Epilepsy Detection and Monitoring - Quantitative EEG Analysis Methods and Clinical Applications - page 159

nel SVMs have been applied to EEG data after removing noise and other artifacts
from the raw signals in the various channels. In one report, the author was able to
detect 97% of the seizures using an online detection method that used a kernel
SVM. Of the seizures that were detected, the author reported that he was able to
predict 40% of the ictal events by an average of 48 seconds before the onset of the
seizure [44].

6.11 Phase Correlation

Methods of measuring phase synchrony include methods based on spectral coher-
ence. These methods incorporate both amplitude and phase information, detection
of maximal values after filtering. For weakly coupled nonlinear equations, phases
are locked, but the amplitudes vary chaotically and are mostly uncorrelated. To
characterize the strength of synchronization, Tass [45] proposed two indices, one
based on Shannon entropy and one based on conditional probability. This
approach aims to quantify the degree of deviation of the relative phase distribution
from a uniform phase distribution.
All of the techniques that have been described thus far approach the problem of
detecting and predicting seizures from a traditional time-series prediction perspec-
tive. In all such cases, the EEG signal is viewed like any other signal that has predic-
tive content embedded in it. The goal, therefore, is to transform the signal using
various mathematical techniques so as to draw out this predictive content. The fact
that an EEG signal is generated in a particular biological context, and is representa-
tive of a particular physical aspect of the system, does not play a significant role in
these techniques.

6.12 Seizure Detection and Prediction

Seizure anticipation (or warning) can be classified into two broad categories: (1)
early seizure detection in which the goal is to use EEG data to identify seizure onset,
which typically occurs a few seconds in advance of the observed behavioral changes
or during the period of early clinical manifestation of focal motor changes or loss of
patient awareness, and (2) seizure prediction in which the aim is to detect preictal
changes in the EEG signal that typically occur minutes to hours in advance of an
impending epileptic seizure.
In seizure detection, since the aim of these algorithms is to causally identify an
ictal state, the statistical robustness of early seizure detection algorithms is very high
[46, 47]. The practical utility of these schemes in the development of an online sei-
zure abatement strategy depends critically on the few seconds of time between the
detection of an EEG seizure and its actual manifestation in patients in terms of
behavioral changes. Recently Talathi et al. [48] conducted a review of a number of
nonparametric early seizure detection algorithms to determine the critical role of
the EEG acquisition methodology in improving the overall performance of these
algorithms in terms of their ability to detect seizure onset early enough to provide a
suitable time to react and intervene to abate seizures.