Automated model calibration with parallel MCMC: Applications for a cardiovascular system model

Area of Science:

• Computational physiology
• Biomedical engineering
• Medical imaging analysis

Background:

• Complex computational physiological models are difficult to calibrate with clinical data due to long computation times and limited data.
• Unique calibration of complex models is challenging, necessitating alternative approaches for patient-specific predictions.

Purpose of the Study:

• To develop an efficient pipeline for calibrating patient-specific computational physiological models.
• To reduce the parameter set for structural identifiability and improve Markov Chain Monte Carlo (MCMC) analysis efficiency.
• To enable reliable uncertainty quantification for task-specific model predictions.

Main Methods:

• Developed a pipeline to reduce fitting parameters for structural identifiability.
• Employed Markov Chain Monte Carlo (MCMC) analysis for optimal parameter estimation and confidence interval determination.
• Demonstrated the pipeline on a cardiovascular system model calibrated with brachial artery pressure, echocardiogram data, and synthetic 4D-flow MRI cerebral blood flow data.

Main Results:

• Reduced the cardiovascular model parameter set from 12 to 8-10 structurally identifiable parameters for three patients.
• Achieved significant efficiency improvements in determining pressure prediction confidence intervals compared to naive MCMC analysis.
• Obtained 95% confidence intervals for middle cerebral artery systolic blood pressure predictions smaller than ±10 mmHg (±1.3 kPa).

Conclusions:

• The developed pipeline effectively addresses challenges in calibrating complex physiological models for clinical application.
• The approach enables automated, efficient calibration using available clinical data and high-performance computing.
• The method provides accurate and reliable patient-specific predictions with quantified uncertainty, useful for risk assessment, such as for hemorrhagic stroke.