To address the needs of geometric modeling and hemodynamic analysis in the noninvasive functional assessment of coronary heart disease, this study establishes a deep learning-based integrated framework for coronary artery segmentation, three-dimensional reconstruction, and hemodynamic analysis. The nnU-Net model was used to achieve automatic coronary artery segmentation, and point cloud modeling techniques based on the Point Cloud Library (PCL) were further applied to construct patient-specific coronary artery models. Based on the reconstructed models, coronary hemodynamic numerical simulations were performed to analyze the variation characteristics of key indicators, including the blood flow velocity field, fractional flow reserve derived from computed tomography (CT-FFR), and wall shear stress (WSS), under different degrees of coronary artery stenosis. The results showed that, with increasing stenosis severity, local flow acceleration and high-velocity jet flow in the stenotic segment were enhanced, while the pressure distal to the stenosis decreased, leading to a gradual reduction in CT-FFR. Meanwhile, WSS increased in the stenotic region and adjacent vessel walls, and the high-WSS region expanded. These results indicate that the combination of machine learning-based three-dimensional reconstruction and hemodynamic analysis can help compensate for the limitations of conventional CT imaging in functional assessment, providing a numerical simulation basis for the noninvasive functional evaluation of coronary artery stenosis.