AI Model Designs 3D-Printed Bone Scaffolds in 12 Seconds, About 700 Times Faster Than Topology Optimization

On October 31, 2026, Nanjixiong learned that researchers had published a PLOS One paper introducing ScaffoldForge, a generative AI model that designs 3D-printed bone scaffolds from target structural properties in approximately 12 seconds. Conventional topology optimization takes about 8,400 seconds per design.



The diffusion model takes eight measurements of trabecular structures, including solid volume, strut thickness and spacing, connectivity, and permeability. Starting from random noise, it generates a 3D voxel grid at a resolution of 64 through 1,000 iterations. Training penalizes three conditions: insufficient compressive strength, overhangs requiring supports, and low permeability. Tests produced scaffolds with approximately 56.6 MPa compressive strength, 8.2% above topology-optimized results, and approximately 700-fold faster generation on one NVIDIA A100 GPU. The team also demonstrated automatic detection of bone defects in patient CT scans, matching scaffold generation, and boundary trimming. Fitting error remained within 1 mm across 20 spinal CT cases.

Nanjixiong's Observations
Results remain computational: the scaffolds have not been physically printed or mechanically tested, and training data mainly consist of software-generated samples. Nevertheless, embedding manufacturability and mechanical constraints directly in generation, rather than validating afterward through simulation, clearly improves personalized implant design efficiency.

Source: 3D Printing Industry.
Original: https://3dprintingindustry.com/news/ai-model-designs-3d-printed-bone-scaffolds-in-12-seconds-256086/

AI-translated from an article by Kuaixiong on Nanjixiong, originally published on 2026-10-31. Text and images retain the original source attribution.
Read the original Chinese article