A novel automated seizure detection system from EMD-MSPCA denoised EEG: Refined composite multiscale sample, fuzzy and permutation entropies based scheme

This paper investigates three complexity measures namely, refined composite multiscale sample entropy (RCMSE), refined composite multiscale fuzzy entropy (RCMFE), and refined composite multiscale permutation entropy (RCMPE) as features for the automated detection of epileptic seizures from electroencephalograms (EEGs). Generally, the EEG signals contain unwanted frequency components and superimposed trends that may influence their complexity evaluation. Therefore, we propose a denoising technique based on empirical mode decomposition (EMD) and multiscale principal component analysis (MSPCA) called EMD-MSPCA, and explore its impact on the performance of RCMSE, RCMFE, and RCMPE features for seizure diagnosis. Additionally, we put forward a novel automated seizure detection methodology based on EMD-MSPCA denoised EEG and combined RCMSE, RCMFE, and RCMPE features to characterize healthy, seizure-free, and seizure EEG signals. The experimental results demonstrate that all the three entropy features can successfully characterize the abnormal dynamics related to epileptic EEG signals with RCMPE being the best feature; applying the proposed EMDMSPCA denoising technique prior to feature extraction using RCMSE, RCMFE and, RCMPE not only improved the performances of various classifiers but also reduced the computational time of these three entropy features significantly; and the proposed seizure detection scheme yielded good classification accuracies on two widely used EEG databases as compared to state-of-the-art works, hence emerges as a robust model for automated detection of epileptic seizures.