A Novel Diffusion-Model-Based Bone Scan Image Inpainting Algorithm

Inpainting degraded region (e.g., due to patient wearing jewelry) in bone scan images is effective for improve image quality, robustness of machine analysis, and accuracy of disease diagnosis. In this paper, a novel diffusion-model-based bone image inpainting model is proposed, which recover missing regions in bone scan images with a pre-trained denoising diffusion probabilistic models(DDPM). Experimental results show that the average PSNR and SSIM values are better than the existing methods. the improvement in disease diagnosis was validated by changes in the accuracy of the classification network.The proposed algorithm presents a promising solution for a better method to inpaint bone scan image missing region.