Localization of epileptic seizure with an approach based on the PSD with an autoregressive model

TL;DR

This paper introduces an automatic method for localizing epileptic seizures using power spectral density analysis of EEG signals processed with autoregressive models, achieving 70% detection accuracy and 80.55% sensitivity.

Contribution

The study presents a novel PSD-based approach combined with AR modeling for automatic seizure detection, improving frequency domain resolution over traditional visual methods.

Findings

Detection accuracy of 70%

Sensitivity of 80.55%

Effective localization of epileptic activity

Abstract

In this study, we present a criterion based on the analysis of EEG signals through the mean of the conventional power spectral density (PSD) in the aim to localize and detect the epileptic area of the brain. Firstly, as the EEG signals are commonly non stationary in practice, we processed the data with technique of differentiation in order to have the stationary which is convenient to model with autoregressive model (AR). For this, we have used many techniques for to determine the order which model better the data in this work. Therefore, we can characterize normal and abnormal activity which correspond to epileptic discharge for the patient. Our contribution in this work is the automatic detection of epilepsy seizure with the PSD novel approach by a better resolution in the frequency domain as the examination of EEG signals is often done with visual inspection of the rhythm (delta, theta, alpha, beta, gamma) by neurologists practitioners. The accuracy of the detection is estimated to 70% with the sensitivity of 80.55% compared with the interpretation of neurologist.