A novel ensemble local graph structure based feature extraction network for EEG signal analysis. (August 2020)

• Record Type: Journal Article
• Authors: Tuncer, Turker; Dogan, Sengul; Ertam, Fatih; Subasi, Abdulhamit
• Is Part Of: Biomedical signal processing and control. Volume 61(2020)
• Publication Date: 2020-08
• Subjects: Ensemble local graph structure (E-LGS) -- Electroencephalogram (EEG) -- Epilepsy -- Epileptic seizure prediction and detection; 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.2020.102006
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Electroencephalogram (EEG) used to diagnose brain diseases such as epilepsy. A novel feature extraction network based on local graph structure (LGS) is utilized. LGSs are utilized for feature extraction and 2D-DWT is used for pooling. ReliefF and neighborhood component analysis (NCA) are used for feature selection. Abstract: Electroencephalogram (EEG) signals have been extensively utilized to identify brain disorders such as epilepsy. In this study, a novel feature extraction network based on local graph structure (LGS) is utilized for EEG signal classification. The aim of this work is to create a framework which utilize ensemble of LGS that uses logically extended LGS, symmetric LGS, vertical LGS, vertical symmetric LGS, zigzag horizontal LGS, zigzag horizontal middle LGS, zigzag vertical LGS and zigzag vertical middle LGS. By using these LGS methods with discrete wavelet transform (DWT), a novel ensemble feature extraction network is formed. In this framework, LGSs are utilized for feature extraction and 2D-DWT is utilized for pooling. In the feature reduction phase, two widely known feature reduction techniques, namely ReliefF and neighborhood component analysis (NCA) are used together. Five different benchmark classifiers are employed to present the strength of the proposed ensemble feature extraction framework. In the experiments, two publicly available EEG datasets have been employed to test the proposed ensemble LGS feature extraction based multilevel EEG signal classification method. The proposed ensemble LGS method achieved 97.20% and 98.67% success rate for these datasets. Six cases were also examined to comprehensively evaluate the used Bonn dataset. Results clearly illustrated the success of the ensemble LGS based EEG classification method.