Sparse representation-based EMD and BLDA for automatic seizure detection

Epilepsy is a serious chronic neurological disorder, which affects more than 50 million people worldwide, and automatic seizure detection on EEG recordings is extremely required in the diagnosis and monitoring of epilepsy. This paper presents a novel seizure detection method using sparse representation-based Earth Mover's Distance (SR-EMD). In the proposed algorithm, wavelet decomposition is executed on the original EEG recordings with five scales, and the scales 3, 4 and 5 are selected to structure the distributions of EEG signals. Then, the Gaussian mixture models (GMMs) of EEG signals are estimated and the distances between GMMs are computed using SR-EMD as EEG features. After that, EEG features are sent to Bayesian linear discriminant analysis classifier for classification. To improve the detection accuracy, the post-processing procedure is employed finally. The long-term intracranial EEG dataset with 21 patients is used to evaluate the performance of the method, and the satisfactory sensitivity of 93.54 %, specificity of 97.57 % and false detection rate of 0.223/h are achieved. The results indicate that SR-EMD is more effective and efficient than the conventional Earth Mover's Distance (EMD). Moreover, the good performance and fast speed of this algorithm make it suitable for the real-time seizure monitoring application.