Prediction and Detection of Epileptic Seizure

Introduction

Seizure is a neurological disorder due to sudden surge of electrical activity in the brain caused by structural abnormalities of the brain, encephalitis, lack of oxygen in the brain, brain injury, tumor, and some sort of dysfunctionalities of the brain (Sanei et al. 2007). Epilepsy is characterized by spontaneously recurrent seizures (Parvez et al., 2012). More than 50 million (i.e., 1% of world population) individuals are diagnosed with epilepsy (Santaniello et al., 2011). Approximately 5% of whole populations are experienced a seizure in their life time (Netoff, et al., 2009). Epileptic patients have intractable seizures which are likely to increase damage to neural tissue. Epilepsy can be controlled by medication in the most cases. Epilepsy causes many injuries such as fractures, submersion, burns, accidents and even death. It is highly possible to prevent this unwanted situation if we correctly and timely predict epilepsy before the actual seizure onset.

In the brain, neurons exploit chemical reaction to generate electricity to control different bodily actions and this ongoing electrical activity can be recorded graphically which is popularly known as Electroencephalogram (EEG). EEG is well accepted tool for epileptic seizure prediction/detection that can measure the voltage fluctuations of the brain (Dorr, et al., 2007; Rosso et al., 2003; Esteller et al., 2004; Guttinger et al., 2005; and Tang et al., 2012). Feature extraction, analysis, and classification of EEG signals are still challenging issues for researchers due to the variations of the brain signals. Variations of EEG signals depend on different brain locations, number of channels, and different patterns of signals from different people. Another challenge for epileptic seizure detection/prediction from EEG signals is to get reasonable accuracy for real time applications.

Generally we can describe a seizure detection technique by a number of steps such as noise/artefacts reduction using raw ictal, interictal, and non-seizure EEG signals, feature extractions, feature reduction, and classification using selected features. On the other hand, we can describe a seizure prediction technique by all steps of detection technique with two extra steps such as regularization of classification outcome and then decision making. The prediction technique normally classifies pre-ictal EEG signals to identify ictal signals from interictal or non-seizure EEG signals. As the similarity between pre-ictal and interictal signals are least compared to that between ictal and interictal signals, the prediction is more challenging compared to detection.

In this chapter we will discuss the fundamental steps (such as artifact removal, features extraction, features selection, classification, regularization of noisy output, and decision function, etc.) of the prediction and detection of epileptic seizure through the analysis of EEG signals. We will also discuss (i) the characteristics of EEG signals for healthy people and epileptic seizure patients, (ii) existing features extraction techniques with their advantages and limitations, (iii) existing classification techniques with their advantages and disadvantages, (iv) other challenging issues in near future for emerging technologies, and (v) future trend of research for high accuracy in detection and prediction.