Optimization of epilepsy detection method based on dynamic EEG channel screening.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Epilepsy detection faces challenges due to interference and limited channel capability in focal epilepsy.
• Effective feature extraction is crucial for accurate epilepsy diagnosis and across-subject generalization.

Purpose of the Study:

• To propose a novel epilepsy detection method using dynamic electroencephalogram (EEG) channel screening.
• To enhance epilepsy feature extraction and enable robust across-subject epilepsy detection.
• To provide an effective approach and theoretical support for clinical epilepsy detection.

Main Methods:

• Dynamic EEG channel screening using Refine Composite Multiscale Dispersion Entropy (RCMDE) to measure signal complexity.
• Identification of common epilepsy features (3-15 Hz) across subjects from screened channels.
• Construction of a Residual Convolutional Long Short-Term Memory (ResCon-LSTM) neural network for detection.

Main Results:

• Single-subject epilepsy detection accuracy reached 98.523%, a 5.298% improvement over non-channel screening.
• Across-subject epilepsy detection achieved an average accuracy of 96.596%.
• The method demonstrated effectiveness in dynamically screening optimal channels for good detection performance.

Conclusions:

• Dynamic EEG channel screening significantly enhances epilepsy detection capability and robustness.
• The proposed method successfully identifies common epilepsy features for across-subject detection.
• This approach offers a promising solution for clinical epilepsy detection across diverse subjects.