Automatic Seizure Detection Using S-transform And Singular Value Decomposition

Epilepsy is a chronic brain disease caused by paroxysmal and excessive neuronal discharges. It is reported that approximately 1% of the world’s population suffers from epilepsy. Long-term and repeated seizures not only cause serious damage to the patients’ physical and mental health, but also bring a heavy burden to their family and society. Electroencephalography (EEG) is an important tool for the diagnosis of epilepsy, which contains abundant physiological and pathological information that provides a effective basis to locate epileptogenic zone. At present, EEG recordings are usually inspected by experts visually to identify seizure activities according to their experience, which is a time-consuming and heavy task due to that the amount of EEG data is very large. As a result, automatic seizure detection technology has great significance to reduce the burden of medical staff and improve the diagnosis efficiency.There have been various kinds of automatic seizure detection methods proposed in recent years, and among them, the method based on feature extraction in combination with classifier has been widely applied. The commonly used EEG features include two categories, linear and nonlinear features. Linear features include fluctuation index, differential variance, relative amplitude and so on, and the nonlinear features include approximate entropy, Lyapunov exponent and correlation dimension, etc. Classifiers play an important role in seizure detection, and some of the commonly used are support vector machine, artificial neural network, and Boosting algorithm, etc. In addition, time-frequency analysis methods have also been applied for seizure detection, such as short-time Fourier transform, wavelet transform and Wigner-Ville distribution.This paper presents a novel seizure detection method based on S-transform and singular value decomposition (SVD) on the basis of predecessors’researches. Specific steps of the method is as follows:S-transform is firstly performed on EEG signals for time-frequency analysis, and the obtained time-frequency matrix is divided into submatrices; Then the singular values of each submatrix are extracted using singular value decomposition (SVD), and 4-dimension feature vectors are constructed by adding the largest singular values in the same frequency band together; Finally, effective feature vectors are fed into BLDA classifier for decision and postprocessing is applied to obtain higher classification accuracy. The EEG data used in this study was obtained from the Epilepsy Center of the University Hospital of Freiburg, and we used EEG recordings containing 82 seizure events from 20 patients to evaluate the system. Results show that the proposed method has a lower computation complexity and achieves a higher sensitivity and a lower false detection rate, which is effective for seizure detection and has certain clinical application value.