A Functional-Genetic Scheme for Seizure Forecasting in Canine Epilepsy

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Biology

Background:

• Epilepsy forecasting remains a challenge, necessitating advanced algorithms for timely intervention.
• Accurate localization of seizure onset zones is crucial for developing effective forecasting models.

Purpose of the Study:

• To develop an accurate seizure forecasting algorithm utilizing brain functional connectivity for optimal electrode selection.
• To enhance the precision of seizure prediction through adaptive connectivity methods and advanced feature selection.

Main Methods:

• Proposed Kmeans-directed transfer function for adaptive functional connectivity and seizure onset zone localization in intracranial EEG.
• Implemented a seizure-forecasting algorithm using selected electrodes from long-term canine epilepsy recordings.
• Employed a precision-recall genetic algorithm for feature selection coupled with a probabilistic support vector machine classifier.

Main Results:

• Identified focal epileptic activity generators and bilateral seizure sinks in canine epilepsy models.
• Achieved improved seizure forecasting performance with an average sensitivity of 84.82% and a time in warning of 0.1.
• Demonstrated enhanced prediction accuracy compared to previous seizure forecasting studies.

Conclusions:

• The developed algorithm demonstrates the feasibility of accurate seizure forecasting in canine epilepsy.
• Improved seizure forecasting capabilities pave the way for EEG-triggered closed-loop seizure intervention systems.
• This research offers a promising advancement for patients with refractory epilepsy.