IoT based Efficient Epileptic Seizure Prediction System Using Deep Learning
Hisham Daoud, Phillip Williams, Magdy Bayoumi
Abstract
Epilepsy is a neurological disorder that affects about 50 million people around the globe. Epileptic seizure prediction reduces the risk associated with the sudden occurrence of the seizure that might threaten a patient’s life. The current seizure prediction methods are computation-intensive due to extracting the complicated hand-crafted features and require large memory for storing their parameters, which make them inappropriate for constrained IoT and wearable devices. In this paper, we introduce an IoT framework for accurate epileptic seizure prediction system based on deep learning. In the proposed system, the feature extraction and classification stages are combined in one integrated system in which raw EEG signals are applied without any preprocessing which further decreases the computation complexity. We developed a Convolutional Neural Network (CNN) based model that extracts the important spatio-temporal features from the non-stationary and nonlinear EEG signals. A channel selection algorithm is proposed to reduce the complexity and the memory required for the system to accommodate the real-time application. An alarm to an incoming seizure is generated and sent to the doctor and any chosen emergency service. Also, the history of the patient is being uploaded to the cloud to be revised by the doctor periodically. The high accuracy of the proposed prediction system of 96.1% makes our system a good candidate for IoT based wearable seizure prediction device.