面向血管形态学分析的点云统计形状模型构建与研究OA

Traditional mesh-based statistical shape models(SSMs)are widely used in medical image analysis but face challenges in handling complex vascular geometric shapes such as bifurcations,curvatures,and plaque deformations due to insufficient precision.A point cloud-based SSM is proposed that directly processes discrete three-dimensional(3D)point cloud data using a fusion algorithm of dynamic graph convolutional neural networks(DGCNNs)and spatial attention mechanisms,establishing a topology-free vascular morphological model.The experimental study utilizes a dataset of 114 carotid artery time-of-flight magnetic resonance angiography(TOF-MRA)images,preprocessed with a hybrid filtering method,to construct a comparative framework of point cloud and mesh models.The results demonstrate that the point cloud model significantly enhances the uniformity of grid quality by 42%(the coefficient of variation of the Jacobi coefficient changes from 0.13 to 0.08),making it more suitable for subsequent fluid dynamics simulation analysis.Additionally,under the condition of preserving 90%of the shape variation,the point cloud model reduces the principal component dimensions by 18%compared to the traditional mesh model(9 vs 11).In subsequent specificity and generalization evaluations,the point cloud model exhibits stronger robustness.