Epileptic Seizure Detection Using a Small Memory Footprint Convolutional Neural Network

Abstract

Automatically detecting epileptic seizures has been explored using machine learning techniques, with promising results in recent years. Epilepsy affects about 51 million people worldwide, and about 30% of them cannot control their symptoms, leading to unpredictable seizures. This means that approximately 15.3 million people continue to experience seizure episodes even after treatment (e.g., drugs or surgery). Many previous studies have focused on developing models for patient-specific tasks. This study explores the problem of seizure detection using a cross-patient approach. We present a technique for generating and preprocessing data from EEG signals to be used as input to a Convolutional Neural Network (CNN). The network architecture and performance are described, achieving results comparable to state of the art, with an average accuracy of around 95%. The proposed network has five convolutional layers, making it a good candidate for embedding in a wearable device. Details of the method and training hyperparameters are also provided.