Epileptic Seizure Prediction from EEG Signals Using DenseNet

Abstract

Epilepsy is a disorder in which the normal electrical pattern in the brain is disrupted causing seizures or loss of consciousness. Seizure is harmful during various events like swimming or driving. The electroencephalogram (EEG) is the measurement of electrical activity received from the nerve cells of the cerebral cortex. Forthcoming seizures can be predicted from scalp EEG signal to improve the quality of life. The study proposes a method of automatic epileptic seizure prediction from raw EEG signal. The raw EEG signal is converted into EEG signal image for automatic extraction of features and classification of inter-ictal and pre-ictal state using Dense Convolutional Network (DenseNet). This classification process is carried out in a manner similar to the process followed by a medical practitioner without resorting to hand-crafted features. The public CHB-MIT EEG database is used for training, validation, and testing. An EEG signal for 1 second duration is taken as one sample. The accuracy for the classification of inter-ictal and pre-ictal state is achieved up to 94% by using 5-Fold cross validation. However, the accuracy is not up to the mark for the presence of common artifacts caused by eye-blinking and muscle activities during EEG recordings. Hence, a 30 seconds pool based technique is used for decision on correct state identification. The proposed pool based technique provides an average specificity of 95.87% and a false prediction rate of 0.0413/hour. It also provide average sensitivities of 100%, 97%, and 90% for the time slots 0 - 5 minutes, 5 - 10 minutes, and 10 - 15 minutes before the seizure event.