Abstracts

Rationale: It has been suggested that seizure initiation is due to significant alterations in neuronal synchrony. However, the role synchrony plays in this process is not well understood. One measure of this synchrony is the coherence between single neuron activity and the local field potential within specific frequency bands, generally referred to as spike field coherence (SFC). This relationship is important because changes in local field potentials and neuronal firing are the defining features of seizures. Therefore, there is a need to develop an accurate and precise measure of this synchrony in order to characterize its role in seizure initiation. A popular method for performing SFC estimation involves measuring the power spectrum of a spike-triggered local field potential (LFP) average, and normalizing this to the average power spectrum of the LFP. However, this method suffers from an inherent bias problem. That is, as the number of estimation trials increases, the average estimated SFC over trials does not approach the true SFC due to the finite number of spikes used in each trial. In fact, the measured value of the coherence will be different for different numbers of spikes used in the estimation. Methods: By examining simple probability models of the LFP and spike times, the true SFC was analytically derived as a linear function of both the estimated SFC and the number of spikes. This function allowed us to define a correction for the bias. Our bias correction was tested using simulated spike time distributions applied to sine wave models of LFP or experimentally recorded LFP. The correction was then tested on experimentally collected data. Finally, we show how this bias correction method can be applied to assess changes in SFC during the initiation of spontaneous seizures in a chronic rat model (kainic acid) of epilepsy. Results: Our results show that this bias correction reduced the bias but generally increased the variance of the estimated coherence. To evaluate the overall impact of these two sources of error, the mean squared error was found and shown to be reduced or unchanged for the data tested. Conclusions: Because the number of spikes recorded can vary greatly between subjects and even within a subject across seizures, a method of bias correction is critically important. Without this correction it would not be possible to compare the estimated coherence across these seizures. This is especially true for clinical assessment of SFC due to the need to pool data from multiple clinical trials. The proposed bias correction enables SFC estimates to be compared between data sets of any size.