چكيده به لاتين
Today, epilepsy has affected many babies around the world. Electroencephalogram signal monitoring is one of the methods to evaluate the neural function of epileptic patients. In this study, we used to determine the presence or absence of epilepsy in all seconds of the electroencephalogram signal separately using effective connectivity criteria. The data used in this study are electroencephalogram signals taken from 79 infants, all of whom had a clinical suspicion of seizures and were admitted to the intensive care unit. In the first step, multivariate auto regressive models with all the degrees in one interval are fitted to the individual signal, and finally the best degree of the model is estimated. The best model degree is the degree to which Schwartz-Bayesian criterion is minimized. In the next step, the signal is divided into small windows (1 second) without overlap and the multivariate auto regressive model is fitted to them with the best model degree and the model coefficients are obtained. Then effective connectivity criteria are obtained by model coefficients for each window. From the obtained effective connectivity matrices, 4 criteria (208 attributes) are calculated and finally, using them, the seconds of individuals (epileptic and non-epileptic) are classified through the linear discriminant analysis classifier. In the end, the results of all people are averaged. In this study, the mean accuracy for partial directional coherence matrix was 74.33% and for directional transfer function matrix was 72.08%. After combining the properties with the temporal properties, the classification accuracy results were obtained 83.76 for the partial directional coherence matrix and 83.05 for the directional transfer function matrix.
