Detection of epileptiform activity in human EEG signals using bayesian neural networks

In this paper, we investigate the application of neural networks to the problem of detecting inter-ictal epileptiform activity in the electroencephalogram (EEG). The proposed detector consists of a segmentation, feature extraction and classification stage. For the feature extraction stage, coefficients of the Discrete Wavelet Transform (DWT), real and imaginary parts of the Fast Fourier Transform and raw EEG data were all found to be well-suited to EEG classification. Principal Component Analysis was used to reduce the dimensionality of the features. For the classification stage, Multi-Layer Perceptron neural networks were implemented according to Maximum Likelihood and Bayesian Learning formulations. The latter was found to make better use of training data and consequently produced better trained neural networks. Rejection thresholds of 0.9 were applied to the network output as a doubt level in order to ensure that only reliable classification decisions are made. A maximum classifier accuracy of 95,10% was achieved with 24,97% of patterns not being classified. Bayesian moderated outputs could not improve on these classification predictions significantly enough to warrant their added computational overhead.