I-Corps: Epileptic Seizure Detection System

Arizona State University, Scottsdale AZ

Investigators

Abstract

The broader impact/commercial potential of this I-Corps project will benefit epileptic patients by improving diagnostic systems for doctors, freeing them from manual scanning of continuous electroencephalogram results. This is expected to bring about significant reduction of cost in health care and enable novel commercial applications of monitoring of epileptic patients. The nonlinear analysis technique applied in this project has the potential to open up new applications in neuroscience and may prove to be essential in recognizing how and why seizures occur as well as other neurological disorders. The high resolution of the approach also indicates a potential in identifying seizure precursors, which would better assess seizure susceptibility and ultimately lead to seizure suppression through external intervention. This I-Corps project further develops a reliable and automated seizure detection mechanism which will potentially aid medical practitioners and patients with epilepsy. The established practice for patients suspected of epilepsy is a manual process, highly labor-intensive, and prone to human error. The technology developed here addresses these problems by utilizing techniques from chaotic systems and nonlinear measures that more accurately describe brain activity and yield significantly more accurate results in seizure detection using EEG data compared to standard linear measures. It combines state-of-the-art solutions from both algorithms and computing hardware. At the heart of the method, there is a recently developed algorithm for the estimation of Lyapunov exponents, a measure of chaoticity, that was shown to provide unparalleled resolution and noise immunity. This method is computationally expensive. To adhere to the real-time constraints of the application, the project employs parallelizable platforms such as multi-core CPUs/GPUs. Furthermore, complete automation will be achieved by testing out a number of shape detection algorithms through the use of neural networks and wavelet matching. The method has been tested on extensive records from animal models and some human data.