Epileptic Seizure Detection in EEG via Fusion of Multi-View Attention-Gated U-Net Deep Neural Networks

Abstract

Electroencephalography (EEG) is an essential tool in clinical practice for the diagnosis and monitoring of people with epilepsy. Manual annotation of epileptic seizures is a time consuming process performed by expert neurologists. Hence, a procedure which automatically detects seizures would be hugely beneficial for a fast and cost-effective diagnosis. Recent progress in machine learning techniques, especially deep learning methods, coupled with the availability of large public EEG seizure databases provide new opportunities towards the design of automatic EEG-based seizure detection algorithms. We propose an epileptic seizure detection pipeline based on the fusion of multiple attention-gated U-nets, each operating on a different view of the EEG data. These different views correspond to distinct signal processing techniques applied on the raw EEG. The proposed model uses a long short term memory (LSTM) network for fusion of the individual attention-gated U-net outputs to detect seizures in EEG. The model outperforms the state-of-the-art models on the TUH EEG seizure dataset and was awarded the first place in the Neureka™ 2020 Epilepsy Challenge.