Noise-Aware Epileptic Seizure Prediction Network via Self-Attention Feature Alignment.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Informatics

Background:

• Deep neural networks are widely used for epileptic seizure prediction from EEG data.
• Existing methods often overlook contextual consistency and noise in multi-layer EEG features.

Purpose of the Study:

• To propose a novel deep learning network, NSFA-Net, for improved epileptic seizure prediction.
• To address limitations in handling feature context and noise in EEG-based prediction.

Main Methods:

• Developed NSFA-Net with a self-attention backbone for multi-layer feature extraction.
• Implemented a time-frequency feature alignment module to maintain contextual consistency.
• Introduced a noise-aware regularizer to reduce the impact of inevitable EEG noise during training.

Main Results:

• Achieved average sensitivities of 98.68% on CHB-MIT and 93.57% on Kaggle datasets.
• Reported average false prediction rates of 0.038/h and 0.060/h on the respective datasets.
• Demonstrated superior performance compared to existing state-of-the-art methods.

Conclusions:

• NSFA-Net effectively extracts and aligns multi-layer EEG features while accounting for noise.
• The proposed method significantly enhances the accuracy and reliability of epileptic seizure prediction.
• This approach offers a promising advancement for epilepsy diagnosis and patient care.