Detection of epileptiform spikes in the EEG using a patient-independent neural network

Abstract

An offline neural network that successfully detects spikes when trained on multiple patients selected from a database of electroencephalogram (EEG) records with spikes marked by experienced electroencephalographers has been developed. This spike detector uses a simple threshold detector to identify potential spikes that appear on four-channel bipolar chains within the montage, and then passes waveform parameters to a three-layer neural network for second-level detection. Results obtained for the neural network with output thresholds arbitrarily set of 0.5 have yielded sensitivities averaging 74% and selectivities averaging 54%. While the selectivities for these trials were only fair, it is noted that substantial improvements could be achieved by raising the output thresholds.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>