[FieldTrip] connectivity analysis

jan-mathijs schoffelen jan.schoffelen at donders.ru.nl
Tue Nov 12 11:57:59 CET 2013


Hi Bart,

If you 7 frequency bins within each frequency bin are considered to belong to the same frequency band, I'd suggest to use the multitaper approach to estimate the spectral parameters per frequency band, using just a single frequency bin to represent the whole band of interest. So, rather than for example estimating from 15 until 21 Hz in steps of 1 Hz, you can also estimate only at 18, using cfg.tapsmofrq = 3 in ft_freqanalysis. I suggest to have a look at the documentation on the FT-wiki that is concerned with frequency analysis. I would certainly not average the spectral representation prior to computing the coherence, neither across the individual frequency bins, and even more certainly not across trials.

Regarding statistics, I don't want to discourage you, but I don't think it makes sense to attempt doing statistics at the channel level to begin with, due to the effects of volume conduction. Also, in any group comparison (but also when doing a comparison across conditions), if there is any between groups in terms of SNR (e.g. more or less alpha power across groups) you are bound to find a statistically significant difference in estimated connectivity as well. It remains to be motivated in such a case, that the change in estimated connectivity actually reflects a change in true connectivity.

If you want to do statistics irrespective of these caveats, you may want to look into FieldTrip's implementation of the non-parametric permutation testing framework. There is ample documentation about this on our wiki as well.

Best wishes,

Jan-Mathijs



On Oct 30, 2013, at 1:22 PM, Bart Michiels wrote:

> Hi,
> 
> I have 30 patients and 30 controls and I'm investigating their coherence (EEG, 128 electrodes). Every patient has ~30 trials resting state eyes open consisting of 7 frequency bins with 7 frequencies in each bin. My goal is to show connectivity differences between different brain regions in the control and patient group (doing electrode-level analysis now, source-level analysis is next step).
> 
> - Is it more appropriate to keep the averaging step as the latest step (ie. calculate all coherence for all the different subjects, for all trials, for all different frequencies in frequency bins) or is it better to do the averaging asap (ie. average all frequencies in 1 frequency bin at the time-frequency analysis step, average all trials at the ft_freqdescriptives step, ...)
> 
> - Is there any better way to do some statistics on the 128x128x7x7 (128 electrodes, 7 frequency bins, 7 frequencies in each bin) besides using the ttest2() matlab function?
> 
> Any tips & tricks are more then welcome!
> 
> Bart
> _______________________________________________
> fieldtrip mailing list
> 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
Telephone: +31-24-3614793

http://www.hettaligebrein.nl

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20131112/67cada43/attachment-0001.html>


More information about the fieldtrip mailing list