Fast prediction of blood flow in stenosed arteries using machine learning and immersed boundary-lattice Boltzmann method

Abstract

A fast prediction of blood flow in stenosed arteries with a hybrid framework of machine learning and immersed boundary-lattice Boltzmann method (IB–LBM) is presented. The integrated framework incorporates the immersed boundary method for its excellent capability in handling complex boundaries, the multi-relaxation-time LBM for its efficient modelling for unsteady flows and the deep neural network (DNN) for its high efficiency in artificial learning. Specifically, the stenosed artery is modelled by a channel for two-dimensional (2D) cases or a tube for three-dimensional (3D) cases with a stenosis approximated by a fifth-order polynomial. An IB–LBM is adopted to obtain the training data for the DNN which is constructed to generate an approximate model for the fast flow prediction. In the DNN, the inputs are the characteristic parameters of the stenosis and fluid node coordinates, and the outputs are the mean velocity and pressure at each node. To characterise complex stenosis, a convolutional neural network (CNN) is built to extract the stenosis properties by using the data generated by the aforementioned polynomial. Both 2D and 3D cases (including 3D asymmetrical case) are constructed and examined to demonstrate the effectiveness of the proposed method. Once the DNN model is trained, the prediction efficiency of blood flow in stenosed arteries is much higher compared with the direct computational fluid dynamics simulations. The proposed method has a potential for applications in clinical diagnosis and treatment where the real-time modelling results are desired.

A Comparison of Newtonian and Non-Newtonian Models for Simulating Stenosis Development at the Bifurcation of the Carotid Artery

“…The LBM in a numerical method which mimics the Navier-Stokes equation based on a kinetic approach [27]. Although the LBM is a more recent development compared to alternative simulation approaches in computational fluid dynamics, such as the finitevolume and finite-element methods, it is now well established and considered as an alternative to the conventional approaches, and it includes blood flow [28][29][30][31][32][33][34] and non-Newtonian [35][36][37][38][39][40][41][42] simulations. It is considered to have a number of advantages over conventional methods [43]: incorporation of microscopic interactions, particularly for simulating complex fluids such as colloidal suspensions, bubbles, solid particle suspension and polymer-solvent systems [44]; suitability for parallel computing; and, of particular reliance to this study, dealing with complex boundaries [43] and implementing non-Newtonian fluid models in an efficient manner, as will be seen in Section 2.1.…”