Nonlinear vector decomposed neural network based EEG signal feature extraction and detection of seizure

Abstract

Electroencephalography is one of the important medical methods to evaluate and treat neurophysiology to combat disease related to seizure. The automatic seizure detection system aims to provide a balanced mechanism by excavating deep knowledge of the basic signal and kinetic domains. Contingent alert is given to initiate treatment to reduce the risk of injury in patients with epilepsy and to overcome contingencies. Moreover, the multi-channel Electroencephalogram (EEG) data of seizure detection in conventional machine learning algorithms cannot effectively accommodate both global and spatial information. Therefore in this work proposed a Nonlinear Vector Decomposed Neural Network (NVDN) to detect the seizure from EEG signal. The proposed NVDN based seizure detection system consist of three major stages they are as follows i) EEG preprocessing ii) Feature Extraction and iii) NVDN Classification. In this work, the NVDN technique is applied to improve the accuracy of seizure detection after using frequency domain feature to obtain the results from EEG waves. The performance of the proposed system is validate through MATLAB simulation. The results of the simulation show that the proposed NVDN method is able to effectively detect the seizure with a sensitivity of 94.7%. Specificity of 94.1% and accuracy 95.60%. As compared with conventional methods the proposed system achieve high detecting ratio.