Information Technology Reference
In-Depth Information
that of seizure-free EEG signals and has been used to detect epileptic seizures in
EEG signals (Altunay et al. 2010 ). The fractional linear prediction (FLP) method
has been employed to model seizure and seizure-free EEG signals and prediction
error energy along with signal energy with support vector machine (SVM) classi
er
has been used to classify the epileptic seizure EEG signals from the seizure-free
EEG signals (Joshi et al. 2014 ). Epileptic seizures have been detected in EEG
signals using principal component analysis in combination with enhanced cosine
radial basis function neural network (Ghosh-Dastidar et al. 2008 ). Arti
cial neural
network (ANN) based methodology has been developed to detect the epileptic
seizure using time-domain as well as the frequency-domain features in Srinivasan
et al. ( 2005 ). Spectral parameters based on the Fourier transformation of EEG
signals have been utilized to detect epileptic seizures in EEG signals (Polat and
G
2007 ).
The EEG signals exhibit non-stationary nature (Boashash et al. 2003 ). In litera-
ture, many time-frequency domain based methods have been proposed to detect
epileptic seizure EEG signals, these methods include time-frequency distribution
(Tzallas et al. 2007 , 2009 ), wavelet transform (Ghosh-Dastidar et al. 2007 ;
Uthayakumar and Easwaramoorthy 2013 ; Ocak 2009 ; Subasi 2007 ; Subasi and
Gursoy 2010 ; Adeli et al. 2007 ; Acharya et al. 2012 ), multi-wavelet transform (Guo
et al. 2010 ), and empirical mode decomposition (EMD) (Pachori 2008 ; Oweis and
Abdulhay 2011 ; Pachori and Patidar 2014 ; Li et al. 2013 ; Bajaj and Pachori 2012 ).
The Fourier-Bessel (FB) series expansion of intrinsic mode functions (IMFs)
extracted from EMD, has been used to compute mean frequency of IMFs and these
mean frequencies have been used as features to discriminate ictal and seizure-free
EEG signals (Pachori 2008 ). In Oweis and Abdulhay ( 2011 ), weighted mean fre-
quency of IMFs has been proposed to detect epileptic seizures from EEG signals.
Ellipse area of second-order difference plot (SODP) of different IMFs with 95 %
con
ü
ne
ş
dence limit has been proposed as a feature to classify epileptic seizure and
seizure-free EEG signals (Pachori and Patidar 2014 ). The coef
cient of variation and
fl
fluctuation index computed from IMFs of EEG signals have been proposed to rec-
ognize patterns of ictal EEG signals (Li et al. 2013 ). The amplitude modulation
(AM) and frequency modulation (FM) bandwidths computed from the IMFs toge-
ther with least square support vector machine (LS-SVM) classi
er have been used
for classi
cation of seizure and nonseizure EEG signals (Bajaj and Pachori 2012 ).
Various non-linear parameters have been proposed as features for classi
cation of
epileptic seizure EEG signals. The Lyapunov exponent parameter with probabilistic
neural network (PNN) in
ler et al. ( 2005 ), correlation integral
in Casdagli et al. ( 1997 ), fractal dimension parameters in Easwaramoorthy and
Uthayakumar ( 2011 ) and Accardo et al. ( 1997 ), multistage nonlinear pre-processing
Ü
beyli ( 2010 ) and G
ü
filter combined with a diagnostic neural network in Nigam and Graupe ( 2004 ),
entropy based measures with adaptive neuro-fuzzy inference system in Kannathal
et al. ( 2005 ), and approximate entropy (ApEn) with ANN in Srinivasan et al. ( 2007 )
have been proposed for discrimination of epileptic seizure EEG signals.
Search WWH ::




Custom Search