Analysis of local field potential signals: a systems approach

Efficient methods for Local Field Potential (LFP) signal analysis amenable to interpretation are becoming increasingly relevant. LFP signals are believed, in part, to reflect neural action potential activity, and LFP frequency modulations are linked to spiking events. Furthermore, LFP signals are increasingly accessible in human brain regions previously unreachable due to a proliferation of deep brain stimulation implantation procedures. Traditional LFP analysis involves computing power spectra densities (PSDs) of these signals, which captures power at various frequencies in the signal. However, PSDs are second order statistics and may not capture non-trivial temporal dependencies that exist in the raw data. In this paper, we propose an LFP analysis method that is useful for describing unique features of temporal dependencies in LFP signals. This method is based on autoregressive (AR) modeling and draws from the systems identification sub-field of systems and control. Specifically, we have built and analysed AR models of LFP activity, and have demonstrated statistically significant differences in temporal dependencies between diseased globus pallidus tissue and control regions in two dystonia patients receiving deep brain stimulation implantation. Differences in the PSDs of LFP signals between these two groups were not statistically significant.