Pseudo-Shadow Removal in CT Images Using DDPM Architecture

An image affected by pseudo-shadow erosion can generally be divided into two parts: the pseudo-shadow erosion part and the unaffected part. Let ' represent the CT image domain affected by pseudo-shadow, and ' represent the CT image domain unaffected by pseudo-shadow. To remove the pseudo-shadow erosion, it is necessary to establish a mapping relationship from ' to ', converting the data affected by pseudo-shadow interference into data unaffected by pseudo-shadow interference.

By utilizing the characteristics of the ddpm architecture in step-by-step denoising and explicit image modeling ability, the complex noise patterns of pseudo-shadow can be removed from the original image.