AI-Powered Automated Model Construction for Patient-Specific CFD Simulations of Aortic Flows

TL;DR

The paper tackles the bottleneck of manual construction of patient-specific vascular models for image-based CFD by introducing an integrated deep-learning framework that automatically yields simulation-ready surfaces from medical images. It combines LoGB-Net for voxel-level segmentation with uncertainty quantification and a GNN-LDDMM-based surface deformation module for unsupervised, gradient-guided refinement, unifying segmentation and surface reconstruction in a single pipeline. The approach achieves state-of-the-art segmentation metrics and produces high-fidelity meshes that yield more accurate hemodynamic predictions than manual workflows, while substantially reducing processing time. This modular, uncertainty-aware framework positions image-based CFD as a practical tool for clinical decision support and can be extended to other vascular regions and imaging modalities.

Abstract

Image-based modeling is essential for understanding cardiovascular hemodynamics and advancing the diagnosis and treatment of cardiovascular diseases. Constructing patient-specific vascular models remains labor-intensive, error-prone, and time-consuming, limiting their clinical applications. This study introduces a deep-learning framework that automates the creation of simulation-ready vascular models from medical images. The framework integrates a segmentation module for accurate voxel-based vessel delineation with a surface deformation module that performs anatomically consistent and unsupervised surface refinements guided by medical image data. By unifying voxel segmentation and surface deformation into a single cohesive pipeline, the framework addresses key limitations of existing methods, enhancing geometric accuracy and computational efficiency. Evaluated on publicly available datasets, the proposed approach demonstrates state-of-the-art performance in segmentation and mesh quality while significantly reducing manual effort and processing time. This work advances the scalability and reliability of image-based computational modeling, facilitating broader applications in clinical and research settings.

Figures (8)

• Figure 1: Comparison of manual and AI-powered automatic model construction workflows for image-based simulations. Representative results at each step of our proposed framework are shown in the bottom panel.
• Figure 2: Comparison of segmentation results for two representative samples across different models. The top row for each sample shows 3D reconstructions of segmented aortic geometries, while the lower rows provide cross-sectional views at the supra-aortic branches (Branch) and the main aorta (Main aorta). LoGB-Net demonstrates superior performance, producing anatomically accurate and continuous segmentations, particularly in challenging regions such as the supra-aortic branches.
• Figure 3: Comparison of manual and automated surface reconstruction for two representative samples. Columns display the input CT image, manual and automated reconstructed surfaces, geometric difference maps, and the resulting pressure and wall shear stress (WSS) distributions from CFD simulations.
• Figure 4: Detailed cross-sectional comparison of reconstructed surfaces and source images for two representative samples. For each cross-section, overlays of the manual surface (red) and the automated surface (green) are shown against three background references: image magnitude, image gradient, and image Laplacian. The lower panels present the surface misalignment loss during training, where the automated surfaces achieve rapid convergence to lower loss values compared to static, manual surfaces.
• Figure 5: Uncertainty estimation and propagation through surface reconstruction and CFD simulation. (a) Segmentation uncertainty visualized across voxel-based predictions, smoothed geometries, and deformed surfaces, with uncertainty maps highlighting regions of variability. (b) Correlation between uncertainty (STD) and image quality (SNR), with corresponding surface contours. (c) Comparison of hemodynamic quantities from CFD simulations for smoothed and deformed surfaces.)