Title:ADEPT: A Noninvasive Method for Determining Elastic Properties of Valve Tissue

Abstract:Valvular heart disease accounts for up to 20% of cardiac surgery in the United States. Computer simulation of "virtual interventions" may inform optimal valve repair for a given patient prior to intervention. However, there is a paucity of methods to noninvasively determine in vivo mechanical properties of valve tissue from clinically acquired 3D images, limiting the accuracy of computer prediction and translational potential of in silico valve repairs. Here, we propose ADEPT, A noninvasive method for Determining Elastic Properties of valve Tissue, to overcome this methodological gap. Our framework combines image registration and physics-informed neural networks (PINNs) to estimate material properties of valve tissue from 3D echocardiograms (3DE). The PINN model was validated on a series of benchmarks before being applied to the 3DE of the tricuspid valve in a child with congenital heart disease. Our approach yielded accurate material parameter estimations in the examples accompanying this work.