Abstract

Co-Authors: Marina Strocchi, Francesco Regazzoni, Christoph Augustin, Luca Dede’, Steven Niederer, Alfio Quarteroni. Cardiac digital twins provide a physics- and physiology-informed framework for predictive and personalized medicine. However, high-fidelity multi-scale and multi-physics cardiac models remain a barrier to adoption due to their high computational cost and the large number of model evaluations required for patient-specific personalization. Artificial intelligence-based methods can enable the creation of fast and accurate whole-heart digital twins. We use Latent Neural Ordinary Differential Equations (LNODEs) to learn the temporal pressure-volume dynamics of a heart failure patient. Our LNODE -based surrogate model is trained from 400 3D-0D whole-heart closed-loop electromechanical simulations, taking into account 43 model parameters describing cell-to-organ scale cardiac electromechanics and cardiovascular hemodynamics. The trained system of LNOD Es provides a compact and efficient representation of the 3D-0D model in a latent space using a feed-forward fully connected artificial neural network that retains 3 hidden layers with 13 neurons per layer, enabling faster than real-time numerical simulations of cardiac function on a single processor. This surrogate model is employed to perform global sensitivity analysis and robust parameter estimation with uncertainty quantification in time frames compatible with clinical practice, still using a single processor. This framework introduces several computational tools for digital twinning in computational cardiology.