EEG single-channel seizure recognition using Empirical Mode Decomposition and normalized mutual information

In this document features taken from Empirical Mode Decomposition (EMD) are selected by mutual information for the discrimination between letal and Seizure-Free EEG single-channel signals. Some features are based on the instantaneous or average frequency and amplitude of each EMD component. Also, skewness, kurtosis and Shannon´s entropy are taken as features from the energy obtained using the Teager Energy Operator (TEO). TEO is calculated over each EMD component. Then a subset of relevant and non-redundant features is selected by normalized mutual information. Finally these selected features are used to train a linear Bayes classifier, and a 5-fold cross validation is performed for different clinical cases. We used a publicly available database to compare each feature extraction approach. Accuracies around 98% are reached by the implemented methodology.