Neural networks that learn how to detect epileptic spikes

L. DIAMBRA; A. CAPURRO; A. PLASTINO

PHYSICS LETTERS A

ELSEVIER SCIENCE BV

Año: 1998 vol. 241 p. 61 - 66

0375-9601

We advance a method based on a neural network environment to detect epileptic spikes in EEG. The concomitant training process of the pertinent feed-forward neural networks involves information theoretic tools. The resulting technique can be used to predict the one-step-forward behavior of the EEG time series. An optimal number of neurons is determined that allows for the maximum correlation between EEG signals and the neural network output. Our approach is illustrated with reference to EEG time series of the interictal activity from focal epileptic patients.