Focal and non-focal EEG classification through deep learning

Abstract

Epilepsyis a neurological condition characterized by the repeated occurrence of seizures, which can severely affect patients' quality-of-life. For this purpose, the accurate classification of Electroencephalography (EEG) signals, which is a non-invasive exam, is of great interest in AI research in health. In this project, we intend to advance the classification of EEG signals for Epilepsyanalysis. We will research a new classification framework based on deep learning that runs on graphs: a data representation model widely used to its ability to represent functional and topological relationships in data and has been employed in the study of Brain Networks. This recently emerged field is denominated Graph Neural Networks and it has been accomplishing state-of-the-art results in many classification tasks due to its ability to represent relational complex systems (such as Brain Networks) natively for deep learning tasks. We will investigate the development of representative Brain Networks and combine them with structured EEG data to classify whether a given EEG signal is Focal or Non-Focal. The Focal signals (FEEG) are related to the onset zone and they are acquired from regions where the first ictal EEG changes are observed. Non-focal EEG signals (NFEEG) are obtained from brain regions that do not contribute to seizure onset. The combination of relational data (Brain Network) and structured data (EEG recordings) may have inherent advantages in classifying the signals and possibly identifying their starting location (onset zone) for surgery. (AU)