Signal processing of neural discharges using intensity-based methods.
Brent Edwards · Deep Blue (University of Michigan) · 1992
Neural discharge data are studied under the signal processing domain using a point process intensity model. Describing the intensity function of the point process signal with a driving function and a recovery function, the post-stimulus time (PST) histogram is modeled as an autoregressive (AR) system. The AR parameters of the system are a function of the driving function, recovery function and time quantization. Using the AR model, the PST histogram can be predicted with knowledge of the driving and recovery functions. AR parameter estimation techniques are used to estimate both the constant driving function magnitude and the recovery function from the PST histogram. This estimation is counter to the common assumption that interval information is lost in the PST histogram. The AR model of the histogram leads to a spectral analysis of neural discharges. The spectra of both the spike train and the PST histogram during steady-state firing are predicted using AR spectral techniques. The locations of spectral peaks are determined by the recovery function while the magnitudes of the peaks are determined by the firing rate. The spectral derivations are verified using eighth nerve data. Again, these results are counter to the common assumption that the spectra of the steady-state portion of the PST histogram and spike train are flat. Finally, an efficient method is developed for estimating connectivity between pairs of neurons in a neural ensemble. Three new intensity functions are introduced which each lead to a different connectivity detection technique. Each technique is analyzed using ROC analysis and performance is derived as a function of connectivity strength and data sample size. The application of these computationally simple techniques results in a significant reduction of the number of neural pairs in an ensemble which need to be analyzed for connectivity estimation.