MedPCFM: Improving Medical Point Cloud Completion by Integrating Point Transformers and Flow Matching

This article introduces PCFM, a flow matching approach for medical point cloud completion that integrates Point Transformer v3 (PTv3) to address insufficiently studied generative modeling in this domain. The method is evaluated on the SkullFix, SkullBreak, and Mandibular Defect datasets against strong deterministic and diffusion baselines.

• PCFM with PTv3 achieves state-of-the-art generative performance across all tested datasets.
• The model requires substantially fewer sampling steps than diffusion-based methods like PCDiff.
• It provides up to a 7x speed-up compared to a PVCNN backbone at best operating points.
• Empirical scaling trends show consistent gains with higher point resolution and informative trade-offs across model scales.

The authors consider this important because it offers competitive accuracy while significantly improving throughput, making it more efficient for clinical workflows than existing diffusion approaches.