tsta

Home / سی و دومین کنفرانس ملی و دهمین کنفرانس بین المللی مهندسی زیست پزشکی ایران

Physics-Informed Neural Networks for Cardiac Flow Estimation in 2D Simplified Human Right Ventricular Geometry

Authors :

Mohammadmahdi Sekhavatpisheh¹ Nasser Fatouraee²

1- دانشگاه صنعتی امیرکبیر (پلی تکنیک تهران) 2- دانشگاه صنعتی امیرکبیر (پلی تکنیک تهران)

Keywords :

Blood Pressure Computation،Computational Fluid Dynamics،Medical Imaging،Non-Invasive Measurement،Physics-Informed Neural Networks،Right Ventricle

Abstract :

One of the significant disorders in the cardiovascular system is pulmonary hypertension associated with the right ventricle, where estimation of pressure resulting from blood flow is critical for assessment. Evaluation of the pressure enhances the diagnostic and therapeutic capacity for the disease. So, pressure measurement without causing wounds or injury to the patient is of paramount importance. 4D magnetic resonance imaging is one of the novel imaging modalities capable of providing useful information to clinicians. However, complete extraction and analysis of blood flow dynamics using existing hardware with raw imaging data is currently not fully accessible. This study demonstrated that non-invasive computation of blood flow pressure in the right ventricle can be achieved using physics-informed neural networks with minimal data of the velocity field obtained from imaging. This method can estimate the information faster and more accurately compared to conventional methods, eliminating the need to define geometry, mesh, and boundary conditions. The capability of this method was demonstrated through solving two problems: first, the "lid-driven cavity" problem, and second, "steady blood flow within a realistic 2D right ventricular geometry," where the relative errors for pressure field computation were 2.7% and 1.9%, respectively.