Epileptic Seizure Prediction Using Attention Augmented Convolutional Network.

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Early seizure prediction is vital for managing epilepsy and preventing injuries.
• Distinguishing pre-ictal from inter-ictal electroencephalogram (EEG) signals is challenging due to subtle differences.

Purpose of the Study:

• To develop a novel epileptic seizure prediction method using a multi-head attention (MHA) augmented convolutional neural network (CNN).
• To overcome the limitations of traditional CNNs in capturing global information from EEG signals.

Main Methods:

• EEG data augmentation to balance pre-ictal and inter-ictal samples.
• Utilizing Stockwell transform (ST) for EEG time-frequency distribution.
• Employing an attention-augmented CNN for feature extraction and classification.
• Implementing post-processing to minimize false prediction rates (FPR).

Main Results:

• Achieved a segment-based sensitivity of 98.24% and an event-based sensitivity of 94.78%.
• Reported a low false prediction rate (FPR) of 0.05/h.
• Demonstrated the system's effectiveness on the CHB-MIT EEG database.

Conclusions:

• The proposed MHA-augmented CNN method shows significant potential for accurate epileptic seizure prediction.
• The findings suggest promising clinical applicability for improving epilepsy patient care and quality of life.