Dear Matthieu,
you're testing for the same thing but you might get slightly different
results as the two approaches have a different error model: namely
pooled vs partitioned error models. This is described in this document:
http://www.fil.ion.ucl.ac.uk/~wpenny/publications/rik_anova.pdf
Best regards,
Guillaume.
On 24/02/15 15:44, Matthieu Vanhoutte wrote:
> Many thanks Guillaume for this precise help !
>
> Last question about design and results : Should I have the same results
> running a full factorial design with T-contrast [1 -1 0 0 0] and on the
> other hand a two sample t-test with t-contrast [1 -1 0 0] ?
>
> Best regards,
>
> -------------------------------------
> Matthieu Vanhoutte, MSc
> Research Engineer - Department of Neuroradiology
> Regional University Hospital, Lille, France
>
> 2015-02-24 16:34 GMT+01:00 Guillaume Flandin <[log in to unmask]
> <mailto:[log in to unmask]>>:
>
> Dear Mathieu,
>
> t or F contrasts [1 -1 0 0 0] and [1 0 -1 0 0] are fine; you could also
> use a 'two-sample t-test' model for these.
> The input images to a GLM should always be contrast images (summary
> statistics approach); the spmT/spmF are your final results.
>
> Concerning your second question, this is described for example in these
> publications:
> http://www.sciencedirect.com/science/article/pii/S1053811911011906
> http://www.nature.com/jcbfm/journal/v26/n6/full/9600231a.html
>
> > SPM locates significant clusters based on a ratio of signal to noise
> > (a ‘contrast’ of the parameters divided by its standard error)
> > meaning that very low noise regions, for example outside the brain,
> > can attain artefactually high statistical values. Similarly, the
> > commonly applied preprocessing step of Gaussian spatial smoothing can
> > shift the peak statistical significance away from the peak of the
> > contrast and towards regions of lower variance.
>
> Best regards,
> Guillaume.
>
>
> On 24/02/15 15:19, Matthieu Vanhoutte wrote:
> > Dear Guillaume,
> >
> > Thank you for helping !
> >
> > 1) So using this design matrix which contrasts should I use to highlight
> > column1 > column2 and column1 > column3 ?
> >
> > 2) In statistical analysis this mean that I have to use spm* images
> > instead of con* images ? How is it possible than con* images looks so
> > fine and not spm* images (defined largely outside brain and peaks not
> > corresponding with those of con* images) ?
> >
> > Best regards,
> >
> > -------------------------------------
> > Matthieu Vanhoutte, MSc
> > Research Engineer - Department of Neuroradiology
> > Regional University Hospital, Lille, France
> >
> > 2015-02-24 16:00 GMT+01:00 Guillaume Flandin <[log in to unmask] <mailto:[log in to unmask]>
> > <mailto:[log in to unmask] <mailto:[log in to unmask]>>>:
> >
> > Dear Matthieu,
> >
> > your design matrix looks fine, I would just not include the
> last column
> > (constant term). For the main effect of group, you can use an
> > F-contrast: [1 -1 0 0 0; 0 1 -1 0 0] (i.e. diff(eye(3))).
> > Contrast images contain a linear combination of the parameter
> estimates
> > (betas) while the spm* images contain test statistics (t or F)
> assessing
> > the significance of such effects. See this video lecture:
> > http://www.fil.ion.ucl.ac.uk/spm/course/video/#Contrasts
> >
> > Best regards,
> > Guillaume.
> >
> >
> > On 23/02/15 16:29, Matthieu Vanhoutte wrote:
> > > Dear SPM's experts,
> > >
> > > I am studying perfusion images on 3 groups (control, left
> disease,
> > right
> > > disease) and 3 covariates (age, sexe, constant).
> > >
> > > My purpose is to see group effect, control > left and
> control > right
> > > contrasts. According to this aim I chose a full factorial
> design with
> > > one factor (group) that has three levels (control, left
> disease, right
> > > disease). I defined for each subject 3 covariates : sex (0 :
> male, 1 :
> > > female), age and constant (=1). You can find joined to this mail
> > my SPM
> > > design matrix.
> > >
> > > All my perfusion images have been normalized onto a
> particular T1
> > > template different from the T1 SPM canonical template.
> > >
> > > I ran the script that computed automatically contrast for
> this full
> > > factorial design, but don't know which of them are the good
> ones.
> > >
> > > I) Firstly is my design matrix well defined ?
> > >
> > > II) Then based on this design matrix, how should I define the
> > contrasts
> > > to see :
> > > 1) group effect : F-contrast [1 1 1 0 0 0] ?
> > > 2) control > left : T-contrast [1 -1 0 0 0 0] ?
> > > 3) control > right : T-contrast [1 0 -1 0 0 0] ?
> > >
> > > III) Finally, what is the difference between con*.nii an
> spmT*.nii
> > > output images ?
> > > Because I could see superimposed on my particular T1
> template that the
> > > con*.nii image was so good looking and values well defined
> inside the
> > > mask of my T1 template on precised anatomical areas.
> Contrary to this
> > > coherent con*.nii image, the spmT*.nii image was not good
> looking, had
> > > values defined outside the T1 template mask and the values
> defined
> > > inside the mask weren't well located on anatomical areas.
> > >
> > > Many thanks in advance for helping !
> > >
> > > Best regards,
> > >
> > > -------------------------------------
> > > Matthieu Vanhoutte, MSc
> > > Research Engineer - Department of Neuroradiology
> > > Regional University Hospital, Lille, France
> >
> > --
> > Guillaume Flandin, PhD
> > Wellcome Trust Centre for Neuroimaging
> > University College London
> > 12 Queen Square
> > London WC1N 3BG
> >
> >
>
> --
> Guillaume Flandin, PhD
> Wellcome Trust Centre for Neuroimaging
> University College London
> 12 Queen Square
> London WC1N 3BG
>
>
--
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
|