736 icat a
TRANSCRIPT
Epileptic State Detection: Pre-ictal,
Ictal, Inter-ictal
Apdullah YAYIK* Yakup KUTLU
Esen YILDIRIM Serdar YILDIRIM
Mustafa Kemal University
*Presenter
Propose of Study
Epileptic State Detection The goal of this study is discrimination of three states
of an epileptic patient: before (pre-ictal), during (ictal)
and after seizure (inter-ictal) using artificial
intelligence techniques. For this purpose, EEG
database obtained from Children's Hospital Boston
(CHB) is used.
What is EEG Signal
Electroencephalography (EEG) is the recording of
electrical activity along the scalp.
EEG measures voltage fluctuations resulting from ionic
current flows within the neurons of the brain
What is Epilepsy
Epilepsy is explained by recurring instant seizures due to the
instantaneous development of synchronous firing in the cerebrals
cortex caused by lasting cerebral abnormality .
Database
Children's Hospital Boston
5 Male (3-22 ages), 17 Female (1.5-19 ages)
Second Order Difference Plot
Second-Order Difference
x axis =X(n +1) -X(n)
y axis = X(n+2)-X(n+1)
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1 4
32
3-3-3-1 6-5-1-0 1-1-1-0 6-5-3-0
46
5
3
0
Circle Region Parameters
Classification
Artificial Neural Network
3 hidden layers (16-9-3) as shown in Figure, sigmoidtransfer functions and back propagation algorithm isused.
Cross-Validation
o Cross validation determines how the results of a statistical analysis will
generalize to a new data set.
Classification Results
Classifier Class Precision RecallROC
Area
Kappa
Statistic
Mean
Absolute
Error
Root Mean
Squared
Error
Overall
Accuracy
(%)
Multi-Layer
Neural
Network
pre-seizure 99.00% 99.00% 0.99
0.98 0.01 0.07 98.70Ictal 97.00% 97.60% 0.99
Inter-Ictal 100.00% 100.00% 0.99
k-Nearest
Neighbor
pre-seizure 94.10% 94.10% 0.95
0.92 0.03 0.18 94.71Ictal 92.10% 92.10% 0.93
Inter-Ictal 98.00% 98.00% 0.99
Naive Bayes
pre-seizure 98.10% 100.00% 0.91
0.95 0.02 0.11 95.14Ictal 97.30% 94.10% 0.83
Inter-Ictal 96.10% 97.00% 0.96
Related Works
Study Year Database FeaturesClassification Problem
(Epileptic States)Classifier
Classification
Accuracy (%)
[20] 2009 Freiburg dataset [19] Wavelet Transform pre-ictal and inter-ictal Convolutional NN 71,00
[1] 2009 Individual Approximate Entropy normal and ictal Elman NN 93,33
[5] 2011National Taiwan
University HospitalApproximate Entropy
normal and inter-ictal and
ictal SVM 98.1
[6] 2012Andrzejak Dataset
[22]
Discrete Wavelet
Transformnormal and ictal ANN 98,87
[7] 2012Andrzejak Dataset
[22]Approximate Entropy normal and ictal ANN 93,43
[8] 2012Andrzejak Dataset
[22]
Fast Independent
Component Analysis normal and ictal ANN 99,50
[9] 2013Andrzejak Dataset
[22]
Discrete Wavelet
Transformnormal and ictal SVM 98,00
[21] 2014Andrzejak Dataset
[22]
Empirical Mode
Decomposition
Second-Order Difference
normal and ictal ANN 95,00
This
Study2014 CHB Dataset [10] Second Order Difference pre-ictal, ictal and inter-ictal ANN 98,70
Results
This study shows that the state of an epileptic
patent can be classified as before, during and
after seizure (pre-ictal, ictal and inter-ictal) using
SODP features and machine learning
algorithms.
Questions?
Apdullah YAYIK
Download this presentation and proceeding
using Mobile Phone QR Code Reader