A self-attention model for cross-subject seizure detection.

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Epilepsy is a neurological disorder causing recurrent seizures, traditionally detected via manual electroencephalography (EEG) analysis.
• Automatic seizure detection from high-dimensional, non-stationary EEG signals presents significant challenges.
• Deep learning (DL) techniques offer advanced solutions for automated seizure detection.

Purpose of the Study:

• To propose a novel deep learning model for automatic epilepsy seizure detection.
• To incorporate a self-attention mechanism (SAT) into a DL model for enhanced EEG signal analysis.
• To develop a model that directly processes raw EEG data, minimizing the need for signal processing expertise.

Main Methods:

• A deep learning model combining a one-dimensional convolutional neural network (CNN) for feature extraction and a long short-term memory (LSTM) module for temporal analysis.
• Integration of a self-attention (SAT) layer within the LSTM encoder to improve feature representation.
• Direct input of raw EEG signal data, eliminating the need for manual feature engineering.

Main Results:

• The proposed DL model achieved high F1-scores: 97.8% for binary and 92.7% for five-class seizure recognition on the UCI dataset.
• The model demonstrated strong performance on the CHB-MIT database with an F1-score of 97.9%, outperforming existing state-of-the-art DL methods.
• The approach showed robustness to inter-subject variability in EEG data.

Conclusions:

• The novel DL model with a self-attention mechanism effectively detects epileptic seizures from EEG signals.
• The model's ability to process raw data simplifies application and enhances performance.
• This method represents a significant advancement in automated seizure detection, offering high accuracy and robustness.