Predicting Hemodynamic Complexity in Thoracic Aortic Aneurysms with a POD–LSTM Surrogate Model

Abstract

Accurate prediction of blood flow dynamics in Thoracic Aortic Aneurysms (TAA) is essential for rupture risk assessment, yet remains computationally demanding with conventional CFD approaches. This study explores hemodynamic factors associated with TAA rupture using a Predictive Surrogate Model (PSM) integrating Proper Orthogonal Decomposition (POD) and Long Short-Term Memory (LSTM) networks. Transient hemodynamic parameters—including wall shear stress, pressure, and blood flow velocity—were simulated via finite volume-based computational fluid dynamics (CFD) across a cardiac cycle. The PSM framework leverages POD for dimensionality reduction and LSTM for temporal dynamics prediction, enabling efficient reconstruction of hemodynamic fields. A comprehensive evaluation of reconstruction errors across varying POD modes quantified deviations between the Full Order Model (FOM) and Reduced Order Model (ROM), validating the accuracy of the surrogate approach. Comparative analyses between CFD and PSM results demonstrated the machine learning model’s capability to capture complex flow patterns, with additional contrasts highlighting hemodynamic distinctions in normal vessels versus TAA cases. Results revealed elevated velocity fluctuations in the TAA region, increasing flow complexity and reconstruction challenges. This work underscores the potential of hybrid POD-LSTM frameworks for rapid hemodynamic assessment in clinical settings, while emphasizing the need for advanced modeling to address turbulence and biomechanical instabilities in aneurysmal pathologies. Our results indicate that the performed hybrid technique could efficiently (more than 85 %) predict the hemodynamic factors on the saccular surface TAA.