基于条件扩散模型的 CT 金属伪影校正

Objective To propose a conditional diffusion model based on template priors for correcting streaky metal artifacts characterized by alternating bright and dark patterns that appear in reconstructed computed tomography (CT) images due to the presence of metallic implants. Methods After isolating metal region using image segmentation technology to generate metal trajectory projection data, the projection data of metal trajectory and the corresponding projection data of template prior images were excluded. Subsequently, a conditional diffusion model was developed to recover the missing portions of the projection data in the metal-affected regions. Finally, the restored projection data were reconstructed using filtered back projection to obtain the corrected image. Results Compared with artifact disentanglement network and diffusion model, the proposed approach improved peak signal to noise ratio by 3.44 and 0.749 dB, and increased structural similarity index by 0.079 and 0.015, respectively. Conclusion The proposed approach outperforms traditional metal artifact correction methods and exhibits superior performance in reducing both streaky and shadow artifacts.