[FieldTrip] Antw: Re: MNE surface data: normalising and averaging?

[Gregor Volberg]

Hi Gio, Jan-Mathijs and Jörn,
 
thank you all for your helpful replies. Gio, a special 'thank you' for your
very detailled response. These are very interesting points, though somewhat
beyond of what I try to do. I will be working through the suggested literature
and the new tutorial and might come back later if there are further questions.

Best regard,
Gregor
 

 
 
-- 
Dr. rer. nat. Gregor Volberg <gregor.volberg at psychologie.uni-regensburg.de> (
mailto:gregor.volberg at psychologie.uni-regensburg.de )
University of Regensburg
Institute for Experimental Psychology
93040 Regensburg, Germany
Tel: +49 941 943 3862 
Fax: +49 941 943 3233
http://www.psychologie.uni-regensburg.de/Greenlee/team/volberg/volberg.html
>>> Gio Piantoni <gio at gpiantoni.com> 05.03.2013 09:20 >>>
Hi Gregor,

I also like the surface approach a lot, but things get really
complicated when you have to average across subjects. I don't know if
you are familiar with statistics in freesurfer, but what they do is to
convert the source points of the single subject to a subject-specific
sphere. Then the sphere is averaged across subjects. The
coregistration between these spheres takes advantage of landmarks.

See my attempt at:
http://mailman.science.ru.nl/pipermail/fieldtrip/2012-September/005585.html
If I understand you correctly, you are at step 3). You can use smudge
to get to the high-resolution smoothwm (or pial). Then the number of
vertices in smoothwm corresponds to the number of vertices in the
sphere.

Here you have two options: use fieldtrip code or MNE code. I tried
using fieldtrip code, but where I got stuck was to do statistics on
the averaged sphere (taken from fsaverage). I used this code:
https://github.com/gpiantoni/eventbased/blob/77c3f85ef94baf9d4ab629d48a31e8046c924518/powsource_grand.m#L136-L158
where data{i1,i2,p,h} is already projected onto the subject-specific sphere.
However, after projecting to the averaged sphere, you need to do
statistics on a 2-d cortical sheet and, from my understanding, the
subfunctions to create clusters do not work all that well. I think
it'd be very powerful to do statistics on a 2d sheet. See
http://mailman.science.ru.nl/pipermail/fieldtrip/2012-January/004749.html
and
http://mailman.science.ru.nl/pipermail/fieldtrip/2012-February/004766.html
for discussion. I did not get far because the clustering methods
needed updated to work on cortical sheets in an efficient manner.

Another route you can try is to use the MNE/freesurfer software to do
that. See for example: http://martinos.org/mne/manual/morph.html You'd
just write the surface to disk and use MNE/freesurfer. I personally
don't like the method to correct for multiple comparisons in
freesurfer (which I think might give false positive) and I think that
the fieldtrip stats is better, but I had some problems getting to work
on 2-d cortical sheets.

For the moment, I decided to go with the volume-based approach,
although I have the feeling that it's less accurate and less
sensitive. See also: Tucholka, A., Fritsch, V., Poline, J.-B., and
Thirion, B. (2012). An empirical comparison of surface-based and
volume-based group studies in neuroimaging. Neuroimage 63, 1443–1453.

Maybe I confused you more than helped, but if you manage to do
statistics on surfaces, please let me know because I'd be interested
in using it.

Hope this helps.

Gio

--
Giovanni Piantoni, MSc
Dept. Sleep & Cognition
Netherlands Institute for Neuroscience
Meibergdreef 47
1105 BA Amsterdam (NL)

+31 20 5665492
gio at gpiantoni.com
www.gpiantoni.com

On Mon, Mar 4, 2013 at 6:45 PM, Gregor Volberg
<Gregor.Volberg at psychologie.uni-regensburg.de> wrote:
> Dear Fieldtrippers,
>
> I need to ask for a helpful hint on MNE source reconstructions. Following
> this tutorial http://fieldtrip.fcdonders.nl/tutorial/minimumnormestimate I
> obtained nice individual cortical meshes and source points; the MNE forward
> and inverse solution for the individuals all work well. I would now like to
> normalize the individual grids to a template brain in order to do
statistics
> and averaging for plotting. I figured out that I could use
> ft_sourceinterpolate to tranform the grid into a volumetric representation,
> and then use ft_volumenormalise to normalise to a standard brain.  But this
> did not work too well for source structures containing time series (like
> 'mne'-stcutures) where the computational load gets very high. I also tried
> to use a grid of one subject as a template for the other subject's grids
> with ft_sourceinterpolate and cfg.interpmethod = 'smudge', but this seems
to
> require two grids with the same number of source points as input(?).
> Is there a way to do a normalisation directly on brain triangulations /
> source grids ? Thanks for any help!
> Best regards,
> Gregor
>
> --
> Dr. rer. nat. Gregor Volberg <gregor.volberg at psychologie.uni-regensburg.de>
> ( mailto:gregor.volberg at psychologie.uni-regensburg.de )
> University of Regensburg
> Institute for Experimental Psychology
> 93040 Regensburg, Germany
> Tel: +49 941 943 3862
> Fax: +49 941 943 3233
> http://www.psychologie.uni-regensburg.de/Greenlee/team/volberg/volberg.html
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20130306/c84f53f2/attachment-0002.html>