Phase-coherence classification: a new wavelet-based method to separate local field potentials into local (in)coherent and volume-conducted components

Abstract

Local field potentials (LFP) reflect the integrated electrophysiological activity of a large group of neurons. To minimize influence of external activity on the analysis, conventionally bipolar recordings are used to eliminate volume-conducted signals. Here we introduce a novel method, called phase-coherence classification (PCC), to separate LFP in time-frequency domain into a volume-conducted, a local incoherent and local coherent signal. The PCC allows to compute the power spectral densities of each signal and to associate each class with possible locations of electro-physiological activity. In order to test the resolution properties and accuracy of the method we generate composite and non-stationary synthetic time series with similar statistical characteristics as measured LFP. The PCC identifies volume-conducted signals with a phase threshold that is determined from probability density functions of non-phase-shifted synthetic time series. We estimate optimal PCC parameters for the analysis of beta band oscillations in LFP and apply the PCC to a test data set obtained from within the subthalamic nucleus of eight patients with Parkinsons disease (PD). We show that PCC can identify activity of multiple local clusters during a tremor episode and quantify the relative power of local and volume-conducted signals. We further analyze the electrophysiological response to an apomorphine injection during rest and show that incoherent activity in the low beta band shows a significant medication-induced decrease. We further find significant movement-induced changes on medication of the local coherent signal, which increased during an isometric hold task and decreased during phasic wrist movement. This indicates a different role of incoherent and coherent signals possibly related to physiologically different networks. This new PCC method can potentially also be applied to EEG and MEG data in order to minimize the influence of spatial leakage on power spectra and coherence estimates.