Hardware Implementation of Automated Seizure Detection System using EEG Signals and Edge Computing

Epilepsy is an unrelenting, constantly recurring neurological disorder characterized by abnormal electrical activity in the brain. Epilepsy acts as a condition with inconsistent symptoms involving spasmodic abnormal electrical activities in the brain. Clinical data pertinent to such abnormalities is nonlinear, non-stationary, and multi-dimensional. The processing and analysis of such data generates a combination of computing research challenges. Electroencephalogram (EEG) is one of the most clinically and scientifically utilized signals recorded from human brain. It is a powerful source for providing valuable insight of the brain dynamics. Accurate and careful analyses of EEG signals play a prominent role in diagnosis of brain disorders such as epilepsy. EEG signal has exceptional temporal resolution, usability, and non-invasive recording set-up with low costs which makes it popular in the field of biomedical research. Currently, trained neurologists take diagnostic decision by observing EEG recordings visually to decide probable treatment plan. The visual inspection of non-stationary and high-dimensional EEG recordings is a time-consuming task and results in errors. An automated seizure detector is an essential system not only for the better understanding of seizure generation but also a boon for medical professionals who are subjected to fatigue in visualizing a continuous and enormous long-term recording. The multichannel automated seizure detection is a bulky system as it involves recording from at least 16 channels. The use of such system on subjects who are remotely located is a challenge. In this paper, a unique embedded system based on envelop generation technique while sifting of EEG signals by Ensemble Empirical Mode Decomposition (EEMD) has been proposed to classify seizure signals. The modifications in the algorithm have resulted in an equivalent performance on multivariate signals when compared to simulation results. The use of 5-channel EMOTIV® headset and Raspberry Pi embedded system using python programming language has resulted in classification accuracy of 88.6% on multivariate signals. The portable embedded hardware may support the neurologists in automated identification of epileptic seizures.