Automatic focal and non-focal EEG detection using entropy-based features from flexible analytic wavelet transform. (March 2020)

• Record Type: Journal Article
• Authors: You, Yang; Chen, Wanzhong; Li, Mingyang; Zhang, Tao; Jiang, Yun; Zheng, Xiao
• Is Part Of: Biomedical signal processing and control. Volume 57(2020)
• Publication Date: 2020-03
• Subjects: Electroencephalogram (EEG) -- Focal (F) and non-focal (NF) -- Flexible analytic wavelet transform (FAWT) -- Entropy -- Classifier; 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.2019.101761
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: We explore the ability of FAWT without reconstruction analysis in the feature extraction of epilepsy signals. Fuzzy distribution entropy is firstly introduced to discriminate the focal and non-focal EEG signals, and a satisfied classification performance is achieved using the combination of log energy entropy and fuzzy distribution entropy. In order to obtain more reliable results, different classifiers have been considered in this work. The entire Bern Barcelona database is used to verify the proposed method which makes the results more statistically significant. Abstract: Surgical treatment is one of the most important methods to cure or control drug-resistant epilepsy, and preoperative localization of epileptic lesions plays an important role in the success of a surgery. Given that the manual diagnosis takes time and effort, an automatic detection system is needed to aid clinical diagnosis. Therefore, in the present study, a new automatic focal electroencephalogram (EEG) detection algorithm combining flexible analytic wavelet transform (FAWT) with entropies was put forward. The differential focal (F) and non-focal (NF) EEG signals were decomposed into 15-level sub-bands using FAWT, and this was followed by computing log energy entropy (LEE) and fuzzy distribution entropy (fDistEn) of the detail coefficients of 15 sub-bands and the differential EEG signal. Kruskal–Wallis one-way analysis of variance (ANOVA) was adopted to select the statistically significant features, and five classifiers including general regression neural network (GRNN), support vector machine (SVM), least squares support vector machine (LS-SVM), K-nearest neighbor (KNN), and fuzzy K-Nearest neighbors (fKNN) were then used to verify the effectiveness of the selected features. The proposed methodology was tested on the Bern Barcelona database, and a maximum accuracy of 94.80 % was achieved in distinguishing F and NF EEG signals via LS-SVM classifier. The results suggest that the proposed method is a valuable approach to aid clinicians in locating the epileptic focus in practical application.