Classificação de estágios do sono pela análise do sinal de EEG

Abstract

This paper addresses the problem of automatic classification of sleep stages by analyzing a signal channel electroencephalogram (EEG). The study was conducted using the MIT-BIH Polysomnographic Database, available from Physionet. The main aspects of this work are feature extraction and classification of signal patterns. In this sense, two adaptive techniques, commonly for characterization of signals, are tested: the first one based on wavelet packet and the second one using the Kalman filter. For the first case it was evaluated three classification techniques: Support Vector Machine, K-Nearest Neighbors and Multilayer Perceptron Neural Network. For the second case, it was selected the Hidden Markov Models in order to perform the classification. For each situation, the classification results are compared with the ratings given by a specialist, using metrics commonly used in the areas of automatic analysis of EEG and Pattern Recognition. The results, here presented and discussed, reinforcing the use of automatic classification of sleep stages.