Non-Linear Boundaries for Differentiation Between Epileptic Transients and Background Activities in EEG

The epileptic transients and background activities of EEG from ten patients are found to group in two separate clusters in a two-dimensional sample space according to their half-wave amplitudes and durations. Assuming that the two clusters are from jointly Gaussian distributions with equal a priori probabilities, an optimum decision boundary is determined for each patient. An estimated decision boundary is also obtained for each patient based only on the information of the individual's background cluster. The results indicate that the performance of such an estimated boundary is good in comparison with the optimum decision boundary.