Output of mscohere function

Hi all,

I have a doubt with regards to mscohere function output in matlab. I have 64 channel EEG data to extract coherence data for. Which means I would need a 64x64 matrix output for a particular frequency band say 8 to 13 Hz. I went through the documentation of mscohere function sveral times. One thing that I saw was that mscohere Cxy part can either be a vector, matrix or 3-D array. When I use the following code to calculate coherence between just two channels, channel 20 and channel 50, I get a 501×1 single column vector:

ch1 = 20;
ch2 = 50;
window = 2*EEG.srate; % create 2 sec epochs
noverlap = 0;    % overlap 
nfft = window*2;    % zero-padding factor 2 for smoothing
fs = EEG.srate;

[Cxy,F] = mscohere(EEG.data(ch1,:),EEG.data(ch2,:),window,noverlap,nfft,fs);
Cxy

Cxy =

  501×1 single column vector

    0.0003
    0.0060
    0.0042
    0.0003
    0.0047 % I copied the first few values only to show the output I am getting

Can someone explain this output to me? Are these coherence values between two channels at 501 different frequency points (I'm just guessing. Not sure!)? Also, how do I input the frequency band of my interest like 8-13 Hz and calculate coherence just for that? Are there more steps in calculating coherence in matlab? How do I get a single value (between 0 to 1) for channel 20 x channel 50 coherence for 8-13 Hz frequency band?

I hope my question makes sense. Please let me know if I need to explain anything further.

Thanks!

Satya

Answer 1

[Daniel M]

Yes, these are the coherence values between the two channels at the frequencies returned in F. You can manually set the frequencies that you wish, for example F = [8:0.5:13]. The help section indicates this.

%   [Cxy,F] = MSCOHERE(X,Y,WINDOW,NOVERLAP,F,Fs) computes the coherence
%   estimate at the physical frequencies contained in the vector F.  F must
%   be expressed in hertz and have at least two elements.

Then you can average the output to get the total coherence over that band.

[Daniel M]

Here are two other well used functions that compute the coherence between EEG signals.

In the Fieldtrip toolbox: ft_connectivityanalysis and follow the case where cfg.method = 'coh'.

Here is the code

https://github.com/fieldtrip/fieldtrip/blob/master/ft_connectivityanalysis.m

And here is a tutorial

http://www.fieldtriptoolbox.org/tutorial/connectivity/#extract-the-virtual-channel-time-series

In the spm12 toolbox: spm_eeg_ft_multitaper_coherence, found at https://github.com/neurodebian/spm12/blob/master/toolbox/MEEGtools/spm_eeg_ft_multitaper_coherence.m

[Satya Gopal]

Thank you for your answer, Daniel! I tried setting the frequencies manually with F and I am getting fewer values. I just have a small doubt. Why don't we use F=[8:13] but put F=[8:0.5:13]? Is it because we get a better frequency resolution or smoothening?

Yes, I will check the Fieldtrip option as well!

[Daniel M]

Because interesting things in EEG don't happen at the just the discrete frequencies, e.g. 8:13. They happen at any frequency. That's why we sample with a smaller step. You can choose a step size that works for you.

[Satya Gopal]

Thanks again, Daniel! That makes sense! I really appreciate your effort in explaining things.