Low-Dose CT Denoising Algorithm Based on Fully Convolutional Neural Network

To address the issue of severe noise caused by reduced radiation dose in low-dose medical CT images, which leads to significant image quality degradation, this paper proposes a denoising algorithm for low-dose CT images based on residual attention mechanisms and adaptive feature fusion. The proposed method utilizes a fully convolutional neural network to complete the denoising task. Within the network architecture, a residual attention mechanism and selective internal core feature fusion module are introduced to filter noise information, extract effective features, and adaptively integrate image characteristics. This approach avoids detail loss during reconstruction and improves image quality, making the denoised image perceptually closer to the original image. Both qualitative and quantitative experiments demonstrate that the proposed method effectively suppresses noise and restores more detailed textures in low-dose CT images. Compared with conventional methods, the proposed algorithm improves peak signal-to-noise ratio by 14.94%, enhances structural similarity by 4.68%, and reduces root mean square error by 40.11%, meeting the diagnostic requirements of medical imaging.