ABSTRACT
This chapter describes artificial neural networks in the context of classification of the Electroencephalogram (EEG) where a feature set can belong only to a limited number of predefined classes. It discusses in more general terms what may be incorporated into a complete detection or prediction system to account for differences observed between and within EEG traces. The long history of EEG seizure detection and prediction has shown that naive methods cannot be successful and creativity in the expert system is required. Decisions can be made for each EEG channel separately, at different points in time, with different degrees of certainty, etc. EEG databases are unlikely to have examples of this rare phenomenon, and adding these seizures to the database may not be possible because typical EEG monitoring lasts days rather than months. Patient specific algorithms in EEG are suitable only when sufficient data are available.
