Information Technology Reference
In-Depth Information
Finally, the adaptive detection method was
developed based on the similarity index which
helps to predict epileptic seizure. The result
showed that this dynamical similarity index is not
dependent on the selection of the radius values of
Gaussian function and the size of EEG recording.
Comparing with the dynamical similarity index
proposed Le, Baulac, and Varela (1999), the results
on twelve rats showed that this method provide
better prediction on epileptic seizure.
of a random predictor. This assessment method in
combination with a statistical test procedure, and
proposed extensions enable a patient statistically
verifiable, and clinically motivated selection of
optimal prediction parameters.
Statistical Analysis Method
Iasemidis, et al (2003) proposed an adaptive pro-
cedure for the analysis of continuous long-term
EEG recordings in the scenario that only the oc-
curring time of the very first seizure is known.
The procedure is based on the convergence and
divergence of short-term maximum Lyapunov
exponents in critical electrode sites which are
selected adaptively. The selection of the critical
groups of electrode sites was based on global
optimization techniques. The algorithm was tested
in continuous 0.76 to 5.84 days intracranial EEG
data sets obtained from five patients with refrac-
tory temporal lobe Epilepsy. The results were 82%
of seizures were predicted with a false positive
rate of 0.16/h when a fixed parameter setting
was applied to all cases. Seizure prediction was
obtained on the average of 71.7 min before ictal
onset. A similar set of results was also obtained
when dividing the available EEG data set into
half training and testing parts. Improvement in
sensitivity (84% overall) and reduction in false
positive rate (0.12/h overall) were achieved by
optimizing these parameters for each patient.
Most importantly, the results have proved that this
adaptive procedure can be applied for implantable
devices for clinical purposes.
In order to establish a statistical framework
for modeling seizure generation and a method for
validating algorithm performance, Wong, Gardner,
Frieger, and Litt (2007) proposed a new random
framework based on a three-state hidden Markov
model. This model represents interictal, preictal,
and seizure states with the feature that periods of
increased seizure probability can transition back
to the interictal state. The model involves clipped
EEG segments and incorporates intuitive notions
about statistical validation. Equations for type I and
Synchronization Method
Abnormal synchronization of neurons is an im-
portant factor generating epileptic seizures. Inves-
tigating the relationships between the dynamics
of different neural populations can possibly help
forecasting epileptic seizures. Winterhalder et al
(2006) applied a phase and a lag synchronization
measure and evaluated changes in synchronization
with respect to seizure prediction on a selected
subset of multi-contact intracranial EEG record-
ings. Their results showed that synchronization
changes in EEG dynamic prior to a seizure are
factors that can be used for seizure prediction.
In the experience, averaged sensitivity values of
60% were achieved with a false prediction rate
of 0.15 false predictions per hour, half an hour
seizure occurrence period, and a prediction hori-
zon of 10 min. A statistically significant predic-
tion performance was observed for at least one
synchronization measure and evaluation scheme
for about half of 21 patients.
Schelter, et al (2007) assessed the impacts of
long prediction horizons in relation to clinical
needs and on patients by analyzing long-term
continuous intracranial EEG data. The assess-
ment was conducted using synchronization
theory. The trade-off between intervention times
and occurrence periods of long-term continuous
intracranial EEG data obtained from four patients
in over several days was investigated. The frac-
tion of false alarms and false warning time were
also investigated. A statistical test was applied to
ensure the achieved performance is better than that
Search WWH ::
Custom Search