Hilbert Vibration Decomposition-based epileptic seizure prediction with neural network

Area of Science:

• Neurology
• Biomedical Engineering
• Signal Processing

Background:

• Epilepsy is a prevalent neurological disorder causing debilitating seizures.
• Existing seizure prediction methods often require patient-specific data.
• Developing a generalized, non-patient specific system is crucial for widespread clinical application.

Purpose of the Study:

• To design and demonstrate a non-patient specific seizure prediction system.
• To utilize Hilbert Vibration Decomposition (HVD) for EEG signal analysis.
• To improve seizure management by providing timely predictions.

Main Methods:

• Surface electroencephalogram (EEG) recordings from 10 patients (CHB-MIT database) were analyzed.
• EEG signals were decomposed into 7 subcomponents using Hilbert Vibration Decomposition (HVD) in sliding windows.
• Features from HVD subcomponents were fed into a Multi-Layer Perceptron (MLP) classifier for simultaneous, non-patient specific classification.

Main Results:

• Initial classification sensitivity averaged 19.89% across patients.
• An alarm algorithm improved sensitivity to an average of 89.8% within 120 minutes.
• A low average false alarm rate of 0.081/h was achieved with a 4-minute prediction horizon.

Conclusions:

• The developed non-patient specific seizure prediction system demonstrates significant potential.
• HVD combined with MLP classification offers an effective approach for generalized seizure prediction.
• The system's high sensitivity and low false alarm rate suggest clinical utility in epilepsy management.