This paper was accepted into the Machine Learning and Physical Sciences workshop at NeurIPS 2023.
During the last decades, hemodynamic simulators have constantly evolved and have become tools of choice for studying cardiovascular systems in-silico. This naturally comes at the cost of increasing complexity, since most modern models are nonlinear partial differential equations that depend on many parameters. While these tools are commonly used to simulate hemodynamics given physiological parameters, solving the related inverse problems (mapping waveforms to physiological parameters) has received comparatively less attention. Motivated by advances in simulation-based inference (SBI), we reconsider inverse problems specified by whole-body hemodynamics as statistical inferences. Unlike traditional analyses, SBI provides a multidimensional representation of uncertainty for individual measurements, encoded by posterior distributions. We performed in-silico uncertainty analysis on a focused set of physiological parameters of clinical interest and compared various measurement modalities. Beyond corroborating known facts, such as the feasibility of estimating heart rate, our study highlights the potential of estimating new physiological parameters from standard-of-care measurements. Furthermore, the SBI reveals practically relevant facts that alternative sensitivity analyzes do not detect, such as the existence of subpopulations for which the parameter estimation exhibits different uncertainty regimes. Finally, we study the gap between in vivo and in silico with the MIMIC-III waveform database and critically discuss how cardiovascular simulations can inform real-world data analysis.