Information Technology Reference
In-Depth Information
or specific recipients when
tonic-clonic seizures
are detected on Epilepsy patients. However, this
product cannot detect non-convulsion seizure such
partial seizures because the detection is based
purely on the physical movement of the patients.
Similar to
Wireless Motion Detection Monitor
and
Emfit tonic-clonic seizure monitor
, this device is
mainly used for in-house when the patients sleep.
Medpage MP5
detects convulsive seizures
happening when the patient sleep by monitoring
the patient's body movements and aural sound.
Similar to Emfit tonic-clonic and Epilepsy bed
sensor, Medpage MP5 detects the abnormal move-
ments of the patients and when such movements
last for more than a preset period of time, the alarm
will be triggered. In addition, the device also can
detect aural sounds such as choking, grunting, or
screams which are used to determine if a seizure
does happen (Medpage, 2009).
The limitation of these devices is the detection
method which majorly relies on a patient's physi-
cal movements. It would be more valuable if the
detection method is based on EGG signals which
can be retained and used for medical analysis. In
2008, University of Chicago Pediatric Epilepsy
Center started a project to develop a device to
detect and predict epileptic seizures. The model
of the device is similar to that of other commercial
devices for epileptic seizure detection. However,
the sensor module is “a compact EEG machine that
monitors brain activity” (Perez & Tressel, 2008).
Therefore, this project is supposed to bring back
great values for Epilepsy patients, the families,
and Epilepsy management researches.
plied on digitalized EGG data. These researches
can be arranged into several groups based on the
detection methods. The most common groups
involve neural network, similarity, synchroniza-
tion, and statistical analysis methods.
Neural Network Method
Subasi, Kiymik, Alkan, and Koklukaya (2005)
examined the use of autoregressive (AR) model
by using utmost likelihood estimation (MLE). The
interpretation and performance of this method
were also investigated to find out classifiable
features from human EEG via Artificial Neural
Networks (ANNs). The approach was based on
the earlier observations in which the EEG spec-
trum contains some characteristic waveforms that
majorly belong to four frequency bands— delta
(< 4 Hz), theta (4-8 Hz), alpha (8-14 Hz), and
beta (14-30 Hz). Fast Fourier transform (FFT)
and AR with MLE parameters were used as inputs
to ANNs. ANNs are then evaluated for accuracy,
specificity, and sensitivity on classification of
each patient into epileptic seizure or non-epileptic
seizure group. A classification rate of 92.3% was
achieved by ANNs with a single hidden unit as a
classifier. The classification of AR with MLE was
above 92% and an average of 91% classification
was achieved by utilizing FFT as preprocessing
in the ANN.
Bao, Lie, and Zhang (2008) proposed an au-
tomated system which uses inter-ictal EEG data
to categorize the Epilepsy patients and detects
seizure activities for patient monitoring and further
diagnosis by doctors. The probabilistic neural
network (PNN) was supplied with four classes
of features extracted from the EEG data. Leave-
one-out cross-validation (LOO-CV) applied on a
commonly used epileptic-normal data set returned
99.3% accuracy of detection which differentiates
normal people's EEG from Epilepsy patients' inter-
ictal EEG. In addition, the system can be used for
seizure detection and seizure focus localization at
96.7% and 76.5% accuracy detection.
Review of Some Significant
Researches on Epileptic Seizure
Detection and Prediction
Automated detection and prediction of epileptic
seizures have been the topics receiving consider-
able amount of research interest for over a decade.
Automated detection methods are generally based
on computerized seizure detection algorithms ap-
Search WWH ::
Custom Search