Seizure Forecasting Using Long-Term Electroencephalography and Electrocardiogram Data.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Epilepsy Research

Background:

• Electroencephalography (EEG) is used for seizure forecasting, but with limited success.
• Electrocardiogram (ECG) is being explored as a complementary tool for seizure prediction.
• Critical slowing down, indicated by increased variance and autocorrelation, may precede seizures in neural and cardiovascular systems.

Purpose of the Study:

• To investigate the efficacy of critical slowing down and circadian rhythm features from EEG and ECG data for seizure forecasting.
• To compare the performance of different forecasting models (critical slowing, circadian, combined) in patients with epilepsy.
• To determine if seizure forecasting models are patient-specific.

Main Methods:

• Analysis of long-term continuous EEG and ECG data from 16 epilepsy patients (average recording length 161.9 hours).
• Calculation of variance and autocorrelation from EEG and ECG signals to identify critical slowing features.
• Development and comparison of forecasting models using critical slowing, circadian, and combined features via receiver-operating characteristic (ROC) curve analysis.

Main Results:

• The best seizure forecasting model was found to be patient-specific, with an average Area Under the Curve (AUC) of 0.68 across all patients.
• Circadian forecasting models performed best in 50% of patients.
• Critical slowing models were optimal in 19% of patients, and combined models in 31%.

Conclusions:

• Seizure forecasting using physiological data (EEG and ECG) shows potential, with patient-specific approaches yielding the best results.
• Both critical slowing and circadian rhythm features contribute to seizure prediction, with varying importance across individuals.
• Further development of these forecasting methods could significantly improve the quality of life for individuals with epilepsy.