Low Dose CT Image Reconstruction Method Based On CNN And Transformer Coupling Network

Computed tomography(CT)technology is the most widely used medical imaging technology.However,patients will be exposed to X-ray radiation in the process of being scanned,increasing the risk of disease.Therefore,low-dose CT has become the current research focus.One way to achieve low-dose scanning is to reduce the radiation dose at each angle.However,this will increase the noise of the projection data,and then make the reconstructed image noisy.In this way,the de-noising problem of low-dose CT image becomes the key problem to realize low-dose scanning.In recent years,the depth learning denoising technology of CT images has become the most advanced denoising technology.The fundamental reason is that depth learning can model the complex nonlinear mapping between noisy images and high-quality images,which is difficult to be described by mathematical models.Most of the existing work uses convolutional neural networks(CNN).Its biggest advantage is that it can efficiently model local information of images,but its ability to use global information is limited.Recently,Transformer,which has been successfully applied in natural language processing and computer vision,can well model the global information for long-range dependence.Obviously,if the two can be organically coupled to build a depth network,it is expected to further improve the performance of low-dose CT image denoising.For this reason,this paper studies low-dose CT image denoising based on CNN and Transformer coupling network.The main work is as follows:(1)A kind of CTC(CNN-transformer-coupling)network is proposed.With CNN and the improved Transformer coupling module as the main components,the local information association ability of CNN and the global information capture ability of Transformer are comprehensively utilized,and the residual connection mechanism and information multiplexing mechanism are used to achieve high-precision noise removal of low-dose CT images.(2)A kind of NCCTC(Non-local CNN-Cross Transformer-coupling)network is proposed.On the basis of the CTC network,the core components are improved,and the Ushaped structure is introduced in the way of non-local CNN and cross transformer coupling to expand the receptive field and further improve the noise removal accuracy.(3)A low-dose CT image denoising system based on CNN and Transformer coupling network is designed and implemented.The system can read and process noisy images and write high-quality images.The organic coupling of CNN and Transformer further improves the deep learning denoising ability of low-dose CT images.This research has certain theoretical significance and practical value.