Classifier Combination Supported by the Sleep-Wake Cycle Improves EEG Seizure Prediction Performance

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Epilepsy affects millions, with nearly 30% experiencing drug-resistant seizures.
• Seizure prediction offers a promising avenue to enhance patient quality of life.

Purpose of the Study:

• To evaluate the impact of integrating sleep-wake cycle information into seizure prediction models.
• To compare different methods of incorporating vigilance state data for improved prediction accuracy.

Main Methods:

• Developed five patient-specific seizure prediction models utilizing sleep-wake data in various ways.
• Compared these models against a control method lacking sleep-wake information.
• Utilized data from 17 epilepsy patients (43 seizures, 482 hours) and developed a sleep-wake classifier from a separate dataset.

Main Results:

• The best performing model, a pool of weighted predictors, achieved above-chance prediction levels in 65% of patients.
• This significantly outperformed the control method, which succeeded in only 41% of patients.
• Individual patient results varied, suggesting personalized strategies may be necessary.

Conclusions:

• Integrating sleep-wake information demonstrably enhances seizure prediction accuracy.
• Further research with long-term, real-world data is needed for clinical acceptance.
• Automated sleep-wake detection feasibility supports integration into future seizure prediction devices.