Professional Documents
Culture Documents
(6-12) A Novel Approach For EEG
(6-12) A Novel Approach For EEG
Volume 5 Issue 1
Abstract
Unprovoked seizure is the symptoms of Epilepsy disorder. An electroencephalogram (EEG) is
a test that perceives electrical activity in your mind using nearly nothing, level metal circles
(anodes) added to your scalp. The EEG signals are used for diagnosis of the patient whether
it is seizure or non-seizure which causes epilepsy. The essential objective of this paper is to
build a classifier that can accurately recognize whether a subject is seizure patient or non-
seizure patient. The proposed system classifies the EEG signals into two classes using
different supervised learning algorithm. Features are extracted using discrete wavelet
transform (DWT) and Genetic algorithm approach. Features dimensions have been reduced
using Principal component analysis (PCA) before feature extraction. Different classification
algorithms like support vector machine, logistic regression and decision tree classifier.
Random forest based classification with regularization gives best results which is best from
the existing system.
Delta: has a frequency of < 3 Hz. It will in rest. It might happen centrally with
general be the most noteworthy in force subcortical sores and all in all
and the wave with biggest timeframe. appropriation with diffuse sores, metabolic
Usually as the predominant musicality in encephalopathy hydrocephalus or
babies up to 1 year and in stages 3 and 4 of profound midline sores. It is typically most
that legitimate treatment and prescription variation present in the data set. This is
can be offered to those epileptic patients achieved by transforming to uncorrelated
experiencing epilepsy and diminish the features called as principal components,
opportunity of physical harm or and which are ordered so that the first few
mishaps[18]. The customary method for retain most of the variation present in all
identifying seizure in EEG signals is the of the original variables.First represent
manual skimming through 10s or even the independent features in terms of
100s of long stretches of recorded EEG matrix. Find the mean of the matrix for
motions alongside the requirement for a each features results in a vector of
specialist for the undertaking. This is features mean. Compute the covariance
extremely a strenuous and tedious matrix of the whole dataset using the
employment and furthermore inclined to formula.
human blunders. Subsequently auto
location of seizure in EEG signals is = ,
essential. Numerous such strategies are
accessible utilizing diverse methodologies
for a similar issue. In this task we made a Where, X & Y are the two features.
framework which dissects the EEG
signals channel astute, and extricates Computer eigen values and eigen vectors
highlights from them utilizing Discrete and sort the eigenvectors by decreasing
Wavelet Transform (DWT) trailed by eigen values and choose k eigenvectors
nourishing those highlights to a neural with the largest eigen values to form a d ×
network for discovery of seizure. The k dimensionalmatrix W. So W is the new
accessible EEG signals have been dimension of the features.
characterized into 3 classes-ordinary,
interictal and ictal. Different classifiers Feature Extraction
Feature extraction is the process of
have been tried against similar highlights
selecting best features from the available
extricated from the EEG signs to pass features. Discrete Wavelet Transform
judgment on their relative execution. (DWT) is used along with Genetic
Algorithm (GA).
PROPOSED WORK
Data Preprocessing Discrete Wavelet Transform (DWT):
Data is preprocessed using DWT is an implementation of the wavelet
standardization and normalization to get transform using a discrete set of the
the proper accuracy after applying the wavelet scales and translations obeying
algorithms. For the experimental setup the some defined rules. The discrete signal
data label is divided into two classes that , is a set of N
are classes with 5,4,3,2 are as non-seizure samples taken from a continuous signal
activity and 1 as seizure activity. + m ∆t), (m = 0,1, …, N-1) for
some initial time and sampling period
Dimension Reduction using Principle
. The basic functions
Component Analysis (PCA)
There are more than 170 features in the and
EEG dataset used for the experiment. For are also
avoiding over fitting of the classification
model reduction of size of the feature is vectors containing N elements.
necessary. PCA is used to reduce the Correspondingly the wavelet expansion
becomes discrete wavelet transform
dimensionality of a data set consisting of
(DWT). The discrete function is
a large number of interrelated and represented as a weighted sum in the
dependent variables while retaining the
database is then standardizes and feed to from the EEG signals particularly help in
the feature extraction algorithm as per the increasing the classification accuracy for
above approach. The accuracy of the detection of seizure. Finally the EEG
classification model is the calculated and records obtained from the database were
compared. classified into 2 classes: (i) Seizure, (ii)
Non Seizure. Classifier used for
Results classifying this data Random Forest
The results obtained are encouraging. showed promising results in terms of
Without even using a recurrent neural accuracy, sensitivity and specificity when
network, the Random Forest algorithm is verified using 75% of the available EEG
able to correctly classify the EEG signals data as training set and the remaining 25%
which detect the seizure and non-seizure as the test set. Different classification
activity. However, the best results were algorithms were compared like logistic
obtained when the network was trained on regression, k nearest neighbor, SVM and
samples from the same recording session. random forest. Random forest will give the
While this may be practical for basic brain maximum accuracy when trained.
research, it would be less practical for use
in BCI technology.This further REFERENCES
recommends utilizing EEG will probably 1. Biomedical signal acquisition.
require an iterative methodology of https://www.medicine.mcgill.ca.
preparing on a huge populace and after Accessed: 2017-04-30.
that adjusting on a particular person. It is 2. The bonneeg database.
consequently suggested that future http://epileptologiebonn.de/cms/frontc
research be done on the conceivable ontent.php?idcat=193lang=3.
utilization of Transfer Learning methods to Accessed: 2017−04−30.
the order of EEG signals. 3. Recognition of speech artifacts in eeg
records.
Different classification algorithms were http://docplayer.com.br/18572674-
compared like logistic regression, k Reconhecimento-de-
nearest neighbor, SVM and random forest. artefactosrelativos-a-fala-em-registos-
Random forest will give the maximum de-eeg.html. Accessed: 2017-06-20.
accuracy when trained. 4. The-mystery-of-gamma-waves.
http://q4lt.com/the-mystery-of-
The performance is evaluated by confusion gammawaves. Accessed: 2017-06-20.
matrix and then finding accuracy. 5. Lehnertz K. Rieke C. Mormann F.
David P. Andrzejak, R. &C. Elger.
# Accuracy = TP + TN / TP + TN + FN + Indications of nonlinear deterministic
FP and finite dimensional structures in
accuracy = (cm[0][0] + cm[1][1]) / time series of brain electrical activity:
(cm[0][0] + cm [0][1] + cm [1][0] + dependence on recording region and
cm[1][1]) brain state. Phys. Rev. E. v64. 061907.
# Accuracy 6. Bradley Lee Barnhart. The Hilbert-
accuracy * 100 Huang Transform: theory, applications,
development. PhD thesis, University of
CONCLUSION Iowa. 2011.
EEG signal records were decomposed 7. Dr Colin Tidy Dr Tim Kenny. Epilepsy
using multi-level DWT and further with focal seizures.
subjected to envelope analysis of the http://www.stillingtondoctors.co.uk.
coefficients obtained from DWT. This Accessed: 2017-04-30.
preprocessing before feature extraction