An EEG based real-time epilepsy seizure detection approach using discrete wavelet transform and machine learning methods. (August 2022)

• Record Type: Journal Article
• Authors: Shen, Mingkan; Wen, Peng; Song, Bo; Li, Yan
• Is Part Of: Biomedical signal processing and control. Volume 77(2022)
• Publication Date: 2022-08
• Subjects: EEG -- Real-time seizure detection -- Discrete wavelet transform -- Support vector machine -- RUSBoosted tree Ensemble; 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.2022.103820
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: DB4-DWT and DB16-DWT were proposed to extract approximate and details of signals and remove redundant information. Improved the robustness of EEG based epilepsy detection via machine learning methods. Proposed a method that can achieve 97% accuracy and 96.67% sensitivity in 3-class classification (health control, seizure free and seizure active) using Dataset UB, and 96.38% accuracy, 96.15% sensitivity and 3.24% false positive rate in the real-time seizure detection using Dataset CHB-MIT. Implemented an automatic seizure detection approach in real-time way. Abstract: Epilepsy is one of the most common complex brain disorders which is a chronic non-communicable disease caused by paroxysmal abnormal super-synchronous electrical activity of brain neurons. This paper proposed an electroencephalogram (EEG) based real-time approach to detect epilepsy seizures. Discrete wavelet transform and eight eigenvalues' algorithms are applied to extract features in different sub-frequency bands. Then support vector machine is employed for three-classes classification of health control, seizure free and seizure active, and finally RUSBoosted tree Ensemble method is used for real-time seizure onset detection. The proposed algorithm is evaluated using two public datasets: one short-term dataset named UB and one long-term dataset named CHB-MIT. The results show that the algorithm achieves 97% accuracy and 96.67% sensitivity in the three-classes classification of health control, seizure-free and seizure-active groups in UB dataset, and 96.38% accuracy, 96.15% sensitivity, 3.24% false positive rate for the real time seizure onset detection in CHB-MIT Dataset.