Cancellous Bone μCT Image Generation with Diffusion Models

Deep learning has demonstrated great advantages in analyzing the microstructural features and biomechanical properties of high-resolution cancellous bone micro-computed tomography (μCT) images, providing reliable diagnostics that enable dynamic monitoring of various clinical orthopedic conditions. However, the high difficulty in acquiring μCT images and concerns over patient privacy have limited the quality and quantity of available datasets for related deep learning tasks. Generating synthetic data offers a promising alternative. In this study, we explored the use of diffusion models to generate synthetic images from high-resolution cancellous bone μCT scans, conditioned on morphological parameters such as bone porosity, trabecular thickness, and pore size. Our results show that the model produces realistic data, and that conditioning variables can be effectively used to control the image generation process.