Bayesian sensitivity analysis of a 1D vascular model with Gaussian process emulators.

Area of Science:

• Computational biology
• Biomedical engineering
• Cardiovascular research

Background:

• One-dimensional cardiovascular models simulate pulse waves but require numerous parameters, hindering patient-specific applications.
• Sensitivity analysis is crucial for understanding parameter influence and output uncertainty in these models.
• Traditional Monte Carlo methods for sensitivity analysis are computationally intensive.

Purpose of the Study:

• To demonstrate the computational efficiency of variance-based sensitivity analysis for 1D vascular models using Gaussian process emulators.
• To compare this emulator-based approach against standard Monte Carlo methods.
• To assess the scalability of the proposed methodology.

Main Methods:

• Gaussian process emulators were employed for sensitivity analysis of 1D vascular models.
• The methodology was tested on four vascular networks of increasing complexity.
• Computational time and sensitivity indices were compared between the emulator and Monte Carlo approaches.

Main Results:

• The Gaussian process emulator approach reduced computational time for sensitivity analysis by 99.96% compared to Monte Carlo.
• Sensitivity indices obtained from both methods were comparable.
• The number of simulations needed for emulator training scaled as O(d), significantly better than Monte Carlo's O(d×103).

Conclusions:

• Gaussian process emulators offer a computationally efficient alternative for sensitivity analysis in 1D vascular models.
• This approach facilitates the development of patient-specific cardiovascular models.
• The method has the potential to yield clinically relevant results by quantifying parameter uncertainty impacts.