Automated identification of epileptic seizures in EEG signals based on phase space representation and statistical features in the CEEMD domain

• Published in: Biomedical signal processing and control Vol. 38; pp. 148 - 157
• Main Authors: Jia, Jian, Goparaju, Balaji, Song, JiangLing, Zhang, Rui, Westover, M. Brandon
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.09.2017
• Subjects: Complete ensemble empirical mode decomposition; EEG; Epileptic seizure detection; Random forest classifier; Complete ensemble empirical mode decomposition; Epileptic seizure detection; Random forest classifier; EEG
• ISSN: 1746-8094; 1746-8108
• DOI: 10.1016/j.bspc.2017.05.015

•We present an automated method for detecting epileptic seizures in electroencephalogram (EEG) signals based on complete ensemble empirical mode decomposition.•The concept of the growth curve is adapted for feature extraction for first time in the epileptic seizure detection.•Efficacy of the method is confirmed by statistical and graphical analyses.•Performance of the proposed scheme when compared to the state-of-the-art algorithms is promising. Epileptic seizure detection based on visual inspection by expert physicians is burdensome, and subject to error and bias. In this work, we present a novel method for the automated identification of epileptic seizure using a single-channel EEG signal. We utilize the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) technique to devise an effective feature extraction scheme for physiological signal analysis, and construct the corresponding growth curve. Then, various statistical features are extracted from the growth curve as the feature set, and this is fed to the random forest classifier for completing the detection. The suitability of the extracted features is established through statistical measures and graphical analysis. The proposed method is evaluated for the well-known problem of classifying epileptic seizure and seizure-free signals using a publically available EEG database from the University of Bonn. To assess the performance of the classification method, 10-fold cross-validation is performed. Compared to state-of-the-art algorithms, the numerical results confirm the superior algorithm performance of the proposed scheme in terms of accuracy, sensitivity, specificity, and Cohen’s Kappa statistics.