Data-driven and Physics-informed Muscle Model Surrogates for Cardiac Cycle Simulations

Health professionals can utilize biomechanical simulations of left ventricle to assess different possible situations and hypothetical scenarios. Understanding of the molecular mechanisms behind muscle contraction has resulted in the development of Huxley-like muscle models. Unlike Hill-type muscle models, Huxley-type muscle models can be used to simulate non-uniform and unstable contractions. However, Huxley models demand considerably more computational resources than Hill models, which limits their practical use in large-scale simulations. To address this, we have developed a data-driven and physics-informed surrogate models that mimic the Huxley muscle model, while requiring significantly less processing power. We collected data from various numerical simulations and trained deep neural networks to replace Huxley's muscle model. Data-driven surrogate model was an order of magnitude faster than the original model, while being quite accurate. Our surrogate models were integrated into a finite element solver and used to simulate a complete cardiac cycle, which would be much harder to do with original Huxley's model.