| For analysis of multivariate time series, like e.g. multi-channel EEG, the equal-time cross-correlation is a classic and computationally efficient measure for quantifying linear interrelations between data channels. When the cross-correlation coefficient is estimated using a finite amount of data points, its system-specific non-random part may be strongly contaminated by a sizable random contribution, such that no reliable conclusion can be drawn about genuine mutual interdependencies. The random correlations are determined by the amount of data points used and the signals' frequency content. It may vary dramatically, especially when the power spectra of the signals change. Here, we introduce methods for EEG analysis that can be employed to study non-random correlations independently of the signals' frequency content. First, scalar measures that allow to quantify random and non-random correlations in the global multi-channel EEG are introduced [1]. Then we extend our work to matrices that allow analyzing spatial patterns of genuine cross-correlation regardless of confounding influences of random correlations also locally [2]. The problem of mixing of random with non-random correlations is explained at the example of non-stationary EEG. Then the performance of the proposed methods is illustrated using model systems with known interdependence patterns. Finally, the benefit of these methods is demonstrated by its application to intracranial EEG of epilepsy patients. References: [1] M. Müller, G. Baier, C. Rummel and K. Schindler, Estimating the strength of genuine and random correlations in non-stationary multivariate time series, Europhys. Lett. 84, 10009 (2008) [2] C. Rummel, M. Müller, G. Baier, F. Amor and K. Schindler, Analyzing spatio-temporal patterns of genuine cross-correlations, submitted (2009) |
|