[FieldTrip] Coherence of single trials

Andrei Medvedev am236 at georgetown.edu
Tue Nov 27 21:06:17 CET 2012


Dear Monika,

Indeed, Coh cannot be calculated on single trials, it requires averaging 
of the cross-spectrum over (some) trials. Customarily, people average 
ALL trials recorded in a subject and end up with just only one coherence 
measurement for this subject. Then obviously stats cannot be run on the 
individual level and only a group analysis can be done (e.g., you have 
20 subjects, they give you 20 coherence values and you run your stats on 
those 20 values). Does this mean that coherence can be analyzed only on 
a group level? Perhaps not.

I suggest the following approach. Coherence can be considered as a 
'sample', or 'population', statistic (similar to means, medians, 
variances etc, which are calculated from a set of data points). Methods 
of estimating statistical properties of those 'population' measures are 
well known, they are based on resampling, jackknifing and bootstrapping. 
Simply put, you can take a SUBSET of your trials (e.g., selecting  
randomly 50 trials from all 100 trials in a particular subject) and 
calculate coherence over those 50 selected trials. Then repeat the 
process of random selection (with replacement, as in jackknifing) a 
certain number of times (N) and you would have a set of N 'sample' 
coherences for this subject. Then you can run your stats on those 
'sample' coherences for this particular subject thus achieving your goal 
of the individual subject analysis.

Any feedback from the fellows Fieldtripers about this approach?

Thank you,

Andrei Medvedev
Georgetown University.


On 11/27/2012 6:49 AM, jan-mathijs schoffelen wrote:
> Dear Monika,
>
> It may have escaped your attention while digging to the previous 
> posts, but the question has been raised (and answered ;-) ) before. 
> Just for the very very last time I will make the following statement:
>
> *Single trialcoherenceestimatescannotbecomputed!*
> *
> *
> Perhaps the font size and color will make it memorable ;-). This is 
> due to the simple fact that coherence is defined across observations. 
> If you have just a single observation, due to the mathematics 
> involved, the coherence value will be 1.
> Since you want to do group analysis, I think you really should compute 
> the coherence per condition, where each condition should consist of a 
> sufficient number of trials. Assuming that the time domain data is 
> organised such that you have all trials (irrespective of the 
> condition) in a single structure, do something like this.
>
> for each subject
>   cfg = [];
>   cfg.output = 'fourier';
>   cfg.method = 'mtmfft'/'mtmconvol'
>   cfg.keeptrials = 'yes';
>   ...
>   freq = ft_freqanalysis(cfg, data);
>
>   cfg = [];
>   cfg.channelcmb = {something here}
>   cfg.trials = [indices of trials belonging to condition 1];
>   coh1 = ft_connectivityanalysis(cfg, freq); % compute coherence 
> across the trials for condition 1
>   cfg.trials = [indices of trials belonging to condition 2];
>   coh2 = ft_connectivityanalysis(cfg, freq); % compute coherence 
> across the trials for condition 2
> end
>
>
> Good luck,
>
> Jan-Mathijs
>
>
> On Nov 27, 2012, at 12:25 PM, Monika Mellem wrote:
>
>> Thank you Sheraz and Jan-Mathijs for your help!
>> Sheraz, if you do have the code to keep single-trial coherence 
>> estimates, I would much appreciate it if you could pass it along 
>> (monikamellem at gmail.com <mailto:monikamellem at gmail.com>).
>> And Jan-Mathijs, okay, the necessary statistics make sense, and I 
>> will z-transform the data too.  But as far as you know, is there no 
>> implementation already in Fieldtrip to keep the single-trial 
>> coherence estimates?  I assumed setting cfg.keeptrials = 'yes' would 
>> do this, but is there something else I'm missing as this doesn't seem 
>> to work?  I would prefer to try what Fieldtrip has implemented first 
>> before integrating Sheraz's code into my version of Fieldtrip.
>>
>> Thank you very much!
>> Monika
>>
>>
>> On Tue, Nov 27, 2012 at 3:14 AM, jan-mathijs schoffelen 
>> <jan.schoffelen at donders.ru.nl <mailto:jan.schoffelen at donders.ru.nl>> 
>> wrote:
>>
>>     Hi Monika,
>>
>>     If you want to test across subjects in  a paired fashion, you can
>>     use depsamplesT. Your dependent variable then should contain for
>>     each subject and condition the coherence (which of course should
>>     be estimated across all trials belonging to that condition). Even
>>     better would be to perform a so-called Z-transformation to the
>>     data. For some inspiration you could have a look at the following
>>     paper: http://www.jneurosci.org/content/31/18/6750.long
>>
>>     Best,
>>
>>     Jan-Mathijs
>>
>>     On Nov 27, 2012, at 12:22 AM, Monika Mellem wrote:
>>
>>>     Dear Fieldtrippers,
>>>
>>>     I am trying to calculate coherence estimates on individual
>>>     trials of our EEG data and then want to perform a
>>>     within-subjects statistical test, but so far I have not been
>>>     able to figure out how to do this.  I am able to calculate
>>>     coherence for our 22 subjects and then perform statistics on the
>>>     grand average, so I understand the basic usage of these
>>>     functions (ft_freqanalysis, ft_connectivityanalysis,
>>>     ft_freqgrandaverage, ft_freqstatistics).  But now we want to do
>>>     statistical testing on individual subjects.  Is it possible to
>>>     keep coherence for individual trials in Fieldtrip as they seem
>>>     necessary for this statistical analysis?
>>>
>>>     I am setting cfg.keeptrials = 'yes' as an input to
>>>     ft_freqanalysis, and get the cross-spectra of individual trials
>>>     (I also tried computing the fourier spectra but it made no
>>>     difference in the following step).  However, when computing the
>>>     coherence in ft_connectivityanalysis, Fieldtrip averages over
>>>     the trials.  Is there a way to output coherence for individual
>>>     trials instead?  Please see the full cfg structure settings below.
>>>
>>>     Also, I did look into the option of cfg.statistic =
>>>     'indepsamplesZcoh' as an input to ft_freqstatistics since
>>>     previous posts suggested this was for single subject coherence
>>>     statistics, but our data is paired as the same stimulus appears
>>>     in the 2 conditions we are comparing.  Is there a paired test
>>>     for single subject coherence testing, or should we use
>>>     cfg.statistic = 'depsamplesT'?
>>>
>>>     Many thanks for any suggestions you may have!
>>>     Monika
>>>     ________________
>>>     Monika Mellem
>>>     PhD Candidate
>>>     Interdisciplinary Program in Neuroscience
>>>     Georgetown University
>>>     Washington, DC
>>>     msm79 at georgetown.edu <mailto:msm79 at georgetown.edu>
>>>     202-687-2687 <tel:202-687-2687>
>>>
>>>     ***********************************************************************
>>>     cfg settings for ft_freqanalysis and ft_connectivityanalysis
>>>     ***********************************************************************
>>>     cfg = [];
>>>     cfg.prestim = -0.5;
>>>     cfg.poststim = 1.5;
>>>     cfg.foilim = [2 30];
>>>     cfg.freqrange = 'low';
>>>     % cfg.output     = 'fourier';
>>>     cfg.output     = 'powandcsd';
>>>     cfg.method = 'mtmconvol';
>>>     cfg.keeptrials = 'yes';
>>>     cfg.channel    = {'all'};
>>>     cfg.channelcmb = {refchan 'all'};
>>>
>>>     cfgcoh=[];
>>>     cfgcoh.channelcmb = cfg.channelcmb;
>>>     cfgcoh.method = 'coh';
>>>
>>>     v = genvarname(['CRSP_' condition '_low']);
>>>     eval([v '= ft_multitaper_powcoh_EGI(cfg, data);']);  %Wrapper
>>>     around ft_freqanalysis
>>>
>>>     w = genvarname(['COH_' condition '_low']);
>>>     eval([w '= ft_connectivityanalysis(cfgcoh, ' v ');']);
>>>
>>>
>>>     _______________________________________________
>>>     fieldtrip mailing list
>>>     fieldtrip at donders.ru.nl <mailto:fieldtrip at donders.ru.nl>
>>>     http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>
>>     Jan-Mathijs Schoffelen, MD PhD
>>
>>     Donders Institute for Brain, Cognition and Behaviour,
>>     Centre for Cognitive Neuroimaging,
>>     Radboud University Nijmegen, The Netherlands
>>
>>     Max Planck Institute for Psycholinguistics,
>>     Nijmegen, The Netherlands
>>
>>     J.Schoffelen at donders.ru.nl <mailto:J.Schoffelen at donders.ru.nl>
>>     Telephone: +31-24-3614793 <tel:%2B31-24-3614793>
>>
>>
>>     _______________________________________________
>>     fieldtrip mailing list
>>     fieldtrip at donders.ru.nl <mailto:fieldtrip at donders.ru.nl>
>>     http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>
>>
>> _______________________________________________
>> fieldtrip mailing list
>> fieldtrip at donders.ru.nl <mailto:fieldtrip at donders.ru.nl>
>> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>
> Jan-Mathijs Schoffelen, MD PhD
>
> Donders Institute for Brain, Cognition and Behaviour,
> Centre for Cognitive Neuroimaging,
> Radboud University Nijmegen, The Netherlands
>
> Max Planck Institute for Psycholinguistics,
> Nijmegen, The Netherlands
>
> J.Schoffelen at donders.ru.nl <mailto:J.Schoffelen at donders.ru.nl>
> Telephone: +31-24-3614793
>
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip

-- 
Andrei Medvedev, PhD
Assistant Professor,
Center for Functional and Molecular Imaging
Georgetown University
Building D, Room 154
4000 Reservoir Rd, NW
Washington DC, 20057
Tel.: 202-687-5126

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20121127/838d5183/attachment.html>


More information about the fieldtrip mailing list