Time-Varying Mutual Information Analysis of Evoked in Vivo Local Field Potentials in Rodents.

The local field potential (LFP) is an extracellular electrophysiology signal corresponding to the local activity of neurons and is commonly used in neuroscience to characterize local oscillatory networks of cortical activity. Here, we use the LFP as a measure of the neuronal response evoked by transcranial pulsed stimulation. Traditionally, LFP correlation analysis is used to measure synchronization between brain regions but is unable to capture nonlinear relationships. We propose a mutual information matrix analysis tool to study the nonlinear interactions between the LFPs from different neuron groups and characterize the effect of electrical stimulation on LFPs. The mutual information matrix has many similar properties to the correlation matrix. However, it captures additional nonlinear relationships between the variables. This makes it a complement to the correlation matrix for highly nonlinear neural activity. The mutual information matrix analysis is extended to timevarying signals to capture the nonstationarity of neural activity. As a proof of concept, the analysis is applied to in vivo rodent electrophysiology data, showing that the mutual information matrix is stable during spontaneous activity (d(M-0, M-t) < 2.5) and can capture time-varying effects of stimulation, i.e., increased synchronization (d(M-0, M-t) > 14.0) in the gamma frequency range lasting for about 800 ms after electrical stimulation. This time-varying mutual information analysis has the potential to complement the traditional correlation matrix analysis for neural activity.