Atherosclerosis, aneurysms, and other vascular pathologies are closely associated with hemodynamic parameters. Non-invasive measurement of these hemodynamic parameters is of great significance for the prevention and treatment of cardiovascular diseases. In this study, eight idealized models of diseased vessels with varying stenosis rates and dilation extents were constructed. Transient simulations were performed under realistic physiological boundary conditions, and spatiotemporal coordinates as well as hemodynamic parameters were extracted from the models over one cardiac cycle to construct the dataset. Subsequently, by integrating hard boundary constraint methods with a physics-informed neural network (PINN), a prediction model for hemodynamic parameters in diseased vessels was established. The superiority of this model was demonstrated through validation on steady-state problems. With only limited supervised data, the proposed model achieves high-accuracy predictions of key hemodynamic parameters in lesioned vessels, including velocity, pressure, wall shear stress (WSS), time-averaged wall shear stress (TAWSS), oscillatory shear index (OSI), and relative residence time (RRT). The results indicate that the developed hemodynamic prediction model can accurately capture flow field characteristics such as velocity and pressure distributions, and exhibits excellent performance in predicting WSS and TAWSS. This study provides a novel approach and scientific basis for mechanistic investigations, clinical diagnosis, therapeutic strategies, and risk assessment of atherosclerosis, aneurysms, and related cardiovascular diseases.
