Assessing parameter identifiability of a hemodynamics PDE model using spectral surrogates and dimension reduction

Area of Science:

• Computational science
• Biomedical engineering
• Mathematical modeling

Background:

• Biomedical simulators face challenges with limited data and high parameter dimensionality.
• Sensitivity analysis is crucial for parameter ranking in medical digital twins but can be heuristic.
• Expensive simulations necessitate emulation for faster computation.

Purpose of the Study:

• To develop an innovative solution for parameter identifiability and sensitivity analysis in computational inverse problems.
• To leverage polynomial chaos expansions (PCEs) for multioutput global sensitivity analysis.
• To enable profile-likelihood analysis for problems governed by partial differential equations.

Main Methods:

• Utilized polynomial chaos expansions (PCEs) for global sensitivity analysis and parameter identifiability.
• Applied dimension reduction to quantify time-series sensitivity in a pulmonary hemodynamics model.
• Constructed univariate profile-likelihood confidence intervals using PCEs.

Main Results:

• Demonstrated efficient quantification of time-series sensitivity for a 1D pulmonary hemodynamics model.
• Showcased how experimental design changes can improve parameter identifiability.
• Successfully enabled profile-likelihood analysis for partial differential equation-governed problems via emulation.

Conclusions:

• PCEs offer a novel and effective approach to determining parameter identifiability in complex biomedical models.
• The proposed method addresses limitations of heuristic sensitivity analysis.
• This work integrates emulation strategies with profile-likelihood analysis for enhanced biomedical simulation.