Seizure detection by brain-connectivity analysis using dynamic graph isomorphism network.

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Epilepsy is a neurological disorder characterized by abnormal brain electrical discharges.
• Electroencephalography (EEG) is crucial for measuring brain activity and detecting seizures.
• Manual EEG analysis is time-consuming, necessitating automated seizure detection, especially for long-term monitoring.

Purpose of the Study:

• To develop advanced algorithms for efficient inter-patient seizure detection using EEG signals.
• To address the limited generalizability of existing patient-specific seizure detection methods.
• To improve long-term ambulatory seizure monitoring capabilities.

Main Methods:

• Utilized dynamic brain networks to capture spatiotemporal connectivity dynamics among brain regions.
• Proposed a novel graph neural network, termed graph isomorphic network, for feature extraction.
• Evaluated the method using the CHB-MIT open dataset with a ten-fold cross-validation.

Main Results:

• The proposed graph isomorphic network achieved excellent performance in inter-patient seizure detection.
• Achieved high accuracy (96.2%), sensitivity (95.4%), and specificity (97.0%).
• Performance significantly surpassed existing methods reported in the literature.

Conclusions:

• The developed algorithm demonstrates high efficacy for inter-patient seizure detection.
• The findings offer valuable insights for advancing long-term ambulatory seizure monitoring.
• The novel approach shows promise for more generalizable and automated epilepsy diagnosis.