CNN classification of variance-based selected topo-maps of EEG

Abstract

Epileptic activity in the EEG record can manifest in different ways over time series. A classifier that would alert physicians to the possibility of different types of epileptic activity would be an effective tool. We created image data from EEG records, which we subsequently classified using the SqueezeNet network, which has a promising potential in the field of image classification based on the results so far. On patients whose data the network did not come into contact with during training and validation, we subsequently assessed the accuracy of the classification. The accuracy for each condition was around 80%.