Classification of Ictal and Preictal Seizure using EEG Signals based on Convolutional Neural Network

Epilepsy is a neurological disorder in which normal brain activity is affected. Electroencephalography (EEG) is a gold standard to predict epilepsy seizures. Manual inspection of EEG signals to detect various seizure phases is a major challenge therefore, need for some mechanism to automate seizure prediction to ease the work of clinicians is required. Moreover, Epilepsy patients suffer difficulties in social gatherings as the seizure is unpredictable which can cause anxiety, fear, etc in them. To overcome these challenges, automatic seizure prediction is vital. The present work encompasses the channel selection and data augmentation methods to create a system that automatically classifies two phases of seizure-ictal and preictal using a 1D Convolutional Neural Network (CNN) model. Instead of using all channels of EEG signals, a subset of channels is used in this work to analyze its effect on performance. Data augmentation is used to increase the datasets which got limited due to channel selection. The model achieves the best performance with the help of the 1D CNN model. It achieves an accuracy of 99.62%, specificity of 99.76%, and sensitivity of 99.70% on the CHB-MIT dataset. This work shows that a subset of channels with greater importance can improve the robustness of the seizure prediction system. This in turn saves the time to set up the electrodes on the scalp of the patients. Various data augmentation schemes can be used to compensate for the limited dataset. Usage of models like 1D CNN are suitable for designing low-power seizure prediction systems.