Dear Tommy,
Just one comment on Vladimir's response: He is right, the volume images in
BESA are already spatially normalized, they are all in Talairach dimensions.
There's no need for additional spatial normalization prior to statistical
analysis.
Note that the new BESA version (BESA Research 5.3) has a direct interface to
MATLAB, and on the BESA website www.besa.de/updates/matlab there's a little
tool that allows to statistically compare two groups of BESA-generated 3D
volume images, making use of the statistics functions of the 'Fieldtrip'
toolbox. You could compare the outcome of this script to the SPM statistics
output.
If you encounter problems or have further questions, please let me know.
Best wishes,
Karsten
--------------------------------------
Dr. Karsten Hoechstetter
MEGIS Software GmbH
Gräfelfing, Germany
HRB München 109956
CEO Dr. Michael Scherg
--------------------------------------
-----Original Message-----
From: Vladimir Litvak [mailto:[log in to unmask]]
Sent: Dienstag, 6. April 2010 10:47
To: [log in to unmask]
Cc: [log in to unmask]; [log in to unmask]
Subject: Re: MEG_Normalization
Dear Tommy,
On Tue, Apr 6, 2010 at 12:28 AM, <[log in to unmask]> wrote:
> I am working on a set of MEG data on motor function and wonder if I could
> ask you a couple of questions?
>
> 1) I have analyzed the data in BESA using beamforming method. I exported
> individual's images and entered them into SPM for group analysis. So, the
> first question is how do I spatially normalized individual image into MNI
> template space. I searched spm_eeg but can't seem to find a GUI for
> normalization.
This functionality is in SPM/fMRI. It is possible to normalize your
individual structural to the template and then apply the same
normalization to the functional images. However, in the case of
beamformer images I'm not sure it is really necessary because it is
quite likely that the output of BESA beamformer is normalized already
and even if not, the level of spatial detail in those images is not
such that normalization would improve things. You should just make
sure the images are in decent registration with each other and perhaps
just smooth them a little.
>
> 2) Output from 2nd level analysis shows activation over broad brain
regions.
> However, when I tried to churn out the t and p values using "whole brain",
> only values of activations in one hemisphere showed up (these are the ones
> with strong t values), and weaker activations in the other hemisphere (my
> region of interest) did not show up. Although, when I used xjview, I could
> use the cursor to source out the t and p values, i.e. they are there but
> kind of drowned by the very strong hemispheric activation.
>
This is not an SPM problem per se but a problem with you images. If
you know in advance that you are interested in a particular ROI you
should not do SPM of the whole brain but either use an explicit mask
or a small volume correction. Also, do you compare two conditions or
just do a single sample t-test on your images? The latter is not valid
because the values in the images are non-negative so it doesn't make
sense to compare them to zero. This has been discussed on the SPM list
before. Finally, there are some diagnostic tools in SPM (the 'plot'
button in the 'Results' interface) that make it possible to check why
you get high or low t-values for a particular voxel (for instance is
it because of a single outlier, weak effect or high inter-subject
variance?). Sometimes it might be useful to use grand mean scaling to
reduce the variance. That's what is presently done for source
reconstruction images generated by SPM.
Best,
Vladimir
> Please advise what I should do to resolve these problems?
>
> Any help is appreciated. Thank you.
>
> Regards
>
> Tommy Ng
> PhD Candidate
> MACCS|Macquarie University
>
> ----------------------------------------------------------------
> This message was sent using IMP, the Internet Messaging Program.
>
>
|