An Efficient 1D CNN Architecture for Multi-Channel EEG Seizure Detection.

Area of Science:

• Neurology
• Artificial Intelligence
• Signal Processing

Background:

• Epilepsy affects over 1% of the global population, necessitating accurate seizure prediction for improved patient outcomes.
• Current seizure detection methods often fail to capture the complex dynamics within multi-channel electroencephalography (EEG) signals.
• Developing computationally efficient seizure detection models is crucial for real-time clinical applications.

Purpose of the Study:

• To design an efficient 1D Convolutional Neural Network (CNN) architecture for seizure detection using multi-channel EEG data.
• To optimize the model for computational efficiency while preserving critical temporal features of EEG signals.
• To evaluate the proposed model's performance against state-of-the-art methods on a publicly available seizure database.

Main Methods:

• A novel 1D CNN architecture was developed, incorporating depthwise-separable convolutions.
• The architecture was specifically tailored for processing multi-channel EEG signals.
• The model was trained and validated using the CHBMIT seizure database.

Main Results:

• The proposed model achieved high performance metrics: 95% accuracy, 91% F1-score, 94% precision, 89% recall, and 87% specificity.
• Depthwise-separable convolutions effectively reduced model complexity while retaining temporal information.
• Multi-channel analysis demonstrated the model's robustness and effectiveness.

Conclusions:

• The developed 1D CNN offers a computationally efficient and highly accurate solution for EEG-based seizure detection.
• The model's performance surpasses existing state-of-the-art approaches.
• This work contributes a promising tool for enhancing early and accurate seizure prediction in epilepsy management.