Global sensitivity analysis based on Gaussian-process metamodelling for complex biomechanical problems.

Area of Science:

• Biomechanical Engineering
• Computational Biology
• Mathematical Modeling

Background:

• Complex biomechanical models require numerous parameters, often with high determination costs and population variability.
• Identifying influential parameters is crucial for efficient model calibration and understanding.
• Traditional sensitivity analysis methods like Sobol's can be computationally prohibitive for complex models.

Purpose of the Study:

• To develop and validate a computationally efficient global sensitivity analysis method for complex biomechanical models.
• To identify influential parameters and quantify their interactions in computationally expensive models.
• To reduce the cost of experimental parameter determination and enhance inverse analysis.

Main Methods:

• Utilized Gaussian processes as a metamodel to approximate computationally expensive biomechanical models.
• Employed Sobol's global variance-based sensitivity analysis on the metamodel.
• Quantified uncertainties introduced by metamodelling and higher-order interaction effects.

Main Results:

• Successfully identified influential parameters in complex biomechanical systems, including nanoparticle drug delivery and arterial remodeling.
• Demonstrated that even limited evaluations of the full model suffice with the proposed metamodelling approach.
• Quantified higher-order parameter interactions and associated uncertainties.

Conclusions:

• Variance-based global sensitivity analysis is feasible and effective for complex, computationally expensive biomechanical models.
• The proposed metamodelling approach significantly reduces computational burden and experimental costs.
• This method provides a transparent and robust framework for parameter importance assessment in biomechanics.