Extracellular Spike Detection Approaches for Noisy Neuronal Data using Singular Spectrum Analysis

Multiple neuron activities in an active brain can be recorded using microelectrode arrays (MEAs) and analyzed. This requires an accurate detection of the neural spikes. In this paper, three approaches based on fractal dimension (FD), smoothed nonlinear energy operator (SNEO) and standard deviation to detect the spikes for noisy neuronal data are proposed. These methods however do not perform well in some cases, especially when the noise level is high or there are many noise sources. Although discrete wavelet transform (DWT) is an influential tool widely used in biomedical signals, it does not separate the components which overlap in time-frequency space. It doesn’t have sufficient speed either. Therefore, we suggest a powerful subspace-based filtering approach, namely, singular spectrum analysis (SSA) as the preprocessing step to reduce the noise. When SSA followed by SNEO is used, the average detection of spikes and CPU time are more desirable and are respectively 100% and 0.072 s for semi-real signals of SNRs as low as 5 dB