Epileptic Seizure Detection Based on Local Mean Decomposition and Dictionary Pair Learning

Abstract

The automatic detection of epileptic Electroencephalogram (EEG) signals is of great significance in alleviating the pressure of neurologists and the clinical therapy of patients. In this paper, a novel method combining the local mean decomposition (LMD) and the projective dictionary pair learning (DPL) is proposed for seizure detection. The LMD is employed to decompose the EEG signals into a set of product functions, and the first three product functions are selected to extract three features to characterize the behavior of EEG recordings. After that, the DPL model can learn a synthesis dictionary and an analysis dictionary from the training samples, and then encode the test samples over the learned dictionary. At last, the test samples are categorized by checking whose class has the minimum reconstructed residual. The proposed model is evaluated on the Freiburg database, achieving the sensitivity of 95.89% with the specificity of 95.10%. The experimental results suggest that the proposed method can become a potential approach for detecting seizures in clinic application.