PUSSM: Point Cloud Upsampling as Implicit Statistical Shape Model

This paper proposes a framework for high-fidelity reconstruction of pelvic structures by integrating medical image segmentation and point cloud upsampling. By point cloud upsampling to learn shape priors from MedShapePelvic without requiring landmarks or PCA, our method functions as an implicit statistical shape model. Evaluations on Pelvic1k show significant improvements in surface quality and anatomical accuracy. This approach is generalizable and applicable to other skeletal regions.