Epileptic Seizure Detection With Eeg Textural Features And Imbalanced Classification

Epilepsy is a chronic and recurrent neurological disorder resulting from sudden and excessive electrical discharges in a group of brain neurons.It can lead to loss of consciousness or awareness,disturbances of movement,sensation,mood,or mental function affecting patients’ lives seriously.Therefore,the Epilepsy seizure detection has an important practical significance for the diagnosis of epilepsy and the quality of patients’ lives.EEG plays an important role in the epilepsy seizure detection.Conventional seizure detection depends on the visual inspection of long-term EEG recordings by neurologists.However,visual inspection is a time-consuming and tedious task because of the increasing amount of EEG data.Hence,automatic seizure detection would be advantageous and can efficiently speed up the inspection process which is of great importance in the epilepsy diagnosis.The traditional detection method has been used to classify the database based on the number of seizure samples equal to the non-seizure samples.Imbalance data classification is a challenging task in automatic seizure detection from EEG recordings when the durations of non-seizure periods are much longer than those of seizure activities as a classic imbalanced question.The result of the imbalanced database in the balanced classifier is inevitably bias to the non-seizure,but the seizure samples may be failed to the non-seizures.This paper proposes an imbalanced learning model to improve the identification of seizure events in long-term EEG signals.To better represent the underlying microstructure distributions of EEG signals while preserving the non-stationary nature,a discrete wavelet transform(DWT)and uniform 1D-LBP feature extraction procedure and GLCM are introduced.A learning framework is then proposed by the ensemble of weakly trained support vector machines(SVMs).Under-sampling isemployed to split the imbalanced seizure and non-seizure samples into multiple balanced subsets where each of them is utilized to train an individual SVM classifier.The weak SVMs are incorporated to build a strong classifier which emphasizes seizure samples while perceiving the imbalanced class distribution of EEG data.Final seizure detection results are obtained by a multi-level decision fusion process by taking into account temporal and frequency factors.The EasyEnsemble imbalanced classification algorithm is as follows: firstly,the raw EEG data is divided into 4-s epochs which is decomposed by DWT for obtaining representing frequency band and denoising.Then the uniform 1D-LBP textural features and GLCM are extracted in the selected frequency band.Thirdly,Under-sampling is employed to split the imbalanced seizure and non-seizure samples into multiple balanced subsets where each of them is utilized to train an individual classifier.The weak classifiers are incorporated to build a strong classifier which emphasizes seizure samples.Lastly,final seizure detection results are obtained by a multi-level decision fusion process.The result validated the model over public two long-term and one short-term EEG databases.Our model achieved G-mean of 97.14%with respect to epoch-level assessment and sensitivity of 96.67% with false detection rate 0.86/h in term of event-based evaluation on the intracranial EEG database.An epoch-based G-mean of 95.28% and event-based false detection rate of 0.81/h were yielded over the scalp EEG database.The comparisons with 14 published methods demonstrated the improved detection performance for imbalanced EEG signals and the generalizability of our model.