An Optimized Artificial Neural Network Approach for Epileptiform Activity Recognition

M. Ayala and M. Adjouadi (USA)


Artificial neural networks, EEG, Spike detection, Programming tools.


This study introduces a new artificial neural network (ANN) system dedicated to the automatic recognition of epileptic foci in electroencephalograph (EEG) recordings. This ANN is based on a trend-adaptive polygon which optimizes the search process and more importantly reduces the size of the training set by an impressive 74% or better, yielding as a consequence a computationally attractive ANN. In recent years, several studies have been published on this important topic. All of them seem to have in common the idea of an EEG scrolling window used to extract the desired pattern in the training data set of the ANN. The dimension of the scrolling window determines in such an approach both the size of the training set, and more importantly, the number of neurons or nodes needed in the input layer of the ANN. The premise of this study is to overcome the computational complexity required of the scrolling window approach and resolve the problem effectively through the use of what we call a trend-adaptive polygon. The merit of such a practical premise is weighed in terms of minimizing the input layer of the ANN, and reducing the size of the training set data.

Important Links:

Go Back