Dynamical graph neural network with attention mechanism for epilepsy detection using single channel EEG

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Epilepsy is a chronic neurological disorder requiring accurate seizure detection for effective patient management.
• Electroencephalogram (EEG) signals are crucial for monitoring brain activity and identifying epileptic seizures.
• Current methods for EEG-based seizure detection can be improved for greater accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate a novel classification algorithm for identifying epileptic seizures from single-channel EEG signals.
• To leverage dynamical graph neural networks with an attention mechanism for enhanced feature learning in EEG analysis.
• To improve the accuracy and reliability of automated seizure detection systems.

Main Methods:

• Empirical Mode Decomposition (EMD) was used to construct graphs from EEG signals.
• An optimal adjacency matrix was obtained through model optimization.
• A multilayer dynamic graph neural network with an attention mechanism was employed for feature extraction.
• An MLP-pooling structure was utilized for fusing graph features.

Main Results:

• The proposed algorithm was evaluated on 12 classification tasks using the University of Bonn epileptic EEG database.
• Achieved an average accuracy of 99.83% across 12 classification tasks.
• Demonstrated high performance with average specificity of 99.91%, sensitivity of 99.78%, precision of 99.87%, and F1-score of 99.47% using 25 runs of ten-fold cross-validation.

Conclusions:

• The developed dynamical graph neural network with an attention mechanism is highly effective for classifying epileptic EEG signals.
• This algorithm shows significant promise for accurate and reliable automated seizure detection.
• The findings suggest a potential for improved interventions and quality of life for epilepsy patients.