A Spatiotemporal Graph Attention Network Based on Synchronization for Epileptic Seizure Prediction

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Accurate early epileptic seizure prediction is crucial for timely patient treatment.
• Previous methods often neglect the combined temporal and spatial dimensions of electroencephalograms (EEGs).
• This limitation hinders the full evaluation of EEG signal properties.

Purpose of the Study:

• To propose an effective spatiotemporal graph attention network (STGAT) for enhanced epileptic seizure prediction.
• To address the limitations of single-dimension analysis in previous EEG studies.
• To fully evaluate the effective properties of EEGs by considering both spatial and temporal correlations.

Main Methods:

• Extracted spatial and functional connectivity from multichannel EEG signals using phase locking values (PLVs).
• Modeled multichannel EEG signals as graph signals.
• Employed a spatiotemporal graph attention network (STGAT) to learn temporal correlations and explore spatial topology.

Main Results:

• The STGAT model achieved high accuracy (98.74%), specificity (99.21%), and sensitivity (98.87%) on the CHB-MIT dataset.
• On a private dataset, all evaluation indices exceeded 98.8%, with an Area Under the Curve (AUC) of 99.96%.
• The proposed method demonstrates superior or comparable performance to existing state-of-the-art models.

Conclusions:

• The STGAT model effectively captures spatiotemporal correlations in EEG data for accurate seizure prediction.
• The end-to-end automatic prediction model shows potential for clinical decision support systems.
• This approach offers a significant advancement in the field of automated epileptic seizure detection.