Information Technology Reference
In-Depth Information
20.1 Introduction
Although automatic monitoring for most organ systems, including heart, lungs,
blood, etc., is common place in general and intensive care units, brain function
monitoring relies almost entirely upon bedside clinical observation by medical and
nursing staff. As a result, a large number of nonconvulsive seizures, with only sub-
tle or nonspecific behavioral changes, go undiagnosed every day. Recent clinical
studies have demonstrated the clinical utility of continuous EEG monitoring in such
inpatient settings as emergency department (ED), intensive care unit (ICU), and
epilepsy monitoring unit (EMU) [16, 17, 18, 5]. EEG-video monitoring is a standard
diagnostic procedure in the EMU for pre-surgical evaluation. Continuous EEG is
also well-established tool for detecting nonconvulsive seizures, cerebral ischemia,
cerebral hypoxia, and other reversible brain disturbances in the ICU and the ED.
Nevertheless, the utility of EEG monitoring currently depends on the availability of
expert technical and medical professionals, and the task of interpreting EEG is labor
intensive. Such experts are only available in tertiary care centers.
Automatic EEG monitoring has several potential practical applications, including
diagnosis and monitoring of patients with epilepsy and other neurological diseases,
monitoring of patients under general anesthesia, etc. An epileptic attack is usually
characterized by dramatic changes in electrical recordings of the brain activity by
multichannel EEG, whereas in the interictal state the EEG recording may appear
completely normal or may exhibit only brief rare abnormalities. Generally, the in-
terictal EEG often provides sufficient information to diagnose epilepsy and may
even contain evidence about the possible type of epilepsy [1].
The role of long-term continuous EEG recordings in clinical practice cannot
be underestimated. Although the main clinical application of continuous EEG in-
volves differential diagnosis between non-epileptic and epileptic seizures [1], there
are other useful applications (e.g., localization of site of onset of epileptic seizures,
detection of seizures with subtle clinical manifestations, finding the frequency of
inconspicuous seizures which may otherwise be overlooked) Development, testing,
and implementation of efficient methods for automatic EEG monitoring can be ex-
tremely useful in application to monitoring brain functions in clinical settings such
as ICU.
In the literature, various combinations of data mining techniques and data pre-
processing methods have been applied to EEG monitoring. For instance, an ap-
proach for extracting information from the video signal in video/EEG monitoring
is presented in [25]. That approach utilizes image compression method to develop
a domain change detection algorithm for automatic tracking of patient's move-
ments. Some popular preprocessing techniques applied to EEG involve such fea-
ture extraction methods as fast Fourier transform (FFT) [4, 19], wavelet transform
(WT) [20, 32], computation of fractal dimensions (FD) [22], calculating different
amplitude and frequency features (e.g., the average dynamic range, and frequency
vector, respectively) [6], symbolic representation of spike events [23], deviation ra-
tio topography [19], and others. The data mining methods include fuzzy classifi-
cation [23], regression of transformed data [4], segmentation (ictal, interictal) [22],
