Graph Neural Network Emulation of Cardiac Mechanics

This paper compares the performance of two graph neural network architectures on emulation a cardiac mechanic model left ventricle heart. These models can be applied directly same computational mesh geometry that is used by expensive numerical forward solver, precluding need for low-order approximation true geometry. Our results show these approaches incur negligible loss in accuracy compared simulator, while making predictions multiple orders magnitude more quickly, raising prospect their use both and inverse problems modelling.