One-step method speeds up cranial implant generation from point clouds
MedPCFM-TED: One-Step Point Cloud Flow Matching for Implant Generation via Teacher-Guided Endpoint Distillation
Computer Vision and Pattern RecognitionMachine Learning
Summary
Making custom cranial implants from 3D scans is important but can be slow because it often requires many steps. The authors developed a one-step method called Teacher-guided Endpoint Distillation (TED) that generates implants quickly while keeping accurate shapes. TED learns from a longer process (the teacher) to predict the implant directly, cutting down time dramatically. This approach works well on standard test datasets for skull implants, matching or beating existing methods in quality.
What this means in practice
- •For medical device designers: Generate accurate custom cranial implants faster during surgical planning using point cloud data and one-step flow matching.
- •For 3d printing services: Produce multiple plausible implant candidates rapidly for evaluation and printing based on 3D skull scans.
Authors
Kamil Kwarciak, Marek Wodzinski
Abstract
Cranial implant generation is an important task in medical imaging. Recent point cloud based generative methods, particularly flow matching, offer strong reconstruction quality and efficient sampling, but still require multiple neural function evaluations during inference. This limits rapid generation of multiple plausible implant candidates. We propose Teacher-guided Endpoint Distillation (TED), a simple one-step distillation framework for conditional cranial implant generation on point clouds. TED trains a one-step student using teacher-guided endpoint supervision and geometric matching losses, while avoiding explicit path straightening. We evaluate TED on the SkullFix and SkullBreak benchmarks. TED achieves the best overall performance on the SkullBreak dataset, remains competitive on SkullFix, and provides the strongest Chamfer distance performance among the compared one-step methods. In addition, TED generates implants in approximately 0.04s per sample. These results show that one-step distillation can substantially accelerate conditional point cloud implant generation without sacrificing reconstruction quality.