Evaluation of techniques for predicting seizure Build up

TL;DR

This study compares filtering techniques for detecting pre-ictal gamma oscillations in EEG data to improve seizure prediction, finding Despikifying more robust than stationary wavelet transforms.

Contribution

It introduces a comparative evaluation of SWT and Despikifying for seizure build-up prediction using time-frequency and spatio-temporal analysis.

Findings

Despikyfying is more robust than SWT in reconstructing pre-ictal gamma oscillations.

Both techniques can detect oscillations, but Despikifying offers better prediction accuracy.

The study demonstrates the effectiveness of preprocessing methods in seizure prediction.

Abstract

The analysis of electrophysiological signal of scalp: EEG (electroencephalography), MEG (magnetoencephalography) and depth (intracerebral EEG) IEEG is a way to delimit epileptogenic zone (EZ). These epileptic signals present two different activities (oscillations and spikes) which can be overlapped in the time frequency plane. Automatic recognition of epileptic seizure occurrence needs several preprocessing steps. In this study, we evaluated two filtering techniques: the stationary wavelet transforms (SWT) and the Despikifying in order to extract pre ictal gamma oscillations (bio markers of seizure build up). Then, we used a temporal basis set of Jmail et al 2017 as a preprocessing step to evaluate the performance of both technique. Moreover, we used time-frequency and spatio-temporal mapping of simulated and real data for both techniques in order to predict seizure build up (in time and space). We concluded that SWT can detect the oscillations, but Despikyfying is more robust than SWT in reconstructing pure pre ictal gamma oscillations and hence in predicting seizure build up.