Seizure Detection Based on Multilayer Brain Network Features

Automated seizure detection can reduce the cost of long-term electroencephalography (EEG) monitoring and assist in the diagnosis of epilepsy. However, existing feature extraction methods primarily focus on the excessive discharge of brain neurons during seizures and fail to capture the altered synchronization in multichannel EEG signals. A frequency-based multilayer brain network was developed to model the complex interactions between different oscillation frequencies and brain regions. Topological features were extracted from each scale of the multilayer network, in combination with single-channel features, to enable patient-independent seizure detection. This approach achieved a sensitivity of 0.7301, a specificity of 0.9392, and an AVC of 0.9337 on the TUSZ dataset. The multilayer network-based method for capturing inter-channel synchronization variations from a phase perspective improves the sensitivity of short-duration seizure detection.