Automated system for detection of epileptiform patterns in EEG by using a modified RBFN classifier

In this paper, we present a two-stage system based on a modified radial basis function network (RBFN) classifier for an automated detection of epileptiform pattern (EP) in an electroencephalographic signal. In the first stage, a discrete perceptron fed by six features are used to classify the peaks into two subgroups: (i) definite non-EPs and (ii) definite EPs and EP-like non-EPs. In the second stage, the peaks falling into the second group are aimed to be separated from each other by a modified RBFN designed by a perturbation method that would function as a post-classifier. If there exist redundant data components in training data set, they can be discarded by analyzing the total disturbance of the RBFN output corresponding to the perturbed inputs. Thus, input dimension size is reduced and network becomes smaller. The classification performance of the system is comparatively evaluated for three different feature sets such as raw EEG data, discrete Fourier transform coefficients, and discrete wavelet transform coefficients.