Bayesian optimisation for efficient parameter inference in a cardiac mechanics model of the left ventricle

Area of Science:

• Computational mechanics
• Biomedical engineering
• Statistical modeling

Background:

• Cardio-mechanic models of the left ventricle are crucial for understanding heart function.
• Current parameter inference methods for these models are computationally expensive, hindering clinical application.
• The Holtzapfel-Ogden (HO) constitutive law is a complex model requiring numerical solutions.

Purpose of the Study:

• To develop a computationally efficient method for parameter inference in left ventricle cardio-mechanic models.
• To improve the clinical applicability of complex heart models, specifically those using the HO constitutive law.
• To integrate in vivo clinical data with ex vivo empirical laws for robust parameter estimation.

Main Methods:

• Utilized Bayesian optimization (BO), an efficient global optimization technique.
• Developed a surrogate model within the BO framework to approximate the black-box function of the HO model.
• Incorporated a penalty term based on ex vivo data to ensure realistic parameter estimates at high pressures.
• Applied the BO procedure to two case studies using real clinical data.

Main Results:

• The proposed Bayesian optimization procedure significantly reduced computational run times compared to traditional methods.
• The BO approach demonstrated superior performance in parameter inference for the HO constitutive law.
• The integrated penalty term ensured the reliability of parameter estimates even for unobservable high-pressure conditions.

Conclusions:

• Bayesian optimization offers an efficient and effective solution for parameter inference in complex cardio-mechanic models.
• This approach enhances the feasibility of using advanced heart models in clinical practice.
• The study validates the utility of BO for optimizing computationally intensive biomedical simulations.