Automatic seizure detection using three-dimensional CNN based on multi-channel EEG.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Automated seizure detection from clinical electroencephalogram (EEG) data aids epilepsy diagnosis and treatment.
• Current methods often rely on limited, manually designed features and struggle with multi-channel EEG data's temporal and spatial information.
• Deep learning offers a promising approach for automatic feature extraction and processing of complex EEG data.

Purpose of the Study:

• To develop an effective system for automatic seizure detection using multi-channel EEG signals.
• To leverage deep learning, specifically a 3D convolutional neural network (CNN), for enhanced seizure detection.
• To overcome the limitations of conventional machine learning algorithms in handling multi-channel EEG data.

Main Methods:

• Collected EEG data from 13 epilepsy patients, previously inspected by experts.
• Converted time-series EEG data into 2D images, then combined them into 3D images representing inter-electrode correlations.
• Developed and applied a 3D CNN with 3D kernels to classify EEG data into inter-ictal, pre-ictal, and ictal stages.

Main Results:

• Multi-channel EEG data analysis significantly increased specificity and sensitivity compared to single-channel analysis.
• The proposed 3D CNN model achieved over 90% accuracy, with 88.90% sensitivity and 93.78% specificity.
• The 3D CNN outperformed both 2D CNN and traditional signal processing methods in seizure detection.

Conclusions:

• This study represents the first application of 3D CNN for seizure detection from EEG.
• The 3D CNN approach enables simultaneous pattern learning from multi-channel EEG signals.
• Deep neural networks combined with 3D kernels offer an effective system for automated seizure detection.