Spike sorting based on approximate entropy

Blind source separation for spike sorting is the foundation of microelectrode array recordings. Spikes initiation and propagation in dendrites with stochastic ion channels contained rich nonlinear dynamic properties. In contrast to the differences of complexity between non-homologous spikes, high-dimensional features were extracted based on approximate entropy firstly. Then, the features were selected and the spikes were projected to the two-dimensional feature space. Finally, K-means cluster method was used to realize the pattern classification of spikes. The results of simulation and experiment show that the features extracted by approximate entropy as the classification basis is utility to discriminate between non-homologous spikes.