[FieldTrip] Source reconstruction and NAI

jan-mathijs schoffelen jan.schoffelen at donders.ru.nl
Thu Apr 7 14:23:58 CEST 2011


Dear Marc,

It seems from your question that you are interested in single trial  
reconstructed time series. If your purpose is to do a statistical  
comparison across a set of conditions I would not do a noise  
normalization at all. You mention that you used a spatial filter  
common to the three conditions. As a consequence, the estimate of the  
noise will be independent of the condition, so any normalization you  
would apply just leads to a scaling of the data, and will not change  
the outcome of your statistical test.
Yet, I understand that it sometimes makes sense to do a normalization  
in order to be able to make a sensible visualization of the data (i.e.  
removing the big central blob). Also, in this case I would compute a  
descriptive statistic across the conditions, rather than normalizing  
with an estimate of the projected noise (which in FieldTrip is rather  
rudimentary because it assumes the noise to be spatially white). You  
could for example compute  an F-value from a one-factor, three level  
anova, or do a pairwise comparison of conditions using a t-test.
Another alternative (which does not affect your statistical test  
either), is to work with norm-normalized leadfields. This will take  
away the blob in the centre of the volume, and facilitates  
visualization.

Best wishes,

Ja-Mathijs


On Apr 5, 2011, at 6:54 PM, Marc Recasens wrote:

> Dear all,
>
> As far as I know it is possible to reconstruct the time-course of  
> the sources (obtained using sourceanalysis) by projecting/ 
> multiplying the filter-weights on the data.
> In my case, the specific code I am using is:
>   for i=1:nvoxels(inside)
>
>         for tr=1:ntrials
>
>             ts(tr,:)=(source.avg.ori{source.inside(i)} 
> (1,:)*source.avg.filter{source.inside(i)}*data2{condit}.trial{tr});
>
>         end
>
>         datvx(i,:,:)= single_trial_time_course
>
> end
>
> I used a common filter with 3 different conditions, thus I think  
> this is the only way I have to reconstruct the activity for one of  
> the conditions.
> My Question is about how to apply the NAI normalization here. Can I  
> just divide the output of the source reconstruction by the projected  
> noise of the filter? I am tempted to do something like that
>
> datvx(i,:,:)= single_trial_time_course ./ repmat(noise(inside), 
> [size(ts,1), size(ts,2)]
>
> That is, I divide the power by the noise estimate (for each voxel)  
> in every trial and time-point. Could anyone tell me whether this is  
> a correct way to procedure? Any other workaround?
>
>
>
> Thanks in advance!
>
> -- 
> Marc Recasens
> Tel.: +34 639 24 15 98
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip

Dr. J.M. (Jan-Mathijs) Schoffelen
Donders Institute for Brain, Cognition and Behaviour,
Centre for Cognitive Neuroimaging,
Radboud University Nijmegen, The Netherlands
J.Schoffelen at donders.ru.nl
Telephone: 0031-24-3614793

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20110407/cb6f4484/attachment.html>


More information about the fieldtrip mailing list