Computational Intelligence Algorithms for Seizure Prediction
Authors
Bruno Miguel Direito Pereira Leitão
Ricardo Martins
Rui Costa
António Dourado
Francisco Sales
Marco Vieira
Ricardo Martins
Rui Costa
António Dourado
Francisco Sales
Marco Vieira
Abstract
Purpose:To develop computational intelligence algorithms for seizure prediction to embed in a transportable device to support refractory epileptic patients.
Methods:
Firstly a set of features is extracted from the EEG, measuring energy, time-frequency and nonlinear dynamic contents. These features are then used for classification of the brain state into four classes: inter-ictal, pre-ictal, ictal, pos-ictal. Two approaches from computational intelligence are applied: (i) artificial neural networks in the original 14 features space (several architectures are compared: feedforward, with and without memory, radial basis function, Elman), (ii) multidimensional scaling to reduce the 14th dimensional space to 3-dimensional space where classification may be done in an easier way.