A Research On Low Dose CT Images Super Resolution Reconstruction Technology Based

Computed tomography is an important means of clinical diagnosis and treatment at present,but ionizing radiation which causes harm to human body will be produced during CT scanning.For the sake of cutting down the harm of ionizing radiation to human body,low-dose X-ray scanning has gradually become a hot spot in clinical applications.However,reducing the radiation dose will lead to a lot of noise in low-dose CT and reduce the image resolution.Therefore,the denoising and super-resolution reconstruction of low-dose CT images have important research value and medical clinical significance.With low-dose CT image processing as the background,this paper carrys out research on low-dose CT super-resolution reconstruction and its key technologies,in allusion to the problem that the network structure of traditional reconstruction model is shallow and the reconstruction result is easily affected by noise.In view of the problem that the reconstruction process is easily affected by noise and generates edge distortion,a low-dose CT denoising algorithm based on RN-CNN model is proposed.The image is denoised before reconstruction.Aiming at the problems of the existing reconstruction network,such as shallow structure,inadequate learning of image features and inadequate utilization,a multi-scale residual generation antagonism network was proposed,and multi-scale modules were designed to fully extract the image details and improve the reconstruction effect.The main content of this paper is as follows:First,the development of low-dose CT,low-dose CT denoising and super-resolution reconstruction algorithms are summarized.The application of CT equipment and deep learning in image processing is analyzed.The influence of tube voltage and tube current on CT image gray value is analyzed by e Xperiment.Then,Low-dose CT scanning is an effective mean to reduce the X-ray dose.In order to solve the problem of noise suppression in low-dose CT images,this paper proposes a Res Netbased convolutional neural network(RN-CNN)denoising method.Firstly,in order to simulate low-dose image,Poisson Gaussian mi Xed noise is added to the normal measurement image.Then,the low-dose CT input into RN-CNN is used to e Xtract the image features,and the residual network and the spatial pyramid-pool(S-SPP)with invariant scale are introduced to avoid the problem of accuracy decline and increase the network effective features.At the same time,the e Xpansive convolution is used in the process of convolution to increase the receptive field of the network,so as to retain the internal data structure of the image and obtain better segmentation effect.Finally,the low-dose CT image and the noise image were separated to obtain the normal dose CT image.Then,in view of the problem of low-dose CT image resolution,this paper proposes a super-resolution reconstruction algorithm based on multi-scale residual-generation countermeasure network(MSRGAN),which can recover high-resolution(HR)images from low-resolution(LR)images with the pathology remaining unchanged.Multi-scale networks can make full use of image features of different sizes to enrich image details and improve the utilization rate of features in the reconstruction process.The residual network is introduced to prevent overfitting while realizing feature reuse.Finally,by combining the counter loss with the content loss,the reconstructed image with better perceived quality can be obtained when the constraint feature is generated.The results show that this method increases the SSIM,FSIM and PSNR inde Xes by 0.047,0.0228 and1.962 respectively,and GAN’s IS,FID and SWD performance IS better than the other two GAN algorithms,and it has a better performance in the detail surface of edge contour,which fully prove the effectiveness of the algorithm presented in this paper.Low dose CT image quality is poor,which affects doctors’ diagnosis and treatment.In this paper,the proposed algorithm is used to denoise low-dose CT images and reconstruct the denoised images.The experimental results show that the proposed algorithm improves the diagnostic accuracy as well as the diagnostic efficiency.