Fast parameter inference in a biomechanical model of the left ventricle by using statistical emulation

Area of Science:

• Biomedical Engineering
• Computational Biology
• Cardiovascular Research

Background:

• Personalized left ventricular modeling requires estimating myocardial material properties from clinical data.
• Current numerical methods for solving biomechanical models are computationally expensive, limiting clinical application.

Purpose of the Study:

• To develop and validate an efficient emulation strategy for inferring left ventricular myocardial properties within a clinically relevant timeframe.
• To compare different emulation approaches and identify the optimal combination for accurate parameter inference.

Main Methods:

• Utilized emulation techniques, replacing computationally intensive differential equation simulations with surrogate models.
• Compared two emulation strategies: output emulation and loss emulation.
• Tested strategies with various interpolation methods (e.g., Gaussian process) and loss functions (e.g., Euclidean loss).

Main Results:

• The output emulation method combined with local Gaussian process interpolation and Euclidean loss demonstrated accurate parameter inference.
• Achieved a computational cost reduction of approximately three orders of magnitude compared to traditional finite element methods.
• Successfully validated the approach on both simulated and real clinical magnetic resonance imaging data.

Conclusions:

• Emulation offers a computationally efficient alternative for personalized left ventricular modeling.
• The validated emulation strategy enables rapid estimation of myocardial properties, supporting timely clinical decision-making and personalized medicine.
• This approach significantly accelerates biomechanical analysis for heart function assessment.