Statistical Method for Detection of Phase-locking Episodes in Neural Oscillations

In many networks of oscillatory neurons, synaptic interactions can promote the entrainment of units into phase-coupled groups. The detection of synchrony in experimental data, especially if the data consist of single-trial runs, can be problematic when, for example, phase entrainment is of short duration, buried in noise, or masked by amplitude fluctuations that are uncorrelated among the oscillating units. In the present study, we tackle the problem of detecting neural interactions from pairs of oscillatory signals in a narrow frequency band. To avoid the interference of amplitude fluctuations in the detection of synchrony, we extract a phase variable from the data and utilize statistical indices to measure phase locking. We use three different phase-locking indices based on coherence, entropy, and mutual information between the phase variables. Phase-locking indices are calculated over time using sliding analysis windows. By varying the duration of the analysis windows, we were able to inspect the data at different levels of temporal resolution and statistical reliability. The statistical significance of high index values was evaluated using four different surrogate data methods. We determined phase-locking indices using alternative methods for generating surrogate data and found that results are sensitive to the particular method selected. Surrogate methods that preserve the temporal structure of the individual phase time series decrease substantially the number of false positives when tested on a pair of independent signals.