An epilepsy detection method based on multi-dimensional feature extraction and dual-branch hypergraph convolutional...

Area of Science:

• Neuroscience
• Computational Biology
• Medical Signal Processing

Background:

• Epilepsy is a neurological disorder characterized by abnormal neural discharges, leading to diverse seizure manifestations and complex pathogenesis.
• Current epileptic seizure detection methods struggle with insufficient feature extraction and susceptibility to data noise.
• Understanding brain network changes is crucial for accurate epilepsy detection and revealing disease mechanisms.

Purpose of the Study:

• To propose a high-precision and robust model for epileptic seizure detection using hypergraph convolution.
• To address limitations of existing methods by enhancing feature extraction and improving model robustness.
• To explore higher-order characteristics and intrinsic commonalities within epilepsy electroencephalogram (EEG) signals.

Main Methods:

• A novel dual-branch parallel approach combining Conv-LSTM for spatio-temporal features and Power Spectral Density (PSD) for frequency domain features.
• Application of hypergraph convolution to capture higher-order relationships and intrinsic commonalities in extracted EEG features.
• Ensemble learning was employed on the dual-branch hypergraph convolution for final epilepsy detection.

Main Results:

• The model achieved high performance on the TUH dataset with leave-one-out cross-validation: 96.9% accuracy, 97.3% F1 score, 98.2% precision, and 96.7% recall.
• Generalization performance was validated on the CHB-MIT scalp EEG dataset, yielding 94.4% accuracy, 95.1% F1 score, 95.8% precision, and 93.9% recall.
• The proposed model demonstrated superior performance compared to existing methods in the literature.

Conclusions:

• The developed hypergraph convolution model offers a significant advancement in high-precision and robust epileptic seizure detection.
• The method effectively extracts rich, multi-dimensional features and reveals intrinsic commonalities in EEG signals.
• The model's strong performance and generalization capabilities provide a valuable reference for clinical applications in epilepsy management.