Computational modelling of the cardiovascular system is a promising future direction for patient-specific healthcare. However, the computational cost of these simulators is a bottleneck for their practical use in clinic for real-time digital twins. Emulation can overcome this, yet an extensive investigation into cardiovascular emulators is warranted. In this study, we emulate two one-dimensional haemodynamics models of the pulmonary circulation and compare two common emulation strategies: Gaussian processes (GPs) and polynomial chaos expansions (PCEs). We start by reducing the parameter space of the models through global sensitivity analysis, and then compare both emulation strategies using a multivariate, time-series output quantity of interest and a reduced representation using principal component analysis. We compare the emulators in both forward emulation on test data, as well as in their ability to infer parameters in the inverse problem. Our results indicate that GPs slightly outperform PCEs consistently across every comparison, and that a similar performance is obtained for the emulators of the time-dependent output and reduced output. This article is part of the theme issue ‘Uncertainty quantification for healthcare and biological systems (Part 1)’.
Paun et al. (2024) studied this question.