Biomedical Engineering Reference
In-Depth Information
There is evidence that when sleep becomes deeper the sources that drive EEG
activity move from the posterior regions of the head (prevalent during awake state
with eyes closed) to the centro-frontal regions [Kaminski et al., 1997].
The sleep pattern changes very much during childhood and adolescence. For new-
born babies REM takes most of the sleep time, and in young children only REM
and non-REM stages can be distinguished. Diminution of deep, slow wave sleep and
increase in wakefulness continues through the entire life span after 1 year of age. In
old age the contribution of stages 3 and 4 decreases markedly and first REM stage
appears later in the night.
A hypnogram describes sleep macrostructure and relies on division of the time
axis into fixed time epochs (20 or 30 s) —this naturally implies some limitations. In
the alternative approaches treating sleep EEG as a continuous process, its microstruc-
ture is described in terms of: evolution of spindles and SWA activity or occurrence
of transient arousals. They are defined as an abrupt shift in EEG frequency, which
may include theta, alpha, and/or frequencies greater than 16 Hz, but not spindles. A
certain number of spontaneous arousals seems to be an intrinsic component of physi-
ological sleep, but their frequent occurrence may be connected with respiratory sleep
disorders, nocturnal myoclonus, and other clinical conditions. The method providing
the continuous description of sleep, called the cyclic alternating pattern (CAP), was
proposed by [Rosa et al., 1999].
Sleep EEG analysis finds application in diagnosis of several disorders: insomnias,
somnambulism, narcolepsy, epilepsy (some epileptic conditions appear mainly dur-
ing sleep), depression, dementia, drug withdrawal. In sleep scoring, usually central
EEG derivations C3 or C4 (referenced to electrodes A1, A2 placed at mastoids), and
additionally two EOG channels and an EMG channel are considered. Sometimes
occipital electrodes are also taken into account.
The evaluation of sleep pattern connected with construction of hypnogram, or find-
ing arousals, is a very tedious and time consuming job involving analysis of a large
volume of polysomnographic recordings. Therefore as early as the 1970s attempts
were made to design a system for automatic sleep scoring. The early works deal-
ing with this problem included hybrid systems [Smith and Karacan, 1971] and pat-
tern recognition techniques [Martin et al., 1972]. Later, expert systems [Ray et al.,
1986, Kubat et al., 1994] and artificial neural networks (ANN) found applications
in sleep analysis [Roberts and Tarassenko, 1992]. ANNs with input coming from an
AR model were proposed for sleep staging by [Pardey et al., 1996]. The problem
was also attacked by model based approaches [Kemp et al., 1985] and independent
component analysis [McKeown et al., 1998]. Wavelet analysis was used for the de-
tection of transient structures in EEG [Jobert, 1992] and for the identification of
microarousals [Carli et al., 1999]. The review of the studies concerning digital sleep
analysis may be found in [Penzel et al., 2007].
There are several problems in automatic sleep analysis; one of them is poor con-
sent between electroencephalographers concerning the classification of stages [Ku-
bicki et al., 1989, Noman et al., 2000]. Another obstacle is the difficulty in the
identification of the transient structures such as sleep spindles, K-complexes, and
vertex waves. The detection of these waveforms is best performed by the methods
 
Search WWH ::




Custom Search