Automatic Seizure Detection Algorithm Based On Earth Movers’ Distance In EEG

Epilepsy is a chronic disease and paroxysmal syndrome, which results from sudden abnormal electrical activities of neurons. Electroencephalogram(EEG) is an efficient method for clinical diagnosis of epilepsy, and it provides accurate information for doctors to locate epileptogenic zone for effective treatment. Because of long term EEG monitoring resulting in the enormous EEG data, and the EEG analysis’s dependency on the doctor’s clinical experience, the seizure detection task is time consuming and heavy. Seizure automatic classification and identification is particularly urgent and important.EEG consists of phase, amplitude, frequency and other components. Seizure detection is defined as analysing some components’ time series and spatial distribution characteristics of seizure signals and non-seizure signals based on digital EEG, and finding the distinguishing feature for classification. At present, numbers of efficient EEG signals researching and analyzing are used in seizure detection work. After looking through all of them, this paper proposes a new seizure detection algorithm based on Support Vector Machine(SVM), wavelet decomposition and Earth Movers’ Distance(EMD). The experiment results show that the algorithms proposed can divide the seizure signals from normal signals efficiently.The EEG data used in this study was obtained from the Freiburg EEG database, containing continuous EEG recordings of21patients suffering from medically focal epilepsy and87seizures. It was recorded using128channels and a16-bit A/D converter, and depth-electrodes surgically inserted inside the brain or placed on the cortex of the patients. This method can be divided into three steps:First, the EEG signals were segmented into epochs and the wavelet decomposition was used to remove the high frequency artifacts of EEG. Second, EMD-L1of each epoch was extracted as the feature. Finally, the features were fed to the classifier and some post-processing technologies were used, such as smoothing, logical judgement and collar technology. Post-processing is used to achieve two objectives. First, it rejects false seizure detections including short-length detections and artifacts. Second, it improves the seizure detection rate.By analyzing the experimental results, we can get the conclusion:the algorithm based on EMD-L1and SVM achieves a high sensitivity, specificity and low false detection rate. It is an efficient automatic seizure detection methods.