A Convolutional Neural Network for Seizure Detection
Omar Kaziha, Talal Bonny
Abstract
In this paper, a software-based neural network is developed for the purpose of detecting seizures from raw EEG signals. Detecting epileptic seizures manually is a long tedious process, creating the need for automatic detections systems. A neural network is designed based on a convolutional neural network (CNN) and trained on the electroencephalogram (EEG) raw signal dataset “CHB-MIT”. Deep learning has not been fully explored in seizure detection, but rather only classical machine learning algorithms that need feature extraction. There is a need to explore deep learning which eliminates manual feature extraction and enables real-time detection of raw signals. We conducted training and inference on the CHB-MIT dataset with a designed CNN in software and achieved an accuracy of 96.74%. The classifier can later be transformed into a portable system on chip (SoC) by realizing it on reconfigurable hardware with the necessary peripherals for acquisition.