Post-processing deep neural network for performance improvement of interictal epileptiform discharges detection

In this study we aimed to further improve interictal epileptiform discharges (IED) detection performance of a deep neural network based algorithm with a second-level post-processing deep learning network. Seventeen interictal ambulatory EEGs were used, 15 with focal and 2 with generalized epilepsy in patients of age range 4-80 years (median 19y, 25th-75th percentile 14-32y). The EEG data was split into 2s non-overlapping epochs. Epochs with an IED probability of at least 0.99, according to a previously developed VGG-C based convolutional neural network, were preselected for the second-level 2D convolutional neural network (CNN). The selected epochs were used as input for the second-level post-processing CNN in Python; these are uploaded as a dataset in this repository. Data was divided into an 80/20 training/validation, resulting in 3049 epochs for training/validation and 580 epochs for testing.