Detection of Spikes with Multiple Layer Perceptron Network Structures

In this work, the spikes in the electroencephalogram (EEG) signals are analyzed by using artificial neural networks (ANN). Multiple layer perceptron (MLP) networks utilizing between 3 and 15 hidden neurons are used in the network architecture. For training the MLP network backpropagation algorithm, backpropagation with adaptive learning rate, Levenberg-Marquardt (LM) algorithm, early stopping and regularization methods are used. Principal components of feature vectors obtained from 41 consecutive sample values of each peak are used for training the networks. Performances of classifiers are examined for two cases depending on both sensitivity-specificity and sensitivity-selectivity properties.