Epileptic seizure classification using shifting sample difference of EEG signals

This paper proposes a novel lightweight shifting sample difference method for efficient epileptic seizure detection from electroencephalogram signals. Unlike most recent seizure detection methods that use complex signal transformations, the shifting sample difference method is based on time-domain and does not require any transformation. The epilepsy detection performances using five popular electroencephalogram features (statistical measures, Hjorth parameters, fractal dimensions, approximate entropy, and sample entropy) are investigated in this study. The proposed shifting sample difference method outperforms widely used discrete wavelet transform and empirical mode decomposition based features in three classification problems from the Bonn university epilepsy dataset. Accuracies of 99%, 98%, and 100% are obtained for normal vs. inter-ictal, inter-ictal vs. ictal, and normal vs. ictal classifications. The highest accuracy and reduced computational complexity show the potential scope of proposed shifting sample differences in epilepsy diagnosis.