Spike separation is a basic prerequisite for analyzing of the cooperative neural behavior and neural code when registering extracellu-larly. Final performance of any spike sorting method is basically defined by the quality of the discriminative features extracted from the spike waveforms. Here we discuss two features extraction approaches: the Principal Component Analysis (PCA), and methods based on the Wavelet Transform (WT). We show that the WT based methods outperform the PCA only when properly tuned to the data, otherwise their results may be comparable or even worse. Then we present a novel method of spike features extraction based on a combination of the PCA and continuous WT. Our approach allows automatic tuning of the wavelet part of the method by the use of knowledge obtained from the PCA. To illustrate the methods strength and weakness we provide comparative examples of their performances using simulated and experimental data.
展开▼
