Past Event: Babuška Forum

A NEURAL NETWORK FINITE ELEMENT APPROACH FOR HIGH-SPEED CARDIAC SIMULATIONS

Professor Michael S. Sacks, UT Austin Biomedical Engineering

Abstract

Due to the complex multiphysics and multiscale nature of cardiac biomechanical function, traditional finite element methods remain prohibitively slow for clinical applications. To meet the requirements of speed as well as accuracy, we have developed and utilized a novel neural network finite element (NNFE) approach for soft tissue simulations that can produce simulation results within clinically relevant timeframes. The NNFE approach is a physics-based approach for rapid simulations that uses the neural network (NN) to represent the nodal displacements, and finite elements to map the displacement output from the NN on the problem domain, as well as to enforce boundary conditions and perform numerical integrations. In other words, this approach does not rely on data generated from physical experiments or simulations for training, rather, the NNFE model is trained to learn the governing PDE. In this work, we present a study using an extension of the NNFE approach towards complete organ level cardiac simulations to predict the P-V loop responses of the left ventricle, accounting for active contraction and transmural fiber distributions. We trained the model over two pressure-volume (P-V) loops and predicted the P-V relationship for a third loop. We compared the results of the NNFE model against an identical simulation setup in FEniCS.  The NNFE model predicted the displacements and corresponding PV loop, with mean nodal error between the NNFE solution and the FE solution was 2.32x10-2 mm.  The trained NNFE model took 2-3 seconds whereas FEniCS took 10-20 min. Consequentially, the NNFE approach is well-suited for the development of high speed cardiac digital twins, where one needs to solve very similar problems repeatedly.