Non-Invasive Young modulus and aortic root flow reconstruction from scarce data using Physics-Informed Neural Networks

Abstract

Two clinical scenarios focused on inference of nonmeasurable cardiovascular signals and parameters within the human arterial system are presented. In the first scenario, we use Physics-Informed Neural Networks (PINNs) constrained with zerodimensional (0D) haemodynamical models to jointly
reconstruct the Young modulus of each of the elastic vessels of a mock arterial circulatory system, from lumen area in-silico data. In the second scenario, we use PINNs to reconstruct the flow rate at the ascending aorta inlet of a healthy adult thoracic network using exclusively an in-vivo pressure signal measured non-invasively with tonometry at the left common carotid artery.