Information Technology Reference
In-Depth Information
Aslan, Bozdemir, Şahin, Oğulata, and Erol
(2008) conducted a study to examine Epilepsy
patients and classify them into partial and primary
generalized Epilepsy. The study utilized Radial Ba-
sis Function Neural Network (RBFNN) and Mul-
tilayer Perceptron Neural Network (MLPNNs).
Four hundred eighteen Epilepsy patients were cor-
rectly classified by two expert neurologists before
being executed by neural networks which were
trained by the parameters from the EEG signals
and clinic properties of the patients. The results
from both of the neural networks suggested that
the predictions corresponding to the learning data
sets were compelling. More precisely, RBFNN
with total classification accuracy of 95.2% is better
than MLPNN with total classification accuracy
of 89.2%. Therefore, it is suggested that RBFNN
model can be used as a decision support tool in
clinical studies to validate the epilepsy group
classification after the development of the model.
Sivasankari and Thanushkodi (2009) proposed
an efficient approach for automatic detection of
the epileptic seizures in EEG signals. The input
EEG signals are analyzed with the aid of fast in-
dependent component analysis (FastICA) which is
a statistical signal processing technique to obtain
the components related to the detection of epileptic
seizures. The back propagation neural network
was then trained with the obtained components
for effective detection of epileptic seizures. The
experimental results portray that the proposed ap-
proach efficiently detects the presence of epileptic
seizure in EEG signals and also showed reasonable
accuracy in seizure detection.
minutes in advances in most of the cases. More-
over, the method can be computerized and used to
track spatio temporal changes in a patient's brain
few minutes before the seizure. The algorithm
utilizes similarity index is quite independent of
the choice of radius values and EEG signal length.
It also reduces the false positive rates and can
produce better result than dynamical similarity
index during inter-ictal state.
Phase-space dissimilarity measures (PSDM)
was ever proposed to be a solution to make pre-
diction of impending epileptic seizure from scalp
EEG. However, this method may create channel
inconsistency in which multiple data sets from a
patient resulting in conflicting prediction of epilep-
tic seizure in the same channel. This problem was
addressed by Hively and Protopopescu in 2003.
According to Hively and Protopopescu (2003),
“channel-consistent total trues are a much more
stringent measure of forewarning performance”
which can be addressed by: 1) establishing a
quantitative measures of “channel consistency
in both true positives (TPs) and true negatives
(TNs) for multiple datasets from each of several
patients; 2) measuring forewarning performance
in terms of channel consistent total trues; 3) de-
veloping a methodology to make the most of this
performance measure; and 4) showing that these
improvements raise the channel-consistent total
trues considerably.”
Li & Ouyang (2006) proposed an improved
dynamical similarity measure to predict epilep-
tic seizures in electroencephalographic (EEG).
Firstly, a phase space of preprocessed EEG re-
cording by using positive zero crossing method
was reconstructed by the method for determining
the minimum embedding dimension of a scalar
time series provided by Cao (1997) and mutual
information. In the next step, Heaveyside function
was replaced by Gaussian function in the cor-
relation integral at calculating a similarity index
which helps eliminates the crips boundary of the
Heavyside function.
Similarity Method
Le, Baulac, and Varela (1999) presented a method
to analyze long-term nonstationarity in EEG by
measuring the dynamical similarity between dif-
ferent parts of the time series. The method was
evaluated on a group of patient with temporal lobe
Epilepsy detected within the cranium. The results
showed that their extraction of dynamical proper-
ties can help to predict coming seizures several
Search WWH ::
Custom Search