Coherence analysis: Methods, solutions and problems

Abstract

A coherence function is a measure of the correlation of two signals and may be used as a measure for functional relationship between brain areas. In studying functional relationships, referenced EEG (REEG) coherence analysis yields important new aspects of brain activities, which complement the data obtained by power spectral analysis. However, REEG-based coherence tends to show a false high value due to volume conduction from un correlated sources (VCUS). Existing signal processing methods address this issue using a Fourier coherence function of scalp Laplacian. Although this method has been proved useful to reveal correlation between EEG signals with minimum VCUS effects, it only provides frequency-domain analysis. Since EEG signals are highly non-stationary, it is more appropriate to use time-frequency methods for coherence analysis of scalp Laplacian. Thus this research applies the wavelet transform on coherence analysis of scalp Laplacian. To verify our technique, already recorded EEG data of event related potentials were obtained from a study of two large groups of alcoholic and abstinent alcoholic subjects, performing visual picture-recognition tasks. The proposed coherence method successfully detected time-frequency correlation between EEG signals with minimum VCUS effects. It showed significant spatial specificity and revealed detailed coherence patterns. Some new important results regarding time-frequency characteristics of VCUS effects on wavelet and short-time Fourier transform (STFT) coherence analysis of REEG signals were deduced. The proposed coherence method was also compared to a conventional wavelet coherence method of REEG signals in the study of coherence difference between coherences of alcoholic and abstinent alcoholic EEG signals. Results of this study provided substantial evidence that VCUS effects are not additive and therefore can not be ignored in comparison of different brain states between groups of subjects.