Epileptic Seizure Detection

Abstract

The purpose of our research project was to use machine learning (ML) and deep learning (DL) algorithms to develop a dedicated RNN model that could analyze live data as a prototype for a light-weight and effective EEG-based home-care system that can immediately detect and report oncoming seizures in patients with chronic seizure disorders, including petit and grand mal epilepsy. A seizure involves excessive nerve and neuron signaling that creates an overload of electro-biochemical activities in the brain resulting in involuntary, autonomic sensations, behaviors, and even physical movements ranging from momentary or complete cessation of breathing and consciousness to dramatic to uncontrolled intellectual and/or physical activity that endanger patients and those around them. Some seizure disorder conditions can be effectively treated; others even cured. Preventative care is essential, however, to prevent or minimize every seizure’s potential damage to the brain. Because it can track how brain cells send and receive electric waves to and through the nodes of the brain, the electroencephalogram (EEG) is the main method of monitoring for seizure events; furthermore, EEG records provide important insights into the types and processes of seizures. Utilizing a Muse 2 headband to read EEG waves and a Raspberry Pi 4 microcomputer to process and provide clean, organized data, the research team wrote machine and DL algorithms to enable a light-weight, portable EEG device which can easily be adapted to any wireless or cable telehealth monitoring device and can identify, record, analyze, and report electrical brain waves within five (5) seconds, before, during, or after a seizure so that physicians, family, and other caregivers are alerted promptly to provide immediate care.