| Computed Tomography(CT)has been widely used in clinical diagnosis,industrial inspection,material science,and biomedicine,bringing great value to human health and industrial development.In addition,the raging epidemic brought by the 2019-nCov has made CT known to the public as an effective assistive technology.However,the associated X-ray radiation from CT brings a potential risk of cancer to the scanned object,so the need to reduce the radiation dose of CT scanning is becoming more and more urgent.Generally speaking,Low Dose Computed Tomography(LDCT)can be achieved by reducing current and voltage,sparse sampling CT,and limited-view CT.Among them,limited-view CT is currently the most common LDCT strategy in the industry due to inevitable mechanical and physical limitations in practice.However,due to the lack of prior information in limited-view CT data,it will cause serious damage such as artifacts and blurring in its imaging.Therefore,the prior information of limited-view CT data must be fully utilized to perform high-quality restoration and reconstruction,so that the final imaging results can be successfully applied to various practical uses.However,the current mainstream algorithms focus on restoring and reconstructing a single CT image on a single domain,but ignore the information gain that different spatial domains can bring to data restoration,as well as the third-dimensional spatial information contained between consecutive CT images.Therefore,in this subject,a pioneering dual-domain limited-view CT restoration and reconstruction algorithm based on spatial information is proposed.The deep learning model is used to effectively fuse the Radon domain and the image domain to provide reliable information gain for data restoration,and also fully excavate and utilize the spatial coherence between consecutive CT images to perform high-quality CT image restoration.Specifically,in the first stage,in order to make better use of the prior information in the Radon domain,it is first sent to the data preprocessing module for preliminary completion of the Radon data;in addition,this subject proposes a lightweight neural network named L-AAE that can achieve high-quality restoration of limited-view Radon data on the premise of greatly reducing the overall computational complexity.In the second stage,the Radon data generated from the previous stage is first transferred to the image domain through domain transformation,and a lightweight neural network LS-AAE is proposed here to achieve high-quality image inpainting based on spatial information between consecutive CT images.The LS-AAE can not only capture residual representations between adjacent images through implicit motion estimation to provide reliable information gain,but also extract feature representations in 3D space through small-scale 3D convolution.Therefore,the LS-AAE can effectively reduce the consumption of computing resources,and its internal cascade structure can help deepen the neural network to enhance its learning ability for complex mappings.So far,the algorithm can achieve high-quality restoration and reconstruction of limited-view CT,not only can effectively remove artifacts and blurring in CT imaging,but also help to clearly restore the texture details in its imaging.In addition,the FBP algorithm is used to replace the SART-TV algorithm in this subject,so that it has better real-time performance as a whole and can be better used in practice.In the experiment,four sorts of limited-view Radon data generated from the LIDC-IDRI dataset are adopted,the algorithm proposed in this subject and various mainstream algorithms are tested on these data for comparison.From the experimental results,it can be seen that the algorithm proposed in this subject has outstanding performance and excellent robustness on severely damaged limited-view data.It is also worth mentioning that for the limited-view CT data missing the last third of information,the algorithm proposed in this subject can increase its PSNR to 39.685 and its SSIM to 0.940,achieving a significant image quality improvement. |
PDF Full Text Request