Leveraging Channel Coherence in Long-Term iEEG Data for Seizure Prediction.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Epilepsy affects millions globally, with unpredictable seizures significantly impacting quality of life.
• Current seizure prediction methods lack the accuracy needed for effective clinical application.
• There is a critical need for advanced techniques to reliably forecast epileptic seizures.

Purpose of the Study:

• To introduce and evaluate the Coherence-based Seizure Prediction (CoSP) method for enhanced seizure forecasting.
• To integrate electroencephalography (EEG) coherence analysis with deep learning for improved prediction efficacy.
• To assess CoSP's performance in a pseudo-prospective setting using real-world patient data.

Main Methods:

• EEG data segmented into 10-second intervals to extract channel pairwise coherence.
• A four-layer convolutional neural network trained on coherence data to predict preictal state probability.
• Pseudo-prospective evaluation using long-term intracranial EEG (iEEG) data from ten epilepsy patients.

Main Results:

• CoSP demonstrated promising predictive performance across various preictal intervals (4-180 minutes).
• Achieved a median Seizure Sensitivity (SS) of 0.79 and a median false alarm rate of 0.15 per hour.
• Significantly outperformed chance (p = 0.001) and baseline methods (p <0.05).

Conclusions:

• The Coherence-based Seizure Prediction (CoSP) method shows significant potential for accurate and reliable seizure forecasting.
• CoSP's integration of EEG coherence and deep learning offers a novel approach to epilepsy management.
• This method could lead to improved seizure warning systems and better patient outcomes.