An Epileptic Seizure Prediction Method Based on CBAM-3D CNN-LSTM Model

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Epilepsy is a common neurological disorder characterized by sudden, recurrent seizures.
• Timely seizure prediction is crucial for reducing patient injury and improving health outcomes.
• Existing deep learning models often overlook spatial features in electroencephalogram (EEG) signals.

Purpose of the Study:

• To develop and evaluate a deep learning model for accurate epilepsy seizure prediction.
• To leverage both temporal and spatial characteristics of EEG signals for improved prediction accuracy.
• To enhance the extraction of critical interictal and pre-ictal features from EEG data.

Main Methods:

• Preprocessing EEG signals using Short-Time Fourier Transform (STFT).
• Utilizing a 3D Convolutional Neural Network (CNN) for feature extraction.
• Employing a Bidirectional Long Short-Term Memory (Bi-LSTM) network for classification.
• Integrating a Convolutional Block Attention Module (CBAM) to focus on relevant spatial and channel information.

Main Results:

• Achieved an accuracy of 97.95% in seizure prediction.
• Demonstrated a sensitivity of 98.40% for seizure detection.
• Reported a low false alarm rate of 0.017 h^-1.
• Validated on the public CHB-MIT scalp EEG dataset with 11 patients.

Conclusions:

• The proposed CBAM-3D CNN-LSTM model effectively predicts epilepsy seizures by integrating temporal and spatial EEG features.
• The attention mechanism enhances the model's ability to extract key discriminative features.
• This approach holds significant potential for improving patient safety and management of epilepsy.