Automated Seizure Detection using Transformer Models on Multi-Channel EEGs

Epilepsy is a prevalent neurological disorder characterized by recurring seizures, affecting approximately 50 million individuals globally. Given the potential severity of the associated complications, early and accurate seizure detection is crucial. In clinical practice, scalp electroencephalograms (EEGs) are non-invasive tools widely used in seizure detection and localization, aiding in the classification of seizure types. However, manual EEG annotation is labor-intensive, costly, and suffers from low inter-rater agreement, necessitating automated approaches. To address this, we introduce a novel deep learning framework, combining a convolutional neural network (CNN) module for temporal and spatial feature extraction from multi-channel EEG data, and a transformer encoder module to capture long-term sequential information. We conduct extensive experiments on a public EEG seizure detection dataset, achieving an unweighted average F1 score of 0.731, precision of 0.724, and recall (sensitivity) of 0.744. We further replicate several EEG analysis pipelines from literature and demonstrate that our pipeline outperforms, current state-of-the-art approaches. This work provides a significant step forward in automated seizure detection. By enabling a more effective and efficient diagnostic tool, it has the potential to significantly impact clinical practice, optimizing patient care and outcomes in epilepsy treatment. Codes available on GitHub
¹
.