A Generative Adversarial Network for Data Augmentation of Ictal Waves from Multi-Electrode Brain Activity

High-density multi-electrode arrays (HD-MEAs) are sophisticated recording devices used in vitro to capture voltage traces of neural activity. HD-MEAs yield data with a high spatial and temporal resolution at the level of individual neurons. In this study, deep convolutional neural networks (CNNs) were trained on HD-MEA data to classify epileptic seizure waves. A major challenge for the prediction of epileptic seizures was the scarcity of data as ictal waves constitute rare events relative to background neuronal activity. To address this issue, we investigated a method of data augmentation using generative adversarial networks (GANs), a class of deep learning models that are trained to generate plausible synthetic images. We show that synthetic exemplars created by GANs can serve as training data for deep convolutional neural networks (CNNs) in order to improve classification accuracy. These results suggest a workflow for combining GANs and CNNs in the detection of ictal waves recorded via HD-MEAs.