Patient-Specific Epileptic Seizure Onset Detection Algorithm Based on Spectral Features and IPSONN Classifier

This paper proposes a patient-specific epileptic seizure onset detection algorithm. In this algorithm, spectral features in five frequency bands (Δ, α, ß, θ and γ) is extracted from small frames of seizure and non-seizure EEG signals by applying Discrete Wavelet Transform (DWT) and Discrete Fourier Transform (DFT). These features can create the maximum distinction between two classes. Then a neural network (NN) classifier based on improved particle swarm optimization (IPSO) is used to determine an optimal nonlinear decision boundary. This classifier allows adjusting the parameter of the NN classifier, efficiently. Finally, the performance of algorithm is evaluated based on three measures, sensitivity, specificity and latency. The results indicate that the proposed algorithm obtain a higher sensitivity and smaller latency than other common algorithms. The proposed algorithm can be used as a seizure onset detector to initiate the just-in time therapy methods.