Back to all papers

Meshing around: evaluating the robustness of tetrahedral meshing algorithms for deep-learning segmented coronary arteries.

October 14, 2026pubmed logopapers

Authors

Fahmy S,Khan MO

Affiliations (2)

  • Department of Electrical, Computer and Biomedical Engineering, Toronto Metropolitan University, Toronto, ON, M5B 2K3, Canada.
  • Department of Electrical, Computer and Biomedical Engineering, Toronto Metropolitan University, Toronto, ON, M5B 2K3, Canada. Electronic address: [email protected].

Abstract

Computational fluid dynamics (CFD) simulations of coronary arteries are increasingly being used to assess coronary artery disease. While deep-learning segmentation methods have accelerated the reconstruction of coronary vasculature from computed tomography angiography (CTA), the integration of these segmented models into volumetric meshing algorithms to generate potentially CFD-usable meshes remains challenging due to complex geometries and surface deformities. This study investigated the robustness of three tetrahedral meshing algorithms, including TetGen, fTetWild and Ansys Fluent, in automatically generating potentially CFD-usable volumetric meshes from deep-learning segmentations. A 3D nnU-Net model was trained on 81 coronary CTA images to segment aorta and coronary artery trees from 29 test cases. The resulting surface labels were fed into the meshing algorithms without any manual pre-processing. Performance was evaluated based on meshing success rate, element count, and element quality metrics including Equiangle Skew, Scaled Jacobian, and Mean Ratio. Results demonstrated that fTetWild achieved the highest success rate (nearly 100%), outperforming TetGen and Ansys Fluent, particularly when handling low-quality geometries. Furthermore, fTetWild produced volume meshes with significantly fewer elements and superior worst-element quality which is critical for simulation convergence and stability. Although TetGen produced lower overall mesh quality with higher deviations, it maintained a higher success rate than Ansys Fluent under default parameters. fTetWild provides a highly robust and consistent solution for generating high-quality tetrahedral meshes directly from automated deep-learning segmentations, but with the caveat of higher computational cost. These findings present a crucial step toward developing fully automated, end-to-end vascular CFD simulation pipelines to streamline hemodynamic assessments in clinical environments.

Topics

Journal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAISlice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.