Hi,
> For structural VBM analysis, I was told to use
> randomise for group analysis. Then my quesion
> is, is randomize speicial for VBM or a general method
> that can also be applied on fMRI cope images
> and DTI FA images?
It is highly recommended for DTI images (see the TBSS manual), while you should use FEAT for fMRI images.
> My second quesion is, can FEAT also
> be used to do group analysis on structural and DTI
> data?
It could but it is *not* recommended and should anyway be used on smoothed data.
Cheers,
Gwenaëlle
I tried on vbm data, and failed with feat
> Higher-level analysis, But succeed with feat
> first-level analysis. Below is details, but why
> error with higher-level? and is it proper to
> use first-level?
> FEAT details:
> I was thinking of using FEAT higher-level
> analysis but failed. I selected "Inputs are 3D cope
>
> images from FEAT directories" and specified processed 3d images of every subject
> (*_GM_to_template_GM_mod.nii.gz), then specify the design metrix. I thought it will treat all
> images as COPEs, but I got this error message: "registration has not been run for all of the FEAT directories that you have selected for group
> analysis"
> Then I merged all these 3d images into a 4D data and try first-level analysis, I turned off all options in pre-stats tab (I am not sure about how to treat "high pass filter" in data tab)
> and specified a design matrix without convolution. I got a result but I am not sure is this correct or not.
>
> Thanks!
>
>
> --
>
> Chunhui Chen
> 在2010-01-08 10:14:20,chen <[log in to unmask]>
> 写道:
>
>
> Thanks Gwenaëlle, and also those who pay attention to
> this quesion!
> Since randomise is the best choice for VBM, I
> would do it this way.
>
> Another reason for my trying FEAT is, I have
> data of other modality like DTI FA, and I would like to
> model all these data with behavior index across subjects, so
> I would like to find a way to do second level analysis for
> all these data. I assume FEAT would do it well, but I am not
> sure why I got error with Higher-level analysis, and not
> sure the way I use first-level analysis is correct or
> not.
> A related question is, how can I calculate
> correlation bettewn VBM danta (or DTI FA) and behavior
> index across subjects? and will the correlation differ from
> FEAT results?
>
> Best wishes!
>
> --
>
> Chunhui Chen
> 在2010-01-08 01:51:30,"Gwenaëlle DOUAUD" <[log in to unmask]> 写道:
> >Hi,
> >
> >we strongly recommend the use of randomise (non-parametric inference) for plenty of different reasons including the possible non-gaussianity and non-stationarity of your data. Having said that, the argument that might convince you is that it tends to be more sensitive to GM volume differences in a VBM analysis that parametric inference (see Battaglini et al., 2009 and Li et al., 2009).
> >
> >So really, you should use randomise to do what you want to do www.fmrib.ox.ac.uk/fsl/randomise (plus in FEAT, you should have used the smoothed data, all the more that you then assume the normal distribution of them).
> >
> >Hope this helps,
> >Gwenaëlle
> >
> >
> >--- En date de : Jeu 7.1.10, chen <[log in to unmask]> a écrit :
> >
> >> De: chen <[log in to unmask]>
> >> Objet: [FSL] how to do second level analysis of fslvbm (instead of using randomise)? Thanks!
> >> À: [log in to unmask]
> >> Date: Jeudi 7 Janvier 2010, 15h30
> >> Dear FSL Experts,
> >>
> >> I am not quiet sure about FSLvbm statistics. FSL
> >> recommonds using randomise to do second level
> >> statistics, but SPM would treat it just like functional
> >> con* images for second level analysis. So is there
> >> other commond to do second level analysis in FSL except
> >> for randomise?
> >>
> >> I was thinking of using FEAT higher-level
> >> analysis but failed. I selected "Inputs are 3D cope
> >> images from FEAT directories" and specified processed
> >> 3d images of every subject
> >> (*_GM_to_template_GM_mod.nii.gz), then specify the
> >> design metrix. I thought it will treat all
> >> images as COPEs, but I got this error message:
> >> "registration has not been run for all of the FEAT
> >> directories that you have selected for group
> >> analysis"
> >> Then I merged all these 3d images into a 4D data and
> >> try first-level analysis, I turned off all options in
> >> pre-stats tab (I am not sure about how
> >> to treat "high pass filter" in data tab)
> >> and specified a design matrix without convolution. I got a
> >> result but I am not sure is this correct or not.
> >>
> >> Any help is appreciated, Thanks!
> >>
> >> --
> >>
> >> Chunhui Chen
> >>
> >>
> >>
> >>
> >
> >
> >
>
>
>
>
>
|