Pathological Electroencephalographic Signals Classification by Using Multi-Resolution Analysis and Neural Network

In this study is proposed a method based in multiresolution analysis in frequencies strip provided by discreet wavelet transform (DWT), characterizing epileptic discharges of absence crisis and also noises, both analyzed into distinct frequencies strip. The methodology uses the DWT, integrated to the auto-regressive (AR) model and backpropagation network (MLP) to compose the classificator. First, the multi-resolution analysis technique (DWT) and the AR are applied to extract the time-frequency distribution characteristics from the signal in different levels. The neural network MLT with the specialist system, classify the characteristics extracts to identify the kind of disturbance occurred in EEG. In this proposal, occurs a significant reduction of the number features extracts from the signal, without losing its original proprieties. The global performance of the proposal method shows consistent results.