A novel multivariate phase synchrony measure: Application to multichannel newborn EEG analysis.

Phase synchrony assessment across non-stationary multivariate signals is a useful way to characterize the dynamical behavior of their underlying systems. Traditionally, phase synchrony of a multivariate signal has been quantified by first assessing all pair-wise phase relationships between different channels and then, averaging their phase coupling. This approach, however, may not necessarily provide a full picture of multiple phase ratios within non-stationary signals with time-varying statistical properties. Several attempts have been made to generalize pair-wise phase synchrony concept to multivariate signals. In this paper, we introduce a new measure of generalized phase synchrony based on the concept of circular statistics. The performance of the measure is evaluated with simulations using the Kuramoto and Rössler models and compared with that of three existing generalized phase synchrony measures based respectively on 1) the concept of co-integration, 2) S-estimator and 3) hyper-dimensional geometry. The simulation results represent the correct degree of synchronization between channels with negligible mean of squared error, i.e. below 4.2e−4. We then use the proposed measure to assess inter-hemispheric phase synchrony in two abnormal multichannel newborn EEG datasets with manually marked seizure/non-seizure and burst–suppression signatures. The EEG results suggest that the proposed measure is able to detect inter-hemispheric phase synchrony changes with higher accuracy than the other existing measures, i.e. 2% for seizure/non-seizure database and 11% for burst/suppression database than the best performing existing multivariate phase synchrony measure.