Epileptic EEG signal analysis and identification based on nonlinear features

In this paper, two non-linear complexity measures, namely approximate entropy and sample entropy are investigated as feature extraction methods for evaluating the regularity of the epileptic EEG signals. Furthermore, in order to obtain more efficient feature extraction for EEG signals, an optimized algorithm for sample entropy measure (O-SampEn) is proposed which removes the calculation redundancy and optimizes the computation procedure for sample entropy measure. Clinical EEG data was obtained from 20 intracranial electrodes placed within the epileptogenic zone in five epilepsy patients during both interictal and ictal periods. In terms of the experimental results, both sample entropy and approximate entropy analysis show lower values during epileptic seizures, which mean an increase of EEG signal regularity during ictal state. Compared with approximate entropy, the feature extraction based on sample entropy measure is more sensitive to EEG signal variety caused by epileptic seizures, approximately 10.14%~20.02% higher than the results using approximate entropy. In addition, the proposed optimized algorithm for sample entropy can run 9.52~36.16 times faster than the original sample entropy algorithm according to the simulation. High discrimination ability and fast computation speed of the proposed optimized sample entropy algorithm demonstrate its huge potential as a novel feature extraction method for real-time epileptic seizure detection.