Overview of Muscle Surrogate Models for Cardiac Cycle Simulations
Please login to view abstract download link
Biomechanical simulations of the left ventricle frequently utilize Huxley-type muscle models to capture the non-uniform and unsteady cardiac contractions. These models are governed by computationally intensive equations that track the distribution of myosin heads and actin-binding sites. The computational overhead often renders multi-scale simulations or full cardiac cycle studies impractical. To address this limitation, this work overviews two distinct surrogate modeling strategies: a data-driven approach and a physics-informed approach.In data-driven strategy, data are collected from existing numerical simulations, and recurrent and convolutional neural networks were trained to predict stress and instantaneous stiffness. Once integrated into a finite element solver, this surrogate model achieves significant speed-up benefits, allowing for the simulation of a complete cardiac cycle with reduced resource consumption. Alternatively, a physics-informed approach was developed to approximate the solution of the Huxley muscle contraction equation directly. In this approach network predicts crossbridge attachment probabilities that are used to calculate stresses and stress derivatives within the macro-level FE analysis. This approximation bypasses the computational requirements of the traditional method of characteristics at the micro-level of FE simulation. Together, these models demonstrate a potential to replace original Huxley-type simulations in large-scale biomechanical simulations.
