Information Technology Reference
In-Depth Information
event detection and classification [6], patient classification (diagnosis), neural net-
works [19, 20, 32]. In many studies, EEG monitoring is aimed at detection of epilep-
tic seizures [32, 22, 6], or at detection of spike events [20, 23].
This chapter introduces the approach for automatic EEG monitoring that is based
on nonlinear dynamic theory, statistics, and optimization. Based on this approach,
a seizure monitoring and alert system (SMAS) is designed as an online system for
generating warnings for impeding seizures by the analysis of patient's EEG record-
ings. The SMAS also incorporates a seizure susceptibility index (SSI) that is based
on the seizure warning algorithm. The SMAS is developed with a purpose of pro-
viding medical staff information as to the likelihood of ensuing seizure and alerting
the staff when seizure occurs.
The remainder of the chapter is organized as follows. Section 20.2 discusses
seizure prediction and warning. Section 20.3 presents the methods involved analysis
of EEG data. In particular, the methods include application of chaos theory to mea-
sure dynamical transitions in the epileptic brain via Lyapunov exponents, statistical
approach to quantifying similarity between pairs of measurements, and application
of quadratic optimization methods to detect the critical channels in multichannel
EEG. In Section 20.4, we propose the SMAS based on an algorithm for generat-
ing automatic warnings about impending seizure from EEG, which incorporates the
above methods. Finally, the conclusion follows.
20.2 Preictal Transition and Seizure Prediction
The studies investigating the possibilities for prediction of epileptic seizures date
back to the late 1970s. During the 1970s and the 1980s, the linear approaches, in-
cluding linear autoregression, spectral analysis, and pattern recognition techniques,
were mostly applied to analysis of epileptic seizures. Some studies conducted at
that time reported changes in EEG characteristic of epileptic seizures, which could
only be detected a few minutes before the seizure onset. Later, beginning in the
late 1980s, various nonlinear approaches based on Lyapunov exponents, correlation
dimension, and different entropy measures were introduced to study the dynami-
cal changes in the brain before, during, and after epileptic seizures. Introduction of
the nonlinear methods resulted in the findings that showed characteristic changes in
EEG minutes to hours before seizure onset. These results were reported in a num-
ber of papers in the 1990s, and interpreted as an evidence of existence of interictal
state. Beginning in the early 2000s, the multivariate approaches become especially
current in analysis of epileptic seizures. Various studies show particular importance
of spatiotemporal relations in multichannel EEG data with respect to the transitions
in epileptic brain. Another developing research area includes assessment of perfor-
mance of different algorithms of seizure prediction. The two different techniques
are proposed, namely a bootstrap based approach proposed by Andrzejak et al. and
a seizure prediction characteristic method introduced by Winterhalder et al.
