Seizure forecasting with ultra long-term EEG signals.

Area of Science:

• Neuroscience
• Epilepsy Research
• Biomedical Engineering

Background:

• Seizure occurrence unpredictability significantly impacts epilepsy patients' quality of life.
• Seizures often synchronize with underlying interictal epileptiform activity cycles.
• Existing methods struggle with reliable, long-term seizure prediction.

Purpose of the Study:

• To evaluate a novel forecasting scheme for predicting seizures days in advance.
• To assess the efficacy of a bandpass filter for capturing mid-term dynamics in seizure activity.
• To determine the patient-specific and cross-patient forecasting capabilities of the scheme.

Main Methods:

• Retrospective analysis of long-term electroencephalogram (EEG) recordings from three different systems: NeuroPace RNS, NeuroVista intracranial, and UNEEG subcutaneous.
• Application of a bandpass filter to identify universal mid-term dynamics.
• Testing of both patient-specific and cross-patient forecasting models.

Main Results:

• The forecasting scheme achieved better-than-chance accuracy in 83% of patients for daily forecasts and 89% for hourly forecasts.
• Meaningful forecasts up to 30 days were possible for 22% of patients with hourly forecast frequency.
• Cross-patient forecasting performance was only marginally reduced and correlated with patient-specific results.
• Similar forecasting performance was observed for NeuroVista and UNEEG data.

Conclusions:

• The study demonstrates the feasibility and potential clinical utility of a novel seizure forecasting scheme.
• Cross-patient forecasting success highlights the universal relevance of mid-term dynamics in epilepsy.
• The scheme shows promise for inter-subject applicability on ultra-long-term EEG recordings, offering hope for improved epilepsy management.