EEG Based Seizure Detection Using Connectivity Analysis

Abstract

Epileptic seizure detection using electroencephalogram (EEG) signals plays a vital role in the diagnosis and monitoring of neurological disorders. This work proposes an EEG connectivity framework that detects epileptic seizures by modeling interactions between different brain regions rather than processing individual EEG channels separately. EEG recordings from the SIENA Scalp EEG Database are first preprocessed and segmented into seizure-related temporal states. Functional and effective connectivity measures, including Mutual Information (MI), Phase Lag Index (PLI), Granger Causality (GC), and Transfer Entropy (TE), are then extracted to capture the relationships between EEG channels. A comparative evaluation identifies PLI as the most suitable connectivity measure because of its robustness to noise and its effectiveness in representing phase synchronization. The resulting PLI connectivity matrices are classified using Support Vector Machine (SVM), Convolutional Neural Network (CNN), and Graph Neural Network (GNN) models. Among the evaluated models, the SVM classifier produced the highest classification accuracy of (98.4%), outperforming CNN (74.0%) and GNN (62.5%). These findings demonstrate that combining PLI-based connectivity analysis with conventional machine learning offers an accurate and computationally efficient approach for automated EEG-based seizure detection.