A Comparison of Two Synchronization Measures for Neural Data

Abstract

In this paper we analyze the capability of two different signal synchronization measures to detect synchronized periodic signal components. We compare the periodic synchronization index (PSI), which was first introduced for the detection of epileptic seizures in EEG signals, with the widely known phase locking value (PLV). The PSI automatically detects the dominant periodic signal components utilizing Fourier series expansions and measures the synchronization between equivalent components in two signals. The PLV averages the difference between the analytic phases of two signals. Applying these two measures to synthetic data and EEG recordings we found clear advantages of the PSI over the PLV. Experiments with synthetic data and high signal to noise ratio (SNR) revealed no significant difference between the PSI and the PLV in their ability to detect small frequency deviations. However, in the presence of noise the detection performance of the PLV drops significantly. In particular, we found that the PLV relies on an SNR of nearly 3dB. In contrast, the performance with the PSI remained almost unchanged for SNR down to -4dB. This noise robustness of the PSI is due to the incorporation of harmonic oscillations resulting in an improved estimation of periodic signal components. When applied to the detection of early seizure phenomena in EEG data we found that the increase of the PSI at seizure onset is more pronounced and the variance of the measure during pre-ictal time periods is lower than for the PLV.