Smart sensor for EEG acquisition and epileptic seizure detection

Abstract

This project investigates a novel and hybrid system for seizure detection and prediction by attempting to build a real time EEG analyser that strives to detect and predict seizures based on the EEG waveforms. With this goal in sight, system proposed herein is developed using signal processing techniques in time and frequency domain and using an innovative machine learning technique and tested using the CHB-MIT scalp EEG database to simulate real live patients. The system offers an environment where users can test this model with minimal effort while allowing them to alter any parameters deemed fit. Evaluation of the proposed system shows that analysis of EEG signals can be useful in detecting and predicting epileptic seizures in real time. Using the signal processing technique an average sensitivity of 100% and an average specificity of 73.65% were attained. The signal processing technique combined with the novel machine learning technique achieved an average sensitivity of 72.4% and an average specificity of 52.3%. These techniques obtained an average warning window of 74s prior to the seizure onset. Even though this study is still in its infancy, the results obtained are promising and further research is justified in order to identify the best way forward in the direction of a real life practically useful seizure detection and prediction machine.