Keywords: Geometric Deep Learning, Blood Flow Simulation, Vascular Graph
TL;DR: This paper presents a novel blood flow solver based on a implicit function learning on graph.
Abstract: Simulating blood flow is paramount in identifying flow-based biomarkers for vascular-related diseases. A segmented vessel graph is used as a domain for the simulation. Traditionally, partial differential equations are solved with numerical methods. Here, we propose an alternative solver for the simulation of blood flow on a vascular graph leveraging geometric deep learning. Specifically, we reformulate the problem as an implicit function on the graph and learn the simulation by imposing the physics in the loss through a message passing layer. The resultant flow is accurate, fast, and applicable to various tasks.