4
5
6
7
8
9
10
1
2
3
4
5
6
7
8
9
10

The randomize command I ran was: randomise -i all_FA_skeletonised -o tbss -m mean_FA_skeleton_mask -d design.mat -t design.con -e design.grp --T2

Does this design and this randomise command seem accurate?

Thank you for any help anyone can provide,

Jared Tanner





--
Arnaud BORE
Research assistant 
Cellulaire : (001) 514-647-8649





--
Arnaud BORE
Research assistant 
Cellulaire : (001) 514-647-8649