Classification of EEG signals for epileptic seizure prediction using ANN

In this paper, we developed a model for classification of EEG signals. The aim of the study is to determine whether this model can be used for epileptic seizure prediction if “pre-ictal” stages were successfully detected. We analyzed long-term Freiburg EEG data. Each of 21 patients contains datasets called “ictal” (seizure) and “inter-ictal” (seizure-free). We extracted 4096-samples (or 16 seconds) long segments from both datasets of each patient. These segments were decomposed into time-frequency representations using Discrete Wavelet Transform (DWT). The statistical features from the DWT sub-bands of EEG segments were calculated and fed as inputs to Multilayer Perceptron (MLP) and Radial Basis Function (RBF) network classifiers using 10-fold cross validation. We also applied multiscale PCA (MSPCA) de-noising method to determine if it can further enhance the classifiers’ performance. MLP-based approach outperformed RBF classifier with or without MSPCA, which significantly improved the classification accuracy of both classifiers. The proposed MLP-approach with MSPCAachieved a classification accuracy of 95.09%. We showed that a high classification accuracy of EEG signals can be accomplished in cases when additional “pre-ictal” class is introduced. Therefore, the proposed approach may become an efficient tool to predict epileptic seizures from EEG recordings. Keywords: Electroencephalogram (EEG); Epileptic seizure; Discrete Wavelet Transform (DWT); Multilayer Perceptron (MLP); Radial Basis Function (RBF) network; Multiscale PCA (MSPCA); Machine learning.