ABSTRACT
In a clinical setting, a vast number of physiological variables contribute to and
affect the EEG. As a result, the distinctions between the different EEG groups
are not very well defined. For the computer model to be clinically effective,
it is required to be robust with respect to EEG variations across subjects
and various mental states. The ultimate objective is a comprehensive tool for
epilepsy diagnosis as well as real-time monitoring of EEGs for seizure detection
and, eventually, seizure prediction. Toward this end, the following classifier
characteristics are desirable: fast training, high classification accuracy for all
three groups, and low sensitivity to training data and classifier parameters,
that is, robustness, to be discussed later in this chapter. Although a high
classification accuracy is reported using LMBPNN in Chapter 8, the class
of RBFNN classifiers is investigated further, in this chapter, to increase the
robustness of classification.
