Predicting Pre-ictal State of Epileptic Seizures from Electroencephalography Signals

Abstract

Predicting the occurrence of epileptic seizures can provide an enormous aid to epileptic patients and enhance their lives' quality significantly. In this thesis, we have examined different approaches of seizure prediction applied to scalp Electroencephalography (EEG) signals. We have introduced a new method that relies on the count of zero-crossings of wavelet detail coefficients of EEG signals as the major feature. This is followed by a binary classifier that discriminates between preictal and interictal states. We have introduced new enhancements to seizure prediction methods making them more practical for real-time applications through reducing the computational complexity as we use an adaptive algorithm for channel selection to identify the optimum number of needed electrodes. Another enhancement is using shorter amount of training data to enhance the setup process of real time applications. In addition, the method was demonstrated to be robust against variations between seizures for the same patient. Applied to data from 8 patients, the proposed method achieved high accuracy and sensitivity with an average accuracy of 94% and an average sensitivity of 96%. These results were obtained using only 10 minutes of training data as opposed to using hours of recordings typically used in traditional approaches. We studied the effects of changing wavelet functions, wavelet levels, different dimensionality reduction approaches, and using different classifiers, on the accuracy. Moreover, we have examined the efficiency of using autoencoders as a channel reduction approach; using different autoencoders number of levels, encoder and decoder activation functions, and different