Low-dose CT image denoising network based on feature refinement

Deep learning-based methods now play a key role in enhancing low-dose CT images by reducing noise. However, overly aggressive denoising typically causes the disappearance of significant texture information in the image. To solve this problem, a novel feature refinement residual coding-decoding network (FRRED) is proposed in this paper. By introducing a feature refinement module, the network further processes the extracted image features to enhance the details such as edges and textures in the image. In addition, we introduce the channel attention module, which is able to automatically learn the importance weights of each channel, focusing more attention on the key channels related to reconstruction, thereby enhancing the feature expression of these channels while suppressing the influence of unimportant channels. The results of the experiments indicate that, unlike the widely - adopted techniques for denoising low - dose CT images, the proposed method has achieved substantial improvements in various evaluation indicators, thereby effectively enhancing the quality of low-dose CT images.