Towards Classifying Epileptic Seizures Using Entropy Variants

Abstract

Epilepsy is the second most common neurological disorder, affecting about 50 Million people of the world's population. Entropy is a nonlinear measure suitable for studying chaotic behavior and randomness, suits well to analyze perceptual time series data such as EEG data. However, the EEG data are non-stationary and prone to numerous noise types that badly affect the classification accuracy of epileptic seizures. To address these issues this paper introduces a novel index for classifying epileptic seizure to extract the hidden information from recorded EEG signal to discriminate seizure states. The experimental results demonstrate that the proposed method shows better anti-noise performance compared to the conventional State of art entropy variants i.e. Approximate Entropy (ApEn), Sample Entropy (SampEn) and Permutation Entropy (PE). Compared to other variants that are quite sensitive to noise, the proposed method maintains its higher accuracy of 95.3% and AUC among the recorded EEG data of 21 subjects. The discriminative abilities of entropy variants were further tested using student t-test. The results on The Freiburg EEG database proved the superiority of the proposed index over the existing state-of-the-art variants. With the excellent classification performance and low computational complexity, PFuzzy entropy can be utilized for practical seizure classification and epilepsy detection in future hardware implementation and also opens future opportunities towards real-time detection and prediction of epileptic seizures.