A general sample-weighted framework for epileptic seizure prediction.

Area of Science:

• Neurology
• Machine Learning
• Biomedical Engineering

Background:

• Effective epileptic seizure prediction aids patients in taking timely preventive measures.
• Current machine learning approaches often treat all electroencephalogram (EEG) training samples equally, overlooking varying predictive values.
• Discrepancies in the predictive impact of EEG samples necessitate advanced modeling techniques.

Purpose of the Study:

• To propose a general sample-weighted framework for patient-specific epileptic seizure prediction.
• To enhance the accuracy and reliability of seizure prediction models by optimizing the contribution of individual training samples.
• To address the limitations of existing methods that do not account for differential sample importance.

Main Methods:

• A novel sample-weighted framework is introduced for epileptic seizure prediction.
• A fitness function is defined to map training sample weights to validation set performance.
• Genetic algorithms are employed to optimize sample weights, followed by model training with optimized weights.

Main Results:

• The proposed framework significantly improves performance across traditional and deep learning methods.
• A Transformer-based model achieved 94.6% average sensitivity, 0.06/h false prediction rate, and 0.939 AUC on the CHB-MIT database.
• Results demonstrate superior performance compared to state-of-the-art methods in predicting epileptic seizures.

Conclusions:

• The study offers novel insights into epileptic seizure prediction by accounting for EEG sample discrepancies.
• The developed sample-weighted framework is versatile and applicable to various classification methods.
• This approach significantly enhances the performance of seizure prediction models, offering a promising tool for clinical application.