Beruflich Dokumente
Kultur Dokumente
EEG signals were recorded with an 128-channel amplifier perform automatic classification known as Normal,
system, 12 bit A/D resolution and digitized with a sampling Background (Pre-ictal) and Epilepsy (ictal).
rate of 173.61 Hz. The methodology of the work is explained
by Fig. 1.
FEATURE
ANOVA TEST EXTRACTION
CLASSIFIERS
E X f 1 X f 2 X f 1 f 2 0.5 1
B norm ( f 1 , f 2 )
(2)
P f 1 P ( f 2 ) P ( f 1 f 2 )
f1
Where P(f) is the power spectra.
Fig. 3. Real Signals Computation at Non-redundant Region
A. Features Extracted
In this study, seven features were being extracted to F. Mean bispectrum magnitude:
differentiate EEG signals. The bispectral entropies are shown
1 (7)
in (3), (4) and (5). mAmp
B f1 , f 2
L
B. Normalized bispectral entropy (Ent1):
Where B f1 , f 2 is the bispectrum of the signal
B f1 , f 2 (3)
pn
B f1 , f 2 wcx
iB ( i , j )
,
B (i, j )
C. Normalized bispectral squared entropy (Ent2):
wcy
iB ( i , j ) (8)
B (i, j )
Ent 2 q n log q n
Where
n hence,
B f1 , f 2 (4) iB ( i , j )
2
qn wc 1
, wc
iB ( i , j )
and
B f1 , f 2
2
2
B (i, j )
B (i, j )
The first order spectral moment of amplitudes of diagonal V. RESULTS AND DISCUSSIONS
elements of the bispectrum H3 is equated as:
H 3 log B f k , f k (11) The result of the extracted features Ent1, Ent2, EntPh
signify the complication or abnormality of EEG signal that
occurs during epilepsy is reduced. The statistical analysis of
IV. TYPES OF CLASSIFIERS these measurements is extremely statistically significant with
p<0.0001.
A. Probabilistic Neural network (PNN)
Due to the random and non-linear characteristic of brain
signals, statistical analysis is being employed. The features are
PNN is a feed forward network consists of three layers, being tested using ANOVA method to choose the statistical
namely input layers, hidden layers and output layers. It is significant features to be fed to various classifiers. Eight
derived from Bayes Theorem. Donald Specht replaced features which have p-value less than 0.0001 are chosen.
sigmoid activation function with a derived statistical activation The classifiers namely, Probabilistic Neural Networks
function [21]. The main advantage of using PNN is the (PNN), K-Nearest Neighbor (KNN), Support vector machine
computational speed where it can be used on real-time and Decision Tree (DT) are being tested against four
classification comparing to typical back propagation neural performance parameters. SVM classifier scores better
networks. Moreover, it has the flexibility of choosing between performance and accuracy when compared to other classifiers
linear decision probability density function (PDF) and moreover Fuzzy and SVM showed comparable performance.
complex nonlinear P. It is proven the entropies estimators, phase entropy and
weighted centers can be used to identify epilepsy.
B. K-Nearest Neighbour (KNN)
TABLE I. CLASSIFIERS PERFORMANCE OUTCOME
The K-nearest-neighbour (KNN) algorithm measures the Classifiers PNN KNN DT SVM
distance between a query scenario and a set of scenarios in the
data set and it is a memory based learning technique used to TN 28 28 28 60
classify samples based on the closest training samples exist in FN 1 1 2 1
the n-dimensional space. It does not perform comparison with TP 59 59 58 59
trained models for classification. A user defined constant, k is FP 2 2 2 1
used to determine the test samples belong to which class based
Accuracy (%) 96 96 95 98
on majority votes. Typically, k is a small positive integer
value [22]. PPV (%) 96 96 96 98
Sensitivity (%) 98 98 96 98
REFERENCES
[16] Quek, C., Pasquier, M., Lim, B, “OP-TRAFFIC: A Novel Fuzzy Neural
Approach to Road Traffic Analysis and Prediction”, IEEE Transactions
on Intelligent Transportation Systems, 7(2), 133-146, 2006.
[17] Schater, S.C, “The Brainstorms companion: epilepsy in our view”, New
York: Raven Press; 994. activity. Phys Rev Lett, 80. 5019-22, 1998.
[18] Singh, Anuradha, “100 questions & answers about epilepsy”, Sudbury,
Mass: Jones and Bartlett Publishers, 2006.
[19] Specht, Donald. F, “Probabilistic Neural Networks. Pergamon Press pie
on Neural Networks”, Vol. 3. 109 118, 1990.
[20] Subha, D. Puthankattil, Joseph, Paul. K., U, Rajendra. Acharya, Lim, C.
M, “EEG Signal Analysis: A Survey”, J Med System; 34: 195-212.
doi:10.1007/s.10916-008-9231.z, 2010.
[21] Swami, A., Mendel, C. M., Nikias C.L, “Higher-Order Spectral Analysis
(HOSA) Toolbox”, version 2.0.3, 2000.
[22] Tomohiro, Takagi. Michio, Sugeno, “Fuzzy identification of systems and
its applications to modeling and control” IEEE Transactions on systems,
man, and cybernetics, Vol. SMC-15. 116-132, 1985.
[23] Types of Seizures
http://www.epilepsyfoundation.org/about/types/types/index.cfm
[24] Vagus Nerve Anatomy
http://en.wikipedia.org/wiki/Vagus_nerve.