Optical Coherent Tomography Driven Wall Shear Stress Prediction with Deep Neural Network in Projection Domain

With the development of engineering technology, the application of computational fluid dynamics (CFD) based on the Navier-Stokes equations is becoming increasingly widespread in fields such as climate and architecture. However, for some applications such as hemodynamics, CFD cannot be widely used due to computational costs and equipment limitations. In order to achieve the simple and fast acquisition of hemodynamic parameters, a neural network based parameter acquisition method is proposed. The proposed method uses the U-Net combined with an attention mechanism to learn the complete solution of wall shear stress calculated by the Navier-Stokes equations on a traditional CFD solver. The numerical results proved that compared with traditional CFD methods, the speed of the proposed model has been increased by three orders of magnitude, with low errors.