Research On Low Dose CT Image Denoising Algorithm Based On Generative Adversarial Network

With the continuous development of medical imaging technology,Computed Tomography（X-CT）has become one of the most widely used imaging methods in clinical diagnosis.Ionizing radiation produced by X-rays during image acquisition can cause serious harm to human body.To solve this problem,Low-dose CT was developed.The reduction of radiation dose inevitably introduces more noise and artifacts,resulting in the decline of image quality,and affecting doctors’ judgment of the condition.In recent years,the research on low-dose CT denoising using deep learning method has become a key topic in the field of medical imaging.In this paper,the low-dose CT image denoising algorithm based on Generative Adversarial Network is studied:（1）A Low-dose CT image denoising algorithm based on depth residual and compound perceptual loss is proposed to enhance the ability of Generative adversarial network to capture the detail information.The optimization is carried out from two aspects of network structure and loss function: the dilated convolution layer and the stacked residual module are used in the generator network to optimize the training process and solve the problem of gradient disappearance;Introduce a gradient penalty mechanism in the adversarial loss function,the weighted fusion of pixel-by-pixel loss and perceptual loss is used as the content loss function to obtain an optimized composite perceptual loss function,which limits the occurrence of gradient explosions,accelerates the network convergence speed,and promotes the denoising CT image perceptually close to the Full-dose CT image.Experimental analysis shows that the network can suppress most of the noise and artifacts in Low-dose CT images,and the generated denoising image retains the integrity of the structural features in the original image to obtain a better visual effect,which effectively verifies the feasibility of the improved method.（2）In order to further restore the texture details of CT images,a Low-dose CT image denoising algorithm based on multi-scale features and residual-attention mechanism is proposed to enhance the ability of Generative Adversarial Network to distinguish noise features.A multi-scale feature extraction module is set in the shallow layer network of the generator to fuse the information of different scales in the image and enhance the expression ability of the underlying features.The attention module is embedded in the residual structure of the generator and the end of the discriminator to improve the sensitivity of the network to noise and artifacts and to avoid the loss of details in the process of feature extraction.The experimental results show that the denoising image generated by the network can obtain excellent image quality,recover more texture details,and perform well in visual effect and evaluation index,which effectively verifies the superiority of the improved algorithm.