A U-shape Diffusion Network for Low-Dose CT Image Enhancement

Low-dose CT (LDCT) images offer a substantial reduction in radiation exposure in patients, mitigating the potential health risks associated with CT scanning. However, reduced dose often results in images with increased noise, blurred edges, and lost details. Hence, we propose a novel LDCT enhancement pipeline to improve image quality. Its core stage employs a diffusion-based approach that maps LDCT images to normal-dose counterparts, effectively suppressing complex noise in LDCT images and enhancing image clarity. The experimental results demonstrate the effectiveness of our method. Specifically, we achieve a 65% reduction in MSE, a 6.5% increase in SSIM, and a 16.7% improvement in PSNR. Additionally, the enhanced images exhibit more prominent lesion areas and clearer details. These findings suggest that our method has the potential to provide physicians with a more reliable basis to make accurate diagnoses, contributing to better clinical decision making and patient care. Code will be released at https://github.com/MIAinCS/ND-Net.