An early attempt
21
to use EEG pattern recognition to predict seizures was successful at
detecting seizures, but impractical because of a high false-positive rate. Synchronization
among different EEG channels tends to decrease just before a seizure
22
and bursts of high-frequency activity appear.
23
Early seizure prediction models
24
utilized a new field called nonlinear dynamics or “chaos
theory” to study the interictal-ictal transition. A standard EEG plot of voltage versus time
does not display information about whether one time segment of the EEG is non-randomly
related to another. Using nonlinear dynamics, Iasemides
25
asserted that “… the next seizure
can be predicted 91.3% of the time, about 91 min prior to its onset, with the issue of 1 false
warning every 8.27 h.” While this level of accuracy has not always been replicated , several
other researchers have successfully employed nonlinear dynamic methods for seizure
prediction.
26272829
One study
30
of 21 patients with implanted mesial temporal electrodes
recording 88 seizures over 582 continuously recorded hours concluded that predictive ability
was better than random, but sensitivity was only 21–42%.
Several EEG characteristics other than nonlinear dynamics have been used to attempt
seizure prediction, including accumulated energy
31
, wavelets
32
and baseline voltage
crossings
33