An epileptic seizure prediction model based on a time-wise attention simulation module and a pretrained ResNet

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Epilepsy affects 1% of the global population, necessitating accurate seizure prediction.
• Unpredictable seizures significantly impact patients' quality of life.
• Current prediction methods require improvement for clinical utility.

Purpose of the Study:

• To develop and validate a novel seizure prediction model, TASM_ResNet.
• To enhance the accuracy and reliability of epilepsy seizure forecasting.
• To explore the efficacy of deep learning for neurological disorder prediction.

Main Methods:

• Proposed TASM_ResNet model integrating a time-wise attention simulation module and a pre-trained ResNet.
• Converted intracranial electroencephalogram (EEG) data into image-like representations for temporal feature extraction.
• Employed an improved focal loss (FL) to address data imbalance and optimized FL parameters.
• Utilized a pre-trained ResNet to minimize training data requirements.

Main Results:

• TASM_ResNet achieved a superior average Area Under the Curve (AUC) of 0.877 compared to a state-of-the-art CNN model.
• Demonstrated successful migration of EEG signals to an image network via the simulation module.
• Validated the effectiveness of the improved focal loss for imbalanced epilepsy datasets.

Conclusions:

• The TASM_ResNet model shows significant promise for accurate and reliable epilepsy seizure prediction.
• The time-wise attention simulation module effectively extracts temporal features from EEG data for deep learning models.
• Migrating EEG data to pre-trained image networks is a viable approach for seizure prediction research.