Spikes Sorting Based On Manual Neural Network Of Neurons Cultured On Multi-Electrode Arrays

The signals of many neurons in a local region were picked up by electrode, in the multi-unit extracellular recordings. Identification and classification of spike events were the first step in all multiple spike train data analyses. High levels of correlated neural noise and superpositions of waveforms were the two difficulties in spike sorting.The manual neural network has become more and more important as a classifier. Learning from the environment adaptively and generalizing were the most advantages of neural network. The radial basis function manual neural network in this paper is a local approximate network, as its basis function in the hidden layer is a local function. When the input signal is near the center of the function, the output of the hidden note is large, whereas the output is approach zero.In this paper, at first spikes of raw data were detected using a revised peak-detection scheme. And then presorted by principle components analysis. The spikes, which have the smaller square error to the center of each classification, were selected as the train set of radial basis function neural network. The manual network was trained and then identifies the spikes to different classifications. Simulation results showed that when the SNR debase or the firing frequency increase, the radial basis function neural network performed more efficiently than the principle components analysis cluster and the shape parameter cluster.And extracellular recording of hippocampal neural network on the multi-electrode arrays also use these three methods for classification. The inter spike intervals show that radial basis function network performed well on the real raw data.