Epileptic Seizure Prediction Based on Permutation Entropy.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Epilepsy is a chronic brain disorder characterized by recurrent seizures due to abnormal neuronal discharge.
• Predicting seizures could significantly enhance patient safety and quality of life by allowing timely interventions.
• The utility of permutation entropy (PE) for human epilepsy prediction from intracranial electroencephalogram (iEEG) signals is not well-established.

Purpose of the Study:

• To investigate the novel application of permutation entropy (PE) for tracking human brain activity dynamics.
• To assess the efficacy of PE in predicting epileptic seizures from iEEG recordings.
• To evaluate a machine learning approach for seizure prediction using PE features.

Main Methods:

• Intracranial electroencephalogram (iEEG) data were collected from 19 epilepsy patients.
• Permutation entropy (PE) was extracted from preprocessed iEEG signals within a sliding time window.
• A support vector machine (SVM) classifier was used to discriminate brain states, followed by a two-step post-processing prediction method.

Main Results:

• The developed method achieved an average sensitivity (SS) of 94% and a false prediction rate (FPR) of 0.111 h⁻¹.
• Exceptional results, including 100% SS and 0 h⁻¹ FPR, were observed in some patients.
• An average prediction horizon of 61.93 minutes was obtained, providing substantial advance warning before seizures.

Conclusions:

• Permutation entropy (PE) is a viable feature for extracting information from iEEG signals for seizure prediction.
• The combination of PE feature extraction and SVM classification demonstrates significant potential for clinical application in human epilepsy prediction.
• This approach offers a promising tool for improving seizure forecasting and patient management.