Epilepsy seizure detection in electroencephalogram signals using robust time space analysis feature extraction method with DLNN classifiers

Abstract

Electroencephalogram records the electrical brain activity and it is an indispensable tool for diagnosing epileptic seizure. Manually reviewing Electroencephalogram (EEG) recordings for detection of epilepsy pattern is a time-consuming process and hence it is, necessary to automate the epileptic seizure detection. It is important to develop seizure detection and prediction methods to assist clinicians. Seizure progression is a dynamic and non-stationary process and the EEG signals are comprised of many frequency bands. The objective of this work is to determine features that differentiate epileptic seizure from a normal activity. EEG recording of epileptic seizure EEG segments and non-seizure EEG segments are used for the study. In order to evaluate the seizures four statistical moment parameters such as mean, standard deviation, skewness, and kurtosis and the three spectral moments such as Spectral Centroid, Variational Coefficient and Spectral Skew are extracted. This thesis work proposes a method for the analysis of EEG for seizure detection and prediction. newlineIn the first stage of the research, Akima Spline Interpolation based Ensemble Empirical Mode Kalman Filter is used for the decomposition of EEG signal analysis. For effective analysis of Epilepsy seizure, University of BONN Germany single channel dataset have been used and it is processed at a sampling rate from 250 Hz to 2500 Hz. The seizure activity has generally been considered to be associated with and#120575;, and#120579; and and#120572; band. Hence the averaged spectral poweris computed from 0.5 Hz to 14 Hz EEG band. This is performed for 25 non seizure EEG segments and 25 EEG segments with seizure condition. newline