Integration of stationary wavelet transform on a dynamic partial reconfiguration: case study separating preictal gamma oscillations from transitory activities for early build up epileptic seizure

TL;DR

This paper presents a method using stationary wavelet transform on a dynamic reconfiguration platform to efficiently separate preictal gamma oscillations from transitory activities, enabling faster early seizure detection.

Contribution

It introduces a novel implementation of stationary wavelet transform on a dynamic partial reconfiguration platform for real-time separation of neural signals related to seizures.

Findings

Faster seizure build-up recognition by about 40 times using the proposed method.

Effective separation of gamma oscillations from transitory activities in large datasets.

Potential for real-time seizure monitoring and neurofeedback applications.

Abstract

To define the neural networks responsible of the epileptic seizure, we had to study the electrophysiological signal in a proper way. The early recognition of the seizure build up could also be defined through the time space mapping of the preictal gamma oscillations. The electrophysiological signals present three types of wave: oscillations, spikes, and a mixture of both. Recent studies prove that spikes and oscillations should be separated efficiently to define the accurate neural connectivity for each activity. However retrieving the transitory activity is a sensitive task due to the frequency interfering between the gamma oscillatory and the transitory activities. Many filtering techniques are highlighted to ensure a good separation: reducing false oscillations in the transitory activity and vis versa. However and for a big data set, this separation necessitate a big consumption in time execution, this constraint will be overcome using embedded architecture. The integration of these filtering techniques would also lead to creating instantaneous monitoring of the seizure recognition, build up and neurofeedback devices. We propose here to implement the stationary wavelet transform as a convenient filtering technique to keep only the preictal gamma oscillations on a partial dynamic configuration, then we will use the same architecture to integrate the time space mapping for an early recognition of the build up seizure. We proved a faster recognition of the build up seizure through the non contaminated preictal gamma oscillations time space mapping (about 40 times faster), obtained by the integration of the wavelet transform.