Dear BST Community,

I am working on an analysis about the reliability of EEG functional connectivity estimation. As a part of this analysis, I computed the correlation between the time series of scouts after performing source estimation with different sensor signals. I have 64 electrodes with real or generated signals, source estimation is performed with default anatomy, the head model is computed with OpenMEEG BEM, cortex surface only (15002 vertices), default values for BEM layers and conductivities, inverse solution is computed with sLORETA (fixed source orientation). Source signals are downsampled to the DKT atlas (62 ROIs). One interesting result is that the correlation values (linear pairwise correlation coefficients) among the 62 reconstructed source signals are relatively stable, even if the sensor signals change drastically. For example, generated random noise on all channels (sensor space) produced correlation values in the source space that are comparable (but of course not equal) to the result of source reconstruction carried out for real EEG signals. In other words, uncorrelated signals in the sensor space can result in correlated signals in the source space. This was very surprising for me at the beginning, but considering the fact that the inverse kernel defines a linear combination of sensor signals, and the head model is always the same, now it seems to be realistic that (because of downsampling to atlas) highly averaged source space signals can be strongly correlated even if signals in the sensor space were uncorrelated.

I would like to ask you to share your opinion about this phenomenon. Is it a well-known phenomenon? I am quite new to source estimation, so I would greatly appreciate any advice.

Best regards,

Peter