EEG Signal Epilepsy Detection With a Weighted Neighbor Graph Representation and Two-Stream Graph-Based Framework.

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Epilepsy is a common neurological disorder requiring accurate seizure detection.
• Electroencephalography (EEG) signals are crucial for clinical diagnosis.
• Current deep learning models for EEG analysis lack interpretability and efficiency.

Purpose of the Study:

• To address the limitations of existing graph representations and GNN models for single-channel EEG epilepsy detection.
• To develop a more efficient and informative graph representation for EEG signals.
• To propose a novel framework for enhanced epileptic seizure detection using multi-domain EEG information.

Main Methods:

• Introduction of a Weighted Neighbour Graph (WNG) representation for EEG signals, optimizing for time and space efficiency.
• Development of a two-stream graph-based framework to simultaneously learn from WNG in both time and frequency domains.
• Application of graph neural network (GNN) models for feature extraction and classification of epileptic seizures from EEG data.

Main Results:

• The proposed WNG representation demonstrates significant improvements in efficiency and informativeness compared to existing methods.
• The two-stream GNN framework effectively integrates information from time and frequency domains, enhancing seizure detection accuracy.
• Extensive experiments validate the effectiveness and efficiency of the proposed approach in epileptic seizure detection.

Conclusions:

• The Weighted Neighbour Graph (WNG) and the two-stream GNN framework offer a promising advancement in automated epileptic seizure detection from single-channel EEG.
• This methodology enhances the interpretability and efficiency of deep learning models in neurological disorder analysis.
• The findings pave the way for more reliable and accessible diagnostic tools for epilepsy management.