Neonatal seizure detection algorithms: The effect of channel count

Abstract The number of electrodes used to acquire neonatal EEG signals varies between institutions. Therefore, tools for automatic analysis, such as seizure detection algorithms, need be able handle different electrode montages in order find widespread use. aim this study was analyse the effect montage on detector performance. A full 18-channel compared reduced 3- and 8-channel using a convolutional neural network detection. Sensitivity decreased by 10 - 18 % while specificity mostly unaffected. Electrode artefacts associated with biological rhythms caused incorrect classification nonseizure activity some cases, but these were filtered out 3-channel montage. Other types had little effect. Reduced result reduction classifier accuracy, performance may still acceptable. Recording limited accuracy.