Spatio-temporal filtering of the EEG via neural network based multireference adaptive noise cancelling

A system is proposed which enhances transient non-stationarities and, in particular, epileptiform discharges in the EEG. It is based around the technique of multireference adaptive noise cancelling (MRANC) which attenuates the background EEG on a primary channel by using spatial and temporal information from adjacent multichannel EEG recording. This implemented by means of a 3-layer perceptron artificial neural network trained by a backpropagation algorithm. System performance was measured as the percentage increase in signal-to-noise ratio (SNR) of predetermined epileptiform discharges in recorded EEG segments. The results obtained show that, due to the nonlinear nature of the artificial neural network, the improvement in SNR is significant when compared to the performance of MRANC utilising a linear model. MRANC is proposed as the first stage of a neural network based multi-stage system to detect epileptiform discharges in the interictal EEG for the diagnosis of epilepsy