Time CNN and Graph Convolution Network for Epileptic Spike Detection in Meg Data

Pauline Mouches, Thibaut Dejean, Julien Jung, Romain Bouet, Carole Lartizien, Romain Quentin

Abstract: Magnetoencephalography (MEG) recordings of patients with epilepsy exhibit spikes, a typical biomarker of the pathology. Detecting those spikes allows accurate localization of brain regions triggering seizures. Spike detection is often performed manually. However, it is a burdensome and error prone task due to the complexity of MEG data. To address this problem, we propose a lightweight 1D temporal convolutional neural network (Time CNN) to classify short time frames of MEG recording as containing a spike or not. Furthermore, we investigate the benefits of coupling the Time CNN with a graph convolutional network (Time CNN-GCN), to account for MEG sensors spatial relationships. Our models produce clinically relevant results and outperform state-of-the-art methods based on deep CNNs. On a balanced dataset, our Time CNN-GCN achieves the best performances (spike class f1-score: 76.7%) while our Time CNN alone performs best when tested on a realistic, highly imbalanced dataset (spike class f1-score: 25.5%).