Point Cloud Diffusion Models for Automatic Implant Generation
Advances in 3D printing of biocompatible materials make patient-specific implants increasingly popular. The design of these implants is, however, still a tedious and largely manual task that is usually carried out by trained experts. In order to simplify and accelerate the design process, we developed a method for automatic implant generation [1].
The proposed method (Fig. 1) is based on a conditional denoising diffusion probabilistic model for point cloud completion (Fig. 2) and a voxelization network leveraging a differentiable Poisson solver. The network learns to generate complete skull anatomies from segmented CT images of patients showing bone defects . A suitable implant design can be derived by subtracting the defective from the complete skull. The presented model is capable of producing high quality implants for a large variety of different defects like complicated fronto-orbital and large bilateral defects (Fig. 3). Regardless of the complexity of the implant to be generated, this only takes about 20 minutes, which is negligible compared to the time required to produce an implant.
More information and relevant links can be found on the project page: https://pfriedri.github.io/pcdiff-implant-io/
Figure 1: Overview of the proposed automatic implant generation method. A binary segmentation mask Sd of a defective skull serves as input to the pipeline, which is trained to produce a complete version Sc of the skull. The implant I derives from the subtraction of complete and defective skull.
