Wearable Real-Time Epileptic Seizure Detection and Warning System

Abstract

Epilepsy is an unpredictable neuronal brain disorder, affecting approximately 70 million people. It is characterized by seizures resulting in the patient losing the capability of controlling his/her actions. Besides being prone to injuries, losing consciousness, and control of the body, patients have a higher risk of experiencing sudden unexpected death in epilepsy (SUDEP). Therefore, continuous multimodal monitoring using electrodermal activity (EDA) and accelerometer (ACC) sensors can help in detecting epileptic seizures based on the seizure types and symptoms. EDA was used to detect emotional activities, while ACC detected physical activities. This paper describes the design and implementation of a real-time wearable epileptic seizure detection and warning system to monitor 12 epilepsy patients during their daily activities at home. The acquired EDA and ACC signals from the patients’ body were sent continuously to the detection and warning subsystem, where it was continuously processed and analyzed. The later block can also automatically alert the parent/caregiver of the patient over the cellular network in case of a seizure event. Among the various machine learning algorithms, support vector machine (SVM) and bagged decision tree classifiers yielded the highest accuracy of 86.9% and 90.7%, respectively, for ACC and EDA data individually. However, for fused ACC-EDA data, the bagged decision tree showed the highest accuracy of 96.7% in detecting epileptic seizures. It was found that fused ACC data with EDA helped to distinguish epileptic onset from daily activities reliably with a very low false alarm rate.