[Eeglablist] Coherence or Correlation among set of electrodes

[Jason Palmer]

Hi Iman,

Oh, sorry I misunderstood. It sounds like you might be able to use first
canonical correlation coefficient, which gives the maximum correlation
between univariate projections of the two subspaces. Specifically, if we
have we have two sets of data, either sets of electrodes, or groups of ICA
activations, say in the matrices x and y, where x is size m by N, and y is n
by N. Then xf and yf be the STFTs of x and y at some frequency of interest.
Define,

Sxx = xf * xf' / N;
Syy = yf * yf' / N;
Sxy = xf * yf' / N;

Sxxi = pinv(sqrtm(Sxx));
Syyi = pinv(sqrtm(Syy));

ccf = norm(Sxxi*Sxy*Syyi);

Doing this for various frequencies and plotting ccf as a function of
frequency will show the maximal correlation between the subspaces and how it
varies with frequency.

Best,
Jason

-----Original Message-----
From: Iman M.Rezazadeh [mailto:irezazadeh at ucdavis.edu] 
Sent: Wednesday, October 17, 2012 9:11 PM
To: japalmer at ucsd.edu; mmiyakoshi at ucsd.edu; 'Scott Makeig'
Cc: mullen.tim at gmail.com
Subject: RE: [Eeglablist] Coherence or Correlation among set of electrodes

Thanks Jason! 
If I understand correctly, the way your described is computing pair by pair
coherence ( mutual information) . However, I am saying that how can we
measure/find the region ( bunch of spatially related electrodes)  to region
coherence. Tim, purposed me SIFT and currently I am working on that. 
BTW, here I have attached some papers which have shown interesting results
re this issue.
Would be happy to have your idea/ thought!
Best,
Iman

Iman M.Rezazadeh, PhD
Postdoctoral Scholar
Center for Mind and Brain
University of California, Davis
irezazadeh at ucdavis.edu
cell:310-490-1808



-----Original Message-----
From: Jason Palmer [mailto:japalmer29 at gmail.com]
Sent: Wednesday, October 17, 2012 5:00 PM
To: mmiyakoshi at ucsd.edu; 'Iman M.Rezazadeh'; 'Scott Makeig'
Cc: mullen.tim at gmail.com; eeglablist at sccn.ucsd.edu
Subject: RE: [Eeglablist] Coherence or Correlation among set of electrodes

Hi Iman,

You can also use mutual information to get component relationships. The
function minfo.m, available here:
http://sccn.ucsd.edu/~jason/
will compute pairwise mutual information between component source
activations. The output matrix element MI(i,j) gives the mutual information
between source i and source j. It can also compute the mutual information at
a specified number of lags, in MI(i,j,-numlags:numlags),  which is similar
to cross-correlation, but always positive. FFT can also be performed on the
minfo output time series to get mutual info as a function of frequency. Or
more simply, you can take the average MI over the lags to get component
dependence measures allowing for delays. The MI matrix can be conveniently
visualized using (assuming no multiple lags):
>> figure, imagesc( MI - diag(diag(MI)) );
Or
>> MImean = mean(MI,3); figure, imagesc( MImean - diag(diag(MImean)) );
to average over multiple lags.
The attached (unsupported) function will attempt to rearrange the component
order so that related components are near each other, attempting to "block
diagonalize" the MI matrix. It outputs the rearraged MI matrix and the new
component order. Computation for a large number of components can be time
consuming, esp. with delays, but if you just use the top 30 variance
components (which should contain most comps of interest) it shouldn't take
that long, and you can inspect the MI relationships directly.

Best,
Jason


-----Original Message-----
From: Makoto Miyakoshi [mailto:mmiyakoshi at ucsd.edu]
Sent: Monday, October 15, 2012 9:36 PM
To: Iman M.Rezazadeh; Jason Palmer; Scott Makeig
Cc: mullen.tim at gmail.com; eeglablist at sccn.ucsd.edu
Subject: Re: [Eeglablist] Coherence or Correlation among set of electrodes

Dear Iman and ICA persons,

> So, how can I pick up topo and source time series among all 128 
> sources

Jason and Scott, this Iman's question is legitimate. I have been asking the
same question too. Other than Tim's new implementation of the Bayesian stuff
in SIFT in future, what other conventional approach can help?

Makoto


2012/10/15 Iman M.Rezazadeh <irezazadeh at ucdavis.edu>:
> Dear Makoto and Tim,
> Thanks for your reply and suggestion  ! I will definitely go through 
> SIFT to see how can it help me ...
> Each source (after BSS) has a time series and a topo map. Suppose that 
> I have 128 channels means 128 sources and I want to know which sources 
> are related to each other. So, how can I pick up topo and source time 
> series among all 128 sources ( after removing some sources in the 
> artifact rejection stage ) ?
> - for an example, I am interested to see if there is any 
> fronto-parital synchrony for an inhibitory task. After BSS, there are 
> more than 1 source which have frontal or parietal major activities in 
> their topo. Now how could I select the appropriate source?
>
> I have found the attached paper. Do you have any thought?
> Best,
> Iman
>
> Iman M.Rezazadeh, PhD
> Postdoctoral Fellow
> Center for Mind and Brain
> University of California, Davis
> irezazadeh at ucdavis.edu
> cell:310-490-1808
>
>
>
> -----Original Message-----
> From: Makoto Miyakoshi [mailto:mmiyakoshi at ucsd.edu]
> Sent: Monday, October 15, 2012 7:53 PM
> To: Iman M.Rezazadeh
> Cc: eeglablist at sccn.ucsd.edu
> Subject: Re: [Eeglablist] Coherence or Correlation among set of 
> electrodes
>
> Dear Iman,
>
> I wonder why Tim did not mention his SIFT.
> http://sccn.ucsd.edu/wiki/SIFT
> Apply SIFT after ICA. You can choose ICs so that they cover areas of 
> interests.
>
> Makoto
>
> 2012/10/15 Iman M.Rezazadeh <irezazadeh at ucdavis.edu>:
>> Hi,
>>
>> Just wonder if there is a way to calculate the coherence measure 
>> between two regions( set of channels-instead of two single channels) 
>> in EEGLAB or any other software?  In other word, how can we find 
>> regions in the brain which their activities are mostley related to 
>> each
> other using EEG ?
>>
>>
>>
>> Best,
>>
>> Iman
>>
>>
>>
>> Iman M.Rezazadeh, PhD
>>
>> Center for Mind and Brain
>>
>> University of California, Davis
>>
>> irezazadeh at ucdavis.edu
>>
>> cell:310-490-1808
>>
>>
>>
>>
>> _______________________________________________
>> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
>> To unsubscribe, send an empty email to 
>> eeglablist-unsubscribe at sccn.ucsd.edu
>> For digest mode, send an email with the subject "set digest mime" to 
>> eeglablist-request at sccn.ucsd.edu
>
>
>
> --
> Makoto Miyakoshi
> JSPS Postdoctral Fellow for Research Abroad Swartz Center for 
> Computational Neuroscience Institute for Neural Computation, 
> University of California San Diego



--
Makoto Miyakoshi
JSPS Postdoctral Fellow for Research Abroad Swartz Center for Computational
Neuroscience Institute for Neural Computation, University of California San
Diego