A Survey of Medical Point Cloud Shape Learning: Registration, Reconstruction and Variation

Point clouds have become an increasingly important representation for 3D medical imaging, offering a compact, surface-preserving alternative to traditional voxel or mesh-based approaches. Recent advances in deep learning have enabled rapid progress in extracting, modeling, and analyzing anatomical shapes directly from point cloud data. This paper provides a comprehensive and systematic survey of learning-based shape analysis for medical point clouds, focusing on three fundamental tasks: registration, reconstruction, and variation modeling. We review recent literature from 2021 to 2025, summarize representative methods, datasets, and evaluation metrics, and highlight clinical applications and unique challenges in the medical domain. Key trends include the integration of hybrid representations, large-scale self-supervised models, and generative techniques. We also discuss current limitations, such as data scarcity, inter-patient variability, and the need for interpretable and robust solutions for clinical deployment. Finally, future directions are outlined for advancing point cloud-based shape learning in medical imaging.

💡 Research Summary

This survey comprehensively reviews medical point‑cloud shape learning research published between 2021 and 2025, focusing on three core tasks: registration, reconstruction, and variation modeling. After a systematic PRISMA‑based literature search yielding 35 relevant studies, the authors first motivate the use of point clouds as a lightweight, surface‑preserving representation for CT, MRI, PET, and other volumetric modalities. They then organize the field into a progressive pipeline—segmentation feeds registration, which in turn supports reconstruction and variation analysis—highlighting how deep learning has supplanted classical ICP and Demons methods. In registration, recent works such as Adaptive Super4PCS, Graph‑ICP, LCNet, and cross‑modal frameworks like CSN‑ICP and PointVoxelFormer demonstrate robust rigid, deformable, and multimodal alignment, often leveraging graph neural networks, belief propagation, or synthetic data to mitigate sparse sampling and modality gaps. Reconstruction advances are driven by transformer‑based models (SA‑PoinTr, MSN‑FM) and hybrid point‑voxel architectures that achieve high F‑scores and low Chamfer distances on large benchmarks (MedShapeNet, MedPointS). Surface and topology recovery methods (Point2Mesh‑Net, CPA‑Conv‑POCO, PointNeuron) integrate occupancy networks, GCNs, and GANs to produce anatomically faithful meshes even from highly incomplete inputs. Variation modeling combines statistical shape models (DeepSSM, Point2SSM++) with deep latent‑space learning, enabling population‑level analysis and disease‑conditioned generation. The survey identifies persistent challenges: limited annotated data, high inter‑patient variability, and the need for interpretable, robust models suitable for clinical deployment. Future directions include weakly‑supervised and self‑supervised learning, synthetic data generation for cross‑modal bridging, end‑to‑end pipelines that jointly optimize registration and reconstruction, and privacy‑preserving techniques. Overall, the paper underscores the growing clinical impact of point‑cloud‑based shape learning across surgical planning, radiation dose reduction, patient communication, and multi‑center data sharing.