A SEIZURE DETECTION METHOD BASED ON WELL-SOLVED NONLINEAR AND NON-STATIONARY PROBLEMS WITH ELECTROENCEPHALOGRAPHIC SIGNALS

Abstract

The main focus of this paper is to solve the nonlinear and non-stationary problems in electroencephalographic (EEG) signals, which has been solved by the proposed method by using convolutional neural networks (CNN) as the classifiers and assembling Local Mean Decomposition (LMD) and cepstral coefficients as the feature extraction methods to achieve epileptic seizure detection with signal analysis and processing. In this proposed method, LMD and cepstral coefficients have been employed to solve the nonlinear and non-stationary problems in feature extraction and infusion, and then, the feature can be employed to feed to the recognition engine named CNN, and finally, the epileptic seizure detection can be achieved by this step. Publicly available EEG database from the University of Bonn (UoB), Germany had been used to verify the effectiveness and robustness of this proposed method on feature extraction. The complete dataset of total 7960 EEG segments, three recognition problems marked as AB versus CD versus E, the average classification accuracy of these segments can be generally obtained as highly as 99.84%, the maximal classification accuracy is 99.87%, and the lowest recognition accuracy is 98.74%. To the best of our knowledge, the excellent performance of the proposed method has shown that this method can be employed to track the patient’s healthy state and monitor the moment of epilepsy seizure.