Scholarlink

<p> Accurate simulation of blood flow dynamics in thoracic aortic aneurysms (TAA) is crucial for patient-specific risk assessment and treatment planning. However, high-fidelity computational fluid dynamics (CFD) models are computationally expensive and often impractical for real-time clinical use. In this study, we propose a Graph Neural Network (GNN)-based surrogate model to predict transient wall shear stress (WSS) components on the aneurysm wall using unstructured mesh data derived from CFD simulations. The model is trained on 70% of the cardiac cycle data and tested on the remaining 30%. We demonstrate that the GNN can accurately reproduce spatial and temporal WSS distributions, with average relative errors of approximately 5.5%, 6.5%, and 9% for <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\hbox {WSS}_x\)</EquationSource> </InlineEquation>, <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\hbox {WSS}_y\)</EquationSource> </InlineEquation>, and <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(\hbox {WSS}_z\)</EquationSource> </InlineEquation> components, respectively. Spatial error analyses reveal strong correlation between prediction error and flow complexity, particularly in regions near the aneurysm inlet where velocity gradients are highest. Our findings suggest that GNN-based models offer a promising, low-cost alternative for real-time hemodynamic assessment in complex vascular geometries.</p> Graphical abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Graph neural networks for surrogate prediction of hemodynamics in thoracic aortic aneurysm: a patient-specific study

  • 期刊论文

  • 发表日期：

    2026年6月15日

查看全文
  • Yazeed Alkhrijah,
  • Raed H. C. Alfilh,
  • Narinderjit Singh Sawaran Singh,
  • Wajdi Rajhi

摘要

Accurate simulation of blood flow dynamics in thoracic aortic aneurysms (TAA) is crucial for patient-specific risk assessment and treatment planning. However, high-fidelity computational fluid dynamics (CFD) models are computationally expensive and often impractical for real-time clinical use. In this study, we propose a Graph Neural Network (GNN)-based surrogate model to predict transient wall shear stress (WSS) components on the aneurysm wall using unstructured mesh data derived from CFD simulations. The model is trained on 70% of the cardiac cycle data and tested on the remaining 30%. We demonstrate that the GNN can accurately reproduce spatial and temporal WSS distributions, with average relative errors of approximately 5.5%, 6.5%, and 9% for \(\hbox {WSS}_x\) , \(\hbox {WSS}_y\) , and \(\hbox {WSS}_z\) components, respectively. Spatial error analyses reveal strong correlation between prediction error and flow complexity, particularly in regions near the aneurysm inlet where velocity gradients are highest. Our findings suggest that GNN-based models offer a promising, low-cost alternative for real-time hemodynamic assessment in complex vascular geometries.

Graphical abstract