Abstract
Background and Objectives: Reconstructing three-dimensional vascular models for computational fluid dynamics (CFD) is time-consuming, limiting clinical application. Conventional centerline-based approaches generate redundant control points, reducing efficiency. We propose a machine learning-accelerated framework for rapid vascular reconstruction via centerline optimization.Methods: A surrogate neural network predicts the Dice similarity coefficient (DSC) for fast estimation of volumetric overlap. A composite score (CS) balancing geometric fidelity and data reduction guides optimal threshold selection.Results: The method achieves >500-fold computation time reduction while maintaining high accuracy. In an abdominal aortic network, control points were reduced by 70% with DSC = 0.8698. CFD simulations confirm hemodynamic consistency with original models.Conclusions: This approach provides an efficient, scalable solution for generating CFD-ready vascular models, with potential for clinical decision-making and interventional planning.

This publication has 15 references indexed in Scilit: