Research On Low-Dose CT Post-Processing Denoising Methods

Computed Tomography(CT)is a common clinical imaging technology,which uses radioactive rays to scan human bodies and obtain high-quality anatomical images of human tissues.But the X-rays used in CT imaging are harmful to the human body,causing genetic damage and even increasing the risk of cancer,especially for pregnant women and children.Reducing the tube current flow is usually used in clinical practice to reduce the dose of X-ray and reduce the harm caused by radiation.However,reducing the dose of X-ray will degrade the CT images where encounter more noise and artifacts,and finally affects the diagnosis and treatment of clinicians.Therefore,low-dose CT denoising has become an important research direction in the field of CT imaging.The post-processing method of low-dose CT denoising has become the main research direction of low-dose CT denoising in recent years because of its advantages such as no original projection data and fast processing speed.In this paper,based on the AAPM-Mayo open dataset,the post-processing method of low-dose CT denoising is studied,aiming to improve the quality of low-dose CT images.Specific research contents are as follows:(1)Aiming at the problems of large number of REDCNN model parameters,this paper designs a denoising model based on residual neural network RESNET,which significantly reduces the noise level and has a small number of parameters.In terms of network model,the image denoising block of the model adopts 1*1 convolution layer and residual connection to transfer shallow layer features directly to deep layer for channel feature fusion,and adds feature extraction layer before image denoising block to provide more useful feature information for it.After the image denoising block,the image reconstruction block is added to reconstruct the image by combining more feature information.In terms of loss function,the edge loss of image is added to correct L1 loss.Finally,the experimental results show that the performance of the proposed method is improved,and the PSNR of it is 0.49 higher than that of REDCNN.(2)Aiming at the large number of discriminator parameters in the existing WGAN model for low-dose CT denoising,this paper designs a GAN network discriminator structure based on SENET,which combined with WGAN-GP model for low-dose CT denoising,and significantly improved the visual effect of CT images.In terms of the discriminator structure,the channel attention mechanism of SE-Res Net is used to increase the dependency between channels,and the two-layer full connection layer is replaced by a convolutional structure,which significantly reduces the number of network parameters.In terms of loss function,in addition to the adversarial loss of WGAN-GP,edge loss is introduced to correct L1 loss and VGG loss is introduced to enhance the visual effect.Finally,the experimental results show that the proposed method has excellent visual effect.(3)A low dose CT denoising system based on CS architecture is designed and implemented.This paper uses Python programming language,combined with QT5,socket programming and Py Torch deep learning framework to design a low-dose CT denoising system with simple interface and easy operation.After testing,the system can realize denoising using different network models.The two proposed methods both improve the quality of CT image.Combined with the low-dose CT denoising system,it is helpful for clinicians to accurately diagnose the disease.