Search

• A novel scheme was proposed for automated seizure detection using electroencephalography (EEG). • Subbands and cumulative sum of subbands were raised and their effect was investigated. • 12 cases were studied and accuracies of nearly 100% and 93.62% for the first 11 and 12th cases, respectively. We here proposed a novel fusion method of frequency slice wavelet transform (FSWT)-based fuzzy entropy (FuzzyEn) and Higuchi's fractal dimension (HFD), t-distributed stochastic neighbor embedding (t-SNE) and K-nearest neighbor (KNN) to achieve the goal of automated detection of epileptic electroencephalography (EEG). Fifteen EEG subbands were first separated from the original EEG via FSWT, then their cumulative sum of subbands (CSoS) including forward CSoS and backward CSoS were computed. FuzzyEn and HFD were then employed to quantify the nonlinear properties of these subbands and CSoS, and followed by mapping the high-dimensional features into the two-dimensional space using t-SNE. Finally, the two-dimensional features were fed into KNN for classification. Eight binary-class, three ternary-class and one quinary-class cases, in total twelve classification tasks of the Bonn EEG database were conducted. The proposed approach attained accuracies of approximately 100% and 93.62% in the first 11 and 12th cases, respectively. Experimental results not only manifest our proposal is superior to most of existing methods, but also demonstrate subbands and CSoS based nonlinear features enhance the performance of epileptic seizure detection. [ABSTRACT FROM AUTHOR]