Convolutional Neural Networks for Seizure Detection: A Study on Training Strategies

This study investigates how various training strategies originally proposed in the context of image processing, can be used to improve the performance of a convolutional neural network designed for classification of seizures from EEG recordings.Random cropping of seizure segments, dropout, mixup and ensembling improved the performance of the baseline classifier, alone and in combination. The best results were obtained by a combination of random cropping, mixup and ensembling, improving the AUC from 0.957 to 0.981 and F1-score from 71.0% to 77.9%.This study shows the importance of optimizing the training of neural networks for seizure detection.