Classification of Normal and Epileptic Seizure EEG Signals Based on Empirical Mode Decomposition - Complex System Modelling and Control Through Intelligent Soft Computations - page 376

method has been implemented using Matlab. The Matlab codes for EMD method are
available at http://perso.enslyon.fr/patrick. andrin/emd.html . In this study, the

proposed methodology has been validated with one online freely available EEG
dataset (Andrzejak et al. 2001 ). As discussed in Sect. 2.1 , this dataset includes EEG
signals which have been recorded from both healthy and epileptic subjects. It con-
tains

fl

ve subsets denoted as Z, O, N, F, and S. Data subsets Z and S have been used
to evaluate the performance of the proposed method for classi cation of normal and

epileptic seizure EEG signals. Data subset Z consists of normal EEG recordings
taken from 5 healthy volunteers and subset S consists of the EEG recordings of
seizure activities. Each of these subsets have 100 single-channel EEG signals of
duration 23.6 s.
The decomposition of EEG signals using EMD method results into IMFs that are
in decreasing order of frequency, in which

first component is associated with
highest frequency. As the IMFs can help to compute the area of analytic signal
representation of the IMFs in the complex plane and ellipse area parameter obtained
from SODP of IMFs, therefore the EMD has been used to decompose the EEG
signals into a set of IMFs. These above mentioned two area parameters have been
used to create the feature space for classi cation between normal and epileptic

seizure EEG signals.
Recently in Pachori and Bajaj ( 2011 ), the ability of the analytic signal repre-
sentation of IMFs to discriminate EEG signals which contains normal and epileptic
seizure EEG signals has been explored. It comes out of this study that the analytic
signal representation of IMFs provides a set of proper rotations which facilitates
accurate identi cation of the centers and estimation of surface areas in the complex

plane. It has been shown that the area parameter of the analytic IMFs has signi cant

potential to differentiate between epileptic seizure and normal EEG signals. The
experimental results of the above mentioned method reveals that the epileptic
seizure EEG signals had evidently greater surface area in comparison to that of the
normal EEG signals. The increased surface area in the complex plane for IMFs of
the epileptic seizure EEG signals could be attributed to large amplitude of EEG
signals for seizure subjects. It should be noted that the use of EMD enabled the
extraction of individual centers of rotation for each IMF. Furthermore, as discussed
in this study, it is evident from experimental analysis, that window size of 2,000
samples has provided better results, therefore the same window size has been used
to compute the area parameters in this work. As the analytic signal representation
has circular geometry, therefore modi

ed CTM has been measured to compute the
area of the analytic signal representation of the IMFs of EEG signals in the complex
plane. The radius of the circular region which covers the 95 % of the CTM has been
used to determine the area parameter for first four IMFs of EEG signals. In Pachori

and Patidar ( 2014 ), the ef cacy of the ellipse area parameters of SODP of IMFs for

classi cation of seizure-free and ictal EEG signals has been examined. This study

has employed the 95 % con dence ellipse area as a feature for discrimination of

ictal EEG signals from the seizure-free EEG signals, and the classi cation per-

formance of the ellipse area parameter have been evaluated for various window
sizes (500, 1,000, 2,000, 4,000 samples) of seizure-free and ictal EEG signals.