Parameter estimation in a whole-brain network model of epilepsy: Comparison of parallel global optimization solvers

Area of Science:

• Computational Neuroscience
• Medical Informatics
• Epileptology

Background:

• Virtual Epileptic Patient (VEP) models integrate patient-specific anatomy and brain dynamics.
• VEPs simulate spatio-temporal seizure patterns for pre-surgical hypothesis testing.
• Accurate calibration is crucial for VEP model reliability in clinical decision-making.

Purpose of the Study:

• To evaluate a global optimization approach for calibrating VEP models.
• To enhance the accuracy and computational efficiency of VEP model parameter estimation.
• To improve the identification of pathological brain areas in drug-resistant epilepsy.

Main Methods:

• Utilized SaCeSS, a parallel cooperative metaheuristic algorithm for global optimization.
• Employed Bayesian optimization for hyperparameter tuning of VEP models.
• Implemented a scalable uncertainty quantification phase for parameter variability assessment.

Main Results:

• The SaCeSS algorithm demonstrated high-quality solutions and superior scalability compared to other parallel solvers.
• Bayesian optimization significantly improved VEP model accuracy and reduced computational cost.
• Uncertainty quantification provided insights into parameter estimation variability.

Conclusions:

• The proposed global optimization approach effectively calibrates VEP models.
• This method enhances the precision of identifying epileptic brain regions, supporting surgical planning.
• The study offers a pathway to improved clinical decision-making for epilepsy patients.