Epileptic Seizure Detection Based on Multidimensional Feature Extraction

Abstract

Epilepsy is an acute and recurrent neurological disease. Its main clinical symptoms are paroxysmal convulsions, sensory disorders, consciousness and mental disorders, which seriously affect the daily life and work of patients. Seizure detection has always been a challenging work. With the increase of the incidence of epilepsy, high-performance epilepsy automatic detection algorithm can reduce the workload of medical workers in clinic, and has important significance in clinical medical research. This paper combines the characteristics of frequency domain and time domain with nonlinear analysis characteristics to extract features of EEG signals. Firstly, the average Periodic diagram method is used to calculate the power spectral density features, and then the wavelet transform is used to obtain the time domain analysis features. Next, the permutation entropy of five different rhythms are obtained by permutation entropy and information entropy algorithms. Finally, the three features with different dimensions are input into a linear classifier as a single feature to identify seizure and non-seizure signals. The performance of the proposed detection algorithm is evaluated by experiments on the epileptic EEG data-set of the University of Bonn. Experimental results show that the proposed method has a high classification accuracy, which can reach 95. 8%.