Exploring Hemodynamics Simulators and Simulation-Based Inference
A new paper has been accepted at the workshop Machine Learning and the Physical Sciences at NeurIPS 2023. Over the years, hemodynamics simulators have evolved into important tools for studying cardiovascular systems in-silico. These models have become more complex, relying on non-linear partial differential equations with many parameters. While these simulators are routinely used to simulate hemodynamics based on physiological parameters, solving the related inverse problems has not received as much attention.
Reconsidering Inverse Problems
Motivated by advances in simulation-based inference (SBI), researchers are now approaching the inverse problems of whole-body hemodynamics as statistical inferences. Unlike traditional analyses, SBI provides a multi-dimensional representation of uncertainty for individual measurements, as encoded by posterior distributions. An uncertainty analysis was performed in-silico on a focused set of physiological parameters of clinical interest, and several measurement modalities were compared. The study highlighted the potential of estimating new physiological parameters from standard-of-care measurements.
Uncovering New Insights
The study not only confirmed known facts, such as the feasibility of estimating heart rate, but also revealed the potential to estimate new physiological parameters from standard-of-care measurements. Additionally, the research showed that simulation-based inference uncovers practical facts missed by alternative sensitivity analyses, such as the existence of sub-populations with distinct uncertainty regimes in parameter estimation.
Informing Real-World Data Analysis
Finally, the study bridged the gap between in-vivo and in-silico with the MIMIC-III waveform database, critically discussing how cardiovascular simulations can inform real-world data analysis. These findings have the potential to significantly impact research and development in the field of hemodynamics and cardiovascular systems.