Separating burst from background spikes in multichannel neuronal recordings using return map analysis.

We propose a preprocessing method to separate coherent neuronal network activity, referred to as “bursts”, from background spikes. High background activity in neuronal recordings reduces the effectiveness of currently available burst detection methods. For long-term, stationary recordings, burst and background spikes have a bimodal ISI distribution which makes it easy to select the threshold to separate burst and background spikes. Finite, nonstationary recordings lead to noisy ISIs for which the bimodality is not that clear. We introduce a preprocessing method to separate burst from background spikes to improve burst detection reliability because it efficiently uses both single and multichannel activity. The method is tested using a stochastic model constrained by data available in the literature and recordings from primary cortical neurons cultured on multielectrode arrays. The separation between burst and background spikes is obtained using the interspike interval return map. The cutoff threshold is the key parameter to separate the burst and background spikes. We compare two methods for selecting the threshold. The 2-step method, in which threshold selection is based on fixed heuristics. The iterative method, in which the optimal cutoff threshold is directly estimated from the data. The proposed preprocessing method significantly increases the reliability of several established burst detection algorithms, both for simulated and real recordings. The preprocessing method makes it possible to study the effects of diseases or pharmacological manipulations, because it can deal efficiently with nonstationarity in the data.