Difference between wPLI from FieldTrip and MNE - Python

igor-casatti · April 26, 2022, 4:47pm

Hello,

I am currently working with the calculation of wPLI on EEG data and I saw that the MNE implementation of the wPLI is a little bit different by the one from MATLAB FieldTrip package. It seems that MNE uses the equation 8 from Vinck et al., 2011, while the FieldTrip uses the same formula, but without the abs() operator on the numerator.

I am calculating the values using MNE - Python, while a colleague is using FieldTrip, and our results are quite different. Does someone know if the different formulation can cause different results for the same set of patients?

richard · April 30, 2022, 8:17am

Hello @igor-casatti and welcome to the forum!

I’m not familiar at all with wLPI – maybe @agramfort has an idea…?

Best wishes,
Richard

agramfort · May 3, 2022, 6:51am

Hi

Can you point to the lines in mne and fieldtrip source codes that are different? I would also add Robert Oostenveld about this. He is likely to be the one who wrote the WPLI code in fieldtrip.

Alex

igor-casatti · May 3, 2022, 12:33pm

Hi,

In the line 119 of the FieldTrip script:

github.com

fieldtrip/fieldtrip/blob/8bebf82310b247519bdefce336ab1e815a2d6852/connectivity/ft_connectivity_wpli.m#L112-L124

      
        
              input    = imag(input);          % make everything imaginary
              outsum   = nansum(input,1);      % compute the sum; this is 1 x size(2:end)
              outsumW  = nansum(abs(input),1); % normalization of the WPLI
              if debias
                outssq   = nansum(input.^2,1);
                wpli     = (outsum.^2 - outssq)./(outsumW.^2 - outssq); % do the pairwise thing in a handy way
              else
                wpli     = outsum./outsumW; % estimator of E(Im(X))/E(|Im(X)|)
              end
              wpli = reshape(wpli,siz(2:end));
            else
              wpli = NaN(siz(2:end)); % for one observation, we should return NaNs
              ft_warning('ft_connectivity_wpli:nTrials', 'computation wpli requires >1 trial, returning NaNs');

And in the lines 428 - 451 of the MNE code:

github.com

mne-tools/mne-connectivity/blob/2d47ae7d281c2ff2fe17f2d55f2e3d7dd0d0a4c5/mne_connectivity/spectral/epochs.py#L428-L451

      
        
            def accumulate(self, con_idx, csd_xy):
                """Accumulate some connections."""
                im_csd = np.imag(csd_xy)
                self._acc[0, con_idx] += im_csd
                self._acc[1, con_idx] += np.abs(im_csd)
            
            
def compute_con(self, con_idx, n_epochs):
                """Compute final con. score for some connections."""
                if self.con_scores is None:
                    self.con_scores = np.zeros(self.csd_shape)
            
            
    num = np.abs(self._acc[0, con_idx])
                denom = self._acc[1, con_idx]
            
            
    # handle zeros in denominator
                z_denom = np.where(denom == 0.)
                denom[z_denom] = 1.
            
            
    con = num / denom

This file has been truncated. show original

Thanks for the reply.

Igor

richard · May 3, 2022, 12:50pm

Hello Igor, I took the liberty to slightly edit your posting such that the relevant bits of code would show up in the forum directly.

agramfort · May 5, 2022, 7:04am

hi @igor-casatti

I just looked at the code. The two implementation are doing the same except that MNE applies an abs at the end following es (8). It’s true that one could argue that it’s still possible to compute the abs afterwards but the paper has it. So in fieldtrip just compute the abs manually and the results should match.

Alex

Topic		Replies	Views
ICA Help MNE vs Fieldtrip 💬 Support & Discussions preprocessing , eeg , ica , resting-state	16	529	April 18, 2023
How to interpret mne_connectivity results 💬 Support & Discussions eeg , connectivity	2	355	April 8, 2023
Different tools read data differently 💬 Support & Discussions eeg	2	105	August 17, 2023
EEG Forward model ✉️ Mailing List Archive (read-only) list-archive	1	116	June 17, 2016
Discrepancy between mne_analyze and mne-python ✉️ Mailing List Archive (read-only) list-archive	1	162	April 5, 2016

Difference between wPLI from FieldTrip and MNE - Python

Related Topics