CNN based framework for detection of epileptic seizures

Epilepsy is a common neurological disease that uses electroencephalogram (EEG) data for its detection purpose. Neurologists make the diagnosis by visual inspection of EEG reports. As it is time-consuming and due to the shortage of specialists worldwide, researchers have proposed automated systems to detect the disease. In the past decade, most of the systems were designed using hand-engineered features. However, identifying appropriate features is always a challenging task in the development of a seizure detector system. Deep learning networks eliminate the problem of selecting the best features but suffer from long training time, generally days or weeks. To overcome this problem, the authors have proposed a new 1D convolutional neural network (CNN) that automatically extracts features at an average of seven epochs, only followed by traditional machine learning (ML) classifier. 1D CNN architectures are intrinsically suitable for the processing of EEG time-series data. The proposed model doesn't require any preprocessing of EEG signal and results in approximately 94% reduced training time than end-to-end deep learning models. Different ML techniques have been applied to extracted features to check the robustness of the proposed 1D CNN. Maximum accuracy of 99.83% has been achieved by most of the classifiers to detect between healthy and seizure patients. The reduced number of processing steps and epochs makes it suitable for real-time clinical applications.

Automated summary

Researchers propose a new one-dimensional convolutional neural network (CNN) model for detecting epilepsy, which automatically extracts features and significantly reduces training time. The model achieves a high accuracy in distinguishing between healthy and epileptic patients, as validated by applying various machine learning techniques.