The use of Low-Dose CT (LDCT) examinations has significantly reduced the radiation dose of CT scans, but it has also led to increased noise and artifacts in CT images, thereby reducing image quality and accuracy, which affects doctors" diagnostic judgments. In recent years, generative models have demonstrated excellent performance in addressing this issue. However, generative models still face challenges such as generating confusing images and lacking interpretability during the generation process. To address this issue, a conditional diffusion denoising network model was constructed. On this basis, a trainable curve correction module was introduced to correct images with different noise levels, and a joint loss function was incorporated. Experimental results show that the proposed algorithm achieves better denoising results compared to baseline algorithms. Additionally, it demonstrates good generalization across LDCT images with different doses, maintaining excellent denoising performance.