[Eeglablist] EEG signal preprocessing prior to coherence analysis

[Kilmartin, Liam]

Hi,
 
I am relatively new to EEG signal analysis though I do have significant experience in signal processing in other application spaces. I plan to undertake a coherence analysis using ERLCOH (using newcrossf) on a common average referenced 64 channel EEG database. This study will examine the response of subjects to visual stimuli. The raw EEG signals have been ocular corrected and epoched in advance of my analysis (with trial rejection due to excessive residual artefacts). My concerns now related to what form of additional pre-processing (if any) would be expected prior to a linear coherence analysis specifically to deal with the volume conduction problem.
 
The specific questions that I have which I would appreciate any comments on:
 
(a) Does the use of a common average reference during acquisition in any way significantly address the volume conduction problem in the context of a coherence analysis ?
 
(b) Some apparently basic approaches that I have come across in the literature which attempt to address the issue of volume conduction resulting in an over-estimation of "real" coherence include:
 
    (i) Considering only the imaginary part of coherence to provide a lower bound estimate for "real" coherence
    (ii) Ignoring coherence magnitude values which have coherence phase angles close to 0 degrees (as these are likely due to volume conduction)
 
Intuitively it would appear to me that techniques such as these are likely to under-estimate "real" coherence (admittedly perhaps not to the same degree which ignoring them will tend to over-estimate coherence). Do any of these techniques offer acceptable solutions to the problem ?
 
(c) More complex approaches which I have also encountered in the literature involve source estimation approaches or the application of  laplacian surface models prior to coherence analysis. Would such approaches be viewed as the "gold standard" approach and hence would the use of such techniques be an  expectation of most reviewers of any future publications in this area ?
 
I appreciate that there may be no definitive answer to this question and which approach is best may be a matter of opinion, however any comments on these issues would be much appreciated.
 
Regards,
 
Liam
 
Liam Kilmartin
College of Engineering and Informatics,
National University of Ireland, Galway,
IRELAND