Seizures classification based on higher order statistics and deep neural network. (May 2020)

• Record Type: Journal Article
• Authors: Sharma, Rahul; Pachori, Ram Bilas; Sircar, Pradip
• Is Part Of: Biomedical signal processing and control. Volume 59(2020)
• Publication Date: 2020-05
• Subjects: EEG -- Seizure -- Higher order statistics -- Autoencoder -- Deep neural network; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2020.101921
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: The seizures are the predominant symptom of epilepsy. The third order cumulant (ToC) is used to initially extract the features EEG signals. The sparse autoencoder is used to extract the structural information from the ToC features. The sparse autoencoder based neural network gives up to 100% accuracy. Abstract: The epileptic seizure is a transient and abnormal discharge of nerve cells in the brain that leads to a chronic disease of brain dysfunction. There are various features-based seizures classification algorithms listed in the literature. But, there is no standardized set of attributes that can perfectly capture the relevant information regarding the signal dynamics. In this paper, a computationally-fast seizure classification algorithm is presented. The obtained results through the proposed algorithm are consistent and repeatable. This paper describes an automated seizures classification technique using the nonlinear higher-order statistics and deep neural network algorithms. The sparse autoencoder based deep neural network is used to extract the essential structural details from the third-order cumulant coefficients matrix. The proposed algorithm achieves a reliable classification accuracy for both categories, i.e., binary classes and three-classes of electroencephalogram (EEG) signals with the softmax classifier. The proposed study is simulated on the publicly-available Bonn university EEG database. The achieved results show the effectiveness of the proposed algorithm for seizures classification.