Efficiency of a predictive surrogate model for hemodynamic predictions of blood flow in an idealized carotid artery stenosis

This study presents a predictive surrogate model (PSM) for predicting hemodynamic variables in idealized carotid artery stenosis. The PSM integrates long short-term memory (LSTM) and proper orthogonal decomposition (POD) techniques. The model's accuracy is evaluated for two different stenosis conditions in the stenosis. For the simulation of the full-order model of the carotid artery stenosis, a fluid–structure interaction (FSI) solver is employed to model the interaction between blood and the vessel wall. The Casson model is used to estimate the viscosity of the non-Newtonian blood flow. These conditions are selected to accurately capture blood hemodynamics across various stenosis conditions. The study examines pressure, wall shear stress (WSS), velocity components, and oscillatory shear index (OSI) variables. The reconstruction error for the reduced order model is calculated based on a chosen number of POD modes. It is noteworthy that the reconstruction error for OSI is significantly higher than that of other velocity components and derivatives (i.e., WSS) for both stenosis conditions. The POD and LSTM model is evaluated under different stenosis conditions, showing promising results despite the inherent complexities of physiological situations. While the model effectively predicts velocity components and WSS indices with reliable error scales, OSI exhibits slightly larger errors.