Efficient low-dose CT reconstruction with ghost-integrated U-Net discriminator

The high-radiation doses associated with traditional computed tomography (CT) scans may elevate cancer risks, leading to significant interest in low-dose CT imaging technology. However, existing low-dose imaging techniques face various challenges, including difficult data acquisition, high computational costs, and poor imaging quality. We propose an innovative generative network architecture that integrates a meticulously designed U-Net with RED-CNN. Introducing the ghost bottleneck module and dynamic fusion of context attention mechanism enhances computational efficiency and image quality. Experimental results on the Mayo dataset showcase the exceptional performance of this method, which also outperforms other denoising algorithms in comprehensive evaluations. This provides a new technical pathway for low-dose CT image reconstruction, aiding in reducing radiation exposure risks while ensuring diagnostic accuracy in medical imaging.