Adaptive Meshing Based on the Multi-level Partition of Unity and Dynamic Particle Systems for Medical Image Datasets

Surface meshes extracted from sparse medical images contain surface artifacts, there will produce serious distortion and generate numerous narrow triangle meshes. In order to eliminate the impact of the above factors, this paper presents a novel method for generating smooth and adaptive meshes from medical image datasets. Firstly, extracting the stack of contours by means of image segmentation and translating the contours into point clouds. The improved Multi-level Partition of Unity (MPU) implicit functions are used to fit the point clouds for creating the implicit surface. Then, sampling implicit surface through dynamic particle systems based on Gaussian curvature, dense particles sampling in the high curvature region, sparse particles sampling in the low curvature region. Finally, generating triangle meshes based on particle distribution by using the Delaunay triangulation algorithm. Experimental results show that the proposed method can generate high-quality triangle meshes with distributed adaptively and have a nice gradation of triangle mesh density on the surface curvature.

从稀疏医学图像中提取的表面网格包含表面伪影，会产生严重变形并生成大量狭窄的三角形网格。为了消除上述因素的影响，本文提出了一种从医学图像数据集生成平滑且自适应网格的新方法。首先，通过图像分割提取轮廓堆栈，并将轮廓转换为点云。使用改进的多层次单位分解（MPU）隐式函数拟合点云以创建隐式曲面。然后，基于高斯曲率通过动态粒子系统对隐式曲面进行采样，在高曲率区域密集采样粒子，在低曲率区域稀疏采样粒子。最后，使用德劳内三角剖分算法根据粒子分布生成三角形网格。实验结果表明，所提出的方法能够生成自适应分布的高质量三角形网格，并且在表面曲率上三角形网格密度具有良好的渐变。