Calibration of Cohorts of Virtual Patient Heart Models Using Bayesian History Matching

Area of Science:

• Biomedical Engineering
• Computational Biology
• Medical Informatics

Background:

• Patient-specific models are crucial for personalized medicine but are computationally expensive.
• Current calibration methods treat each patient independently, hindering clinical scalability.

Purpose of the Study:

• To develop and validate a simulation reuse strategy for accelerating patient-specific model calibration.
• To reduce the computational burden of creating patient-specific models for widespread clinical use.

Main Methods:

• Utilized a Statistical Shape Model for anatomical representation and electrophysiological simulations for functional representation.
• Employed Bayesian History Matching (BHM) with Gaussian Process Emulators (GPEs) trained on 14 biomarkers.
• Iteratively ruled out implausible parameter space regions based on simulation outcomes.

Main Results:

• Achieved 87.41% identification of non-implausible parameter combinations without additional simulations.
• Demonstrated that reducing measurement uncertainty from 10% to 5% decreased the parameter space by six orders of magnitude.
• Successfully accelerated the patient-specific model calibration pipeline.

Conclusions:

• The proposed simulation reuse technique significantly enhances the efficiency of patient-specific model calibration.
• This method offers a computationally lighter approach, paving the way for broader clinical adoption.
• Optimizing measurement uncertainty is key to drastically reducing the parameter space and improving model fitting.