Classification of Both Seizure and Non-Seizure Based on EEG Signals Using Hidden Markov Model

Abstract

In this paper, we propose a novel feature extraction method, a slope of counting wavelet coefficients over various thresholds (SCOT) method based hidden markov model (HMM) for seizure detection. The purpose of the proposed method is to aid in the diagnosis of epilepsy, which requires long-term electroencephalography (EEG) monitoring. The interpretation of long-term EEG monitoring takes a lot of time and requires the assistance of experienced experts. In order to overcome these limitations, it is important to apply the optimized feature extraction algorithm to the seizure detection system. The proposed SCOT method based HMM has a robust detection accuracy, and a short feature extraction time; whereas the existing methods require a large amount of training data and a long feature extraction time for learning the seizure detection model. Experimental result shows that with the proposed method, the average detection accuracies are 96.5% and 98.4% using the HMM in seizure and non-seizure, respectively. In addition, the proposed method has robust detection performance regardless of the given window sizes (0.15, 0.25, 0.5, 1, and 2 seconds) are used.