Detection and Prediction of Epileptic Seizure by Analysing EEG Signals

An epileptic seizure is a brief episode of symptoms due to abnormal excessive or synchronous neuronal activity in the brain caused by structural abnormalities, encephalitis, lack of oxygen, injury, tumour, and some dysfunctions of the brain. More than 65 million individuals are diagnosed with epilepsy (i.e., spontaneous and recurrent seizures) worldwide. Approximately 325 million of the world population experience a seizure in their life time. Electroencephalogram (EEG) is a widely used tool to diagnose a possible epileptic seizure. Automated seizure (i.e., epileptic seizure) detection and prediction techniques have great potential in epilepsy monitoring, diagnosis, and rehabilitation. Much research has been devoted to detecting and predicting seizures by analysing EEG signals. The
detection and prediction accuracy of the existing techniques are not accurate enough due to the challenges in terms of non-abrupt phenomena and inconsistent signals in different brain locations with different patients, types (general/partial) of seizures, and hospital settings. To overcome the
limitations of the existing methods, generic seizure detection approaches have been developed through innovative feature extraction of ictal (actual period of seizure onset) and interictal (period between two adjacent seizures) EEG signals by identifying a specific range of frequencies, which can distinguish ictal and interictal signals using various established transformation and decomposition
techniques. The experimental results show that the proposed detection methods outperform the state of-the-art methods in terms of accuracy, sensitivity, and specificity with greater consistency using large benchmark data set in different brain locations. Real time seizure prediction techniques have also been developed through extracting innovative spatial and temporal features from preictal (before seizure onset)/ictal and interictal EEG signals. The experimental results reveal that the proposed prediction methods provide high prediction accuracy with low false alarms compared to the state-ofthe-art methods considering a wide range of patients. The proposed theoretical contributions may provide an opportunity to develop a clinical device to predict forthcoming seizures in real time applications.