Radiation dose reduction in computed tomography(CT)can be achieved by decreasing the number of projections.However,reconstructing CT images via filtered back projection algorithm from sparse-view projections often con...Radiation dose reduction in computed tomography(CT)can be achieved by decreasing the number of projections.However,reconstructing CT images via filtered back projection algorithm from sparse-view projections often contains severe streak artifacts,affecting clinical diagnosis.To address this issue,this paper proposes TransitNet,an iterative unrolling deep neural network that combines model-driven data consistency,a physical a prior constraint,with deep learning’s feature extraction capabilities.TransitNet employs a novel iterative architecture,implementing flexible physical constraints through learnable data consistency operations,utilizing Transformer’s self-attention mechanism to model long-range dependencies in image features,and introducing linear attention mechanisms to reduce self-attention’s computational complexity from quadratic to linear.Extensive experiments demonstrate that this method exhibits significant advantages in both reconstruction quality and computational efficiency,effectively suppressing streak artifacts while preserving structures and details of images.展开更多
Grating-based X-ray phase-contrast imaging enhances the contrast of imaged objects,particularly soft tissues.However,the radiation dose in computed tomography(CT)is generally excessive owing to the complex collection ...Grating-based X-ray phase-contrast imaging enhances the contrast of imaged objects,particularly soft tissues.However,the radiation dose in computed tomography(CT)is generally excessive owing to the complex collection scheme.Sparse-view CT collection reduces the radiation dose,but with reduced resolution and reconstructed artifacts particularly in analytical reconstruction methods.Recently,deep learning has been employed in sparse-view CT reconstruction and achieved stateof-the-art results.Nevertheless,its low generalization performance and requirement for abundant training datasets have hindered the practical application of deep learning in phase-contrast CT.In this study,a CT model was used to generate a substantial number of simulated training datasets,thereby circumventing the need for experimental datasets.By training a network with simulated training datasets,the proposed method achieves high generalization performance in attenuationbased CT and phase-contrast CT,despite the lack of sufficient experimental datasets.In experiments utilizing only half of the CT data,our proposed method obtained an image quality comparable to that of the filtered back-projection algorithm with full-view projection.The proposed method simultaneously addresses two challenges in phase-contrast three-dimensional imaging,namely the lack of experimental datasets and the high exposure dose,through model-driven deep learning.This method significantly accelerates the practical application of phase-contrast CT.展开更多
As a complement to X-ray computed tomography(CT),neutron tomography has been extensively used in nuclear engineer-ing,materials science,cultural heritage,and industrial applications.Reconstruction of the attenuation m...As a complement to X-ray computed tomography(CT),neutron tomography has been extensively used in nuclear engineer-ing,materials science,cultural heritage,and industrial applications.Reconstruction of the attenuation matrix for neutron tomography with a traditional analytical algorithm requires hundreds of projection views in the range of 0°to 180°and typically takes several hours to complete.Such a low time-resolved resolution degrades the quality of neutron imaging.Decreasing the number of projection acquisitions is an important approach to improve the time resolution of images;however,this requires efficient reconstruction algorithms.Therefore,sparse-view reconstruction algorithms in neutron tomography need to be investigated.In this study,we investigated the three-dimensional reconstruction algorithm for sparse-view neu-tron CT scans.To enhance the reconstructed image quality of neutron CT,we propose an algorithm that uses OS-SART to reconstruct images and a split Bregman to solve for the total variation(SBTV).A comparative analysis of the performances of each reconstruction algorithm was performed using simulated and actual experimental data.According to the analyzed results,OS-SART-SBTV is superior to the other algorithms in terms of denoising,suppressing artifacts,and preserving detailed structural information of images.展开更多
Neutron computed tomography(NCT)is widely used as a noninvasive measurement technique in nuclear engineering,thermal hydraulics,and cultural heritage.The neutron source intensity of NCT is usually low and the scan tim...Neutron computed tomography(NCT)is widely used as a noninvasive measurement technique in nuclear engineering,thermal hydraulics,and cultural heritage.The neutron source intensity of NCT is usually low and the scan time is long,resulting in a projection image containing severe noise.To reduce the scanning time and increase the image reconstruction quality,an effective reconstruction algorithm must be selected.In CT image reconstruction,the reconstruction algorithms can be divided into three categories:analytical algorithms,iterative algorithms,and deep learning.Because the analytical algorithm requires complete projection data,it is not suitable for reconstruction in harsh environments,such as strong radia-tion,high temperature,and high pressure.Deep learning requires large amounts of data and complex models,which cannot be easily deployed,as well as has a high computational complexity and poor interpretability.Therefore,this paper proposes the OS-SART-PDTV iterative algorithm,which uses the ordered subset simultaneous algebraic reconstruction technique(OS-SART)algorithm to reconstruct the image and the first-order primal–dual algorithm to solve the total variation(PDTV),for sparse-view NCT three-dimensional reconstruction.The novel algorithm was compared with other algorithms(FBP,OS-SART-TV,OS-SART-AwTV,and OS-SART-FGPTV)by simulating the experimental data and actual neutron projection experiments.The reconstruction results demonstrate that the proposed algorithm outperforms the FBP,OS-SART-TV,OS-SART-AwTV,and OS-SART-FGPTV algorithms in terms of preserving edge structure,denoising,and suppressing artifacts.展开更多
The most critical part of a neutron computed tomography(NCT) system is the image processing algorithm,which directly affects the quality and speed of the reconstructed images.Various types of noise in the system can d...The most critical part of a neutron computed tomography(NCT) system is the image processing algorithm,which directly affects the quality and speed of the reconstructed images.Various types of noise in the system can degrade the quality of the reconstructed images.Therefore,to improve the quality of the reconstructed images of NCT systems,efficient image processing algorithms must be used.The anisotropic diffusion filtering(ADF) algorithm can not only effectively suppress the noise in the projection data,but also preserve the image edge structure information by reducing the diffusion at the image edges.Therefore,we propose the application of the ADF algorithm for NCT image reconstruction.To compare the performance of different algorithms in NCT systems,we reconstructed images using the ordered subset simultaneous algebraic reconstruction technique(OS-SART) algorithm with different regular terms as image processing algorithms.In the iterative reconstruction,we selected two image processing algorithms,the Total Variation and split Bregman solved total variation algorithms,for comparison with the performance of the ADF algorithm.Additionally,the filtered back-projection algorithm was used for comparison with an iterative algorithm.By reconstructing the projection data of the numerical and clock models,we compared and analyzed the effects of each algorithm applied in the NCT system.Based on the reconstruction results,OS-SART-ADF outperformed the other algorithms in terms of denoising,preserving the edge structure,and suppressing artifacts.For example,when the 3D Shepp–Logan was reconstructed at 25 views,the root mean square error of OS-SART-ADF was the smallest among the four iterative algorithms,at only 0.0292.The universal quality index,mean structural similarity,and correlation coefficient of the reconstructed image were the largest among all algorithms,with values of 0.9877,0.9878,and 0.9887,respectively.展开更多
Inspired by total variation(TV), this paper represents a new iterative algorithm based on diagonal total variation(DTV) to address the computed tomography image reconstruction problem. To improve the quality of a reco...Inspired by total variation(TV), this paper represents a new iterative algorithm based on diagonal total variation(DTV) to address the computed tomography image reconstruction problem. To improve the quality of a reconstructed image, we used DTV to sparsely represent images when iterative convergence of the reconstructed algorithm with TV-constraint had no effect during the reconstruction process. To investigate our proposed algorithm, the numerical and experimental studies were performed, and rootmean-square error(RMSE) and structure similarity(SSIM)were used to evaluate the reconstructed image quality. The results demonstrated that the proposed method could effectively reduce noise, suppress artifacts, and reconstruct highquality image from incomplete projection data.展开更多
Reconstructing from multi-view images is a longstanding problem in 3D vision,where neural radiance fields(NeRFs)have shown great potential and get realistic rendered images of novel views.Currently,most NeRF methods e...Reconstructing from multi-view images is a longstanding problem in 3D vision,where neural radiance fields(NeRFs)have shown great potential and get realistic rendered images of novel views.Currently,most NeRF methods either require accurate camera poses or a large number of input images,or even both.Reconstructing NeRF from few-view images without poses is challenging and highly ill-posed.To address this problem,we propose CAD-NeRF,a method reconstructed from less than 10 images without any known poses.Specifically,we build a mini library of several CAD models from ShapeNet and render them from many random views.Given sparse-view input images,we run a model and pose retrieval from the library,to get a model with similar shapes,serving as the density supervision and pose initializations.Here we propose a multi-view pose retrieval method to avoid pose conflicts among views,which is a new and unseen problem in uncalibrated NeRF methods.Then,the geometry of the object is trained by the CAD guidance.The deformation of the density field and camera poses are optimized jointly.Then texture and density are trained and fine-tuned as well.All training phases are in self-supervised manners.Comprehensive evaluations of synthetic and real images show that CAD-NeRF successfully learns accurate densities with a large deformation from retrieved CAD models,showing the generalization abilities.展开更多
基金National Natural Science Foundation of China under grant (62071281)Local Science and Technology Development Fund Project Guided by the Central Government under grant (YDZJSX2021A003)。
文摘Radiation dose reduction in computed tomography(CT)can be achieved by decreasing the number of projections.However,reconstructing CT images via filtered back projection algorithm from sparse-view projections often contains severe streak artifacts,affecting clinical diagnosis.To address this issue,this paper proposes TransitNet,an iterative unrolling deep neural network that combines model-driven data consistency,a physical a prior constraint,with deep learning’s feature extraction capabilities.TransitNet employs a novel iterative architecture,implementing flexible physical constraints through learnable data consistency operations,utilizing Transformer’s self-attention mechanism to model long-range dependencies in image features,and introducing linear attention mechanisms to reduce self-attention’s computational complexity from quadratic to linear.Extensive experiments demonstrate that this method exhibits significant advantages in both reconstruction quality and computational efficiency,effectively suppressing streak artifacts while preserving structures and details of images.
基金supported by the National Natural Science Foundation of China(Nos.U2032148,U2032157,11775224)USTC Research Funds of the Double First-Class Initiative(No.YD2310002008)the National Key Research and Development Program of China(No.2017YFA0402904),the Youth Innovation Promotion Association,CAS(No.2020457)。
文摘Grating-based X-ray phase-contrast imaging enhances the contrast of imaged objects,particularly soft tissues.However,the radiation dose in computed tomography(CT)is generally excessive owing to the complex collection scheme.Sparse-view CT collection reduces the radiation dose,but with reduced resolution and reconstructed artifacts particularly in analytical reconstruction methods.Recently,deep learning has been employed in sparse-view CT reconstruction and achieved stateof-the-art results.Nevertheless,its low generalization performance and requirement for abundant training datasets have hindered the practical application of deep learning in phase-contrast CT.In this study,a CT model was used to generate a substantial number of simulated training datasets,thereby circumventing the need for experimental datasets.By training a network with simulated training datasets,the proposed method achieves high generalization performance in attenuationbased CT and phase-contrast CT,despite the lack of sufficient experimental datasets.In experiments utilizing only half of the CT data,our proposed method obtained an image quality comparable to that of the filtered back-projection algorithm with full-view projection.The proposed method simultaneously addresses two challenges in phase-contrast three-dimensional imaging,namely the lack of experimental datasets and the high exposure dose,through model-driven deep learning.This method significantly accelerates the practical application of phase-contrast CT.
基金supported by the National Key Research and Development Program of China(No.2022YFB1902700)the National Natural Science Foundation of China(No.11875129)+3 种基金the Fund of the State Key Laboratory of Intense Pulsed Radiation Simulation and Effect(No.SKLIPR1810)the Fund of Innovation Center of Radiation Application(No.KFZC2020020402)the Fund of the State Key Laboratory of Nuclear Physics and Technology,Peking University(No.NPT2020KFY08)the Joint Innovation Fund of China National Uranium Co.,Ltd.,State Key Laboratory of Nuclear Resources and Environment,East China University of Technology(No.2022NRE-LH-02).
文摘As a complement to X-ray computed tomography(CT),neutron tomography has been extensively used in nuclear engineer-ing,materials science,cultural heritage,and industrial applications.Reconstruction of the attenuation matrix for neutron tomography with a traditional analytical algorithm requires hundreds of projection views in the range of 0°to 180°and typically takes several hours to complete.Such a low time-resolved resolution degrades the quality of neutron imaging.Decreasing the number of projection acquisitions is an important approach to improve the time resolution of images;however,this requires efficient reconstruction algorithms.Therefore,sparse-view reconstruction algorithms in neutron tomography need to be investigated.In this study,we investigated the three-dimensional reconstruction algorithm for sparse-view neu-tron CT scans.To enhance the reconstructed image quality of neutron CT,we propose an algorithm that uses OS-SART to reconstruct images and a split Bregman to solve for the total variation(SBTV).A comparative analysis of the performances of each reconstruction algorithm was performed using simulated and actual experimental data.According to the analyzed results,OS-SART-SBTV is superior to the other algorithms in terms of denoising,suppressing artifacts,and preserving detailed structural information of images.
基金supported by the National Key Research and Development Program of China(No.2022YFB1902700)the Joint Fund of Ministry of Education for Equipment Pre-research(No.8091B042203)+5 种基金the National Natural Science Foundation of China(No.11875129)the Fund of the State Key Laboratory of Intense Pulsed Radiation Simulation and Effect(No.SKLIPR1810)the Fund of Innovation Center of Radiation Application(No.KFZC2020020402)the Fund of the State Key Laboratory of Nuclear Physics and Technology,Peking University(No.NPT2023KFY06)the Joint Innovation Fund of China National Uranium Co.,Ltd.,State Key Laboratory of Nuclear Resources and Environment,East China University of Technology(No.2022NRE-LH-02)the Fundamental Research Funds for the Central Universities(No.2023JG001).
文摘Neutron computed tomography(NCT)is widely used as a noninvasive measurement technique in nuclear engineering,thermal hydraulics,and cultural heritage.The neutron source intensity of NCT is usually low and the scan time is long,resulting in a projection image containing severe noise.To reduce the scanning time and increase the image reconstruction quality,an effective reconstruction algorithm must be selected.In CT image reconstruction,the reconstruction algorithms can be divided into three categories:analytical algorithms,iterative algorithms,and deep learning.Because the analytical algorithm requires complete projection data,it is not suitable for reconstruction in harsh environments,such as strong radia-tion,high temperature,and high pressure.Deep learning requires large amounts of data and complex models,which cannot be easily deployed,as well as has a high computational complexity and poor interpretability.Therefore,this paper proposes the OS-SART-PDTV iterative algorithm,which uses the ordered subset simultaneous algebraic reconstruction technique(OS-SART)algorithm to reconstruct the image and the first-order primal–dual algorithm to solve the total variation(PDTV),for sparse-view NCT three-dimensional reconstruction.The novel algorithm was compared with other algorithms(FBP,OS-SART-TV,OS-SART-AwTV,and OS-SART-FGPTV)by simulating the experimental data and actual neutron projection experiments.The reconstruction results demonstrate that the proposed algorithm outperforms the FBP,OS-SART-TV,OS-SART-AwTV,and OS-SART-FGPTV algorithms in terms of preserving edge structure,denoising,and suppressing artifacts.
基金supported by the National Key Research and Development Program of China (No. 2022YFB1902700)the National Natural Science Foundation of China (No. 11875129)+3 种基金the Fund of the State Key Laboratory of Intense Pulsed Radiation Simulation and Effect (No. SKLIPR1810)Fund of Innovation Center of Radiation Application (No. KFZC2020020402)Fund of the State Key Laboratory of Nuclear Physics and Technology,Peking University (No. NPT2020KFY08)the Joint Innovation Fund of China National Uranium Co.,Ltd.,State Key Laboratory of Nuclear Resources and Environment,East China University of Technology (No. 2022NRE-LH-02)。
文摘The most critical part of a neutron computed tomography(NCT) system is the image processing algorithm,which directly affects the quality and speed of the reconstructed images.Various types of noise in the system can degrade the quality of the reconstructed images.Therefore,to improve the quality of the reconstructed images of NCT systems,efficient image processing algorithms must be used.The anisotropic diffusion filtering(ADF) algorithm can not only effectively suppress the noise in the projection data,but also preserve the image edge structure information by reducing the diffusion at the image edges.Therefore,we propose the application of the ADF algorithm for NCT image reconstruction.To compare the performance of different algorithms in NCT systems,we reconstructed images using the ordered subset simultaneous algebraic reconstruction technique(OS-SART) algorithm with different regular terms as image processing algorithms.In the iterative reconstruction,we selected two image processing algorithms,the Total Variation and split Bregman solved total variation algorithms,for comparison with the performance of the ADF algorithm.Additionally,the filtered back-projection algorithm was used for comparison with an iterative algorithm.By reconstructing the projection data of the numerical and clock models,we compared and analyzed the effects of each algorithm applied in the NCT system.Based on the reconstruction results,OS-SART-ADF outperformed the other algorithms in terms of denoising,preserving the edge structure,and suppressing artifacts.For example,when the 3D Shepp–Logan was reconstructed at 25 views,the root mean square error of OS-SART-ADF was the smallest among the four iterative algorithms,at only 0.0292.The universal quality index,mean structural similarity,and correlation coefficient of the reconstructed image were the largest among all algorithms,with values of 0.9877,0.9878,and 0.9887,respectively.
基金supported in part by the National Natural Science Foundation of China(No.61401049)the Chongqing Foundation and Frontier Research Project(Nos.cstc2016jcyjA0473,cstc2013jcyjA0763)+3 种基金the Graduate Scientific Research and Innovation Foundation of Chongqing,China(No.CYB16044)the Strategic Industry Key Generic Technology Innovation Project of Chongqing(No.cstc2015zdcy-ztzxX0002)China Scholarship Councilthe Fundamental Research Funds for the Central Universities Nos.CDJZR14125501,106112016CDJXY120003,10611CDJXZ238826
文摘Inspired by total variation(TV), this paper represents a new iterative algorithm based on diagonal total variation(DTV) to address the computed tomography image reconstruction problem. To improve the quality of a reconstructed image, we used DTV to sparsely represent images when iterative convergence of the reconstructed algorithm with TV-constraint had no effect during the reconstruction process. To investigate our proposed algorithm, the numerical and experimental studies were performed, and rootmean-square error(RMSE) and structure similarity(SSIM)were used to evaluate the reconstructed image quality. The results demonstrated that the proposed method could effectively reduce noise, suppress artifacts, and reconstruct highquality image from incomplete projection data.
基金supported in part by the National Natural Science Foundation of China (NSFC) (Grant Nos. 62325221, 62132021, and 62372457), the Young Elite Scientists Sponsorship Program by CAST, China (Grant No. 2023QNRC001)the Natural Science Foundation of Hunan Province, China (Grant Nos. 2021RC3071 and 2022RC1104)the National University of Defense Technology Research Grants, China (Grant No. ZK22-52).
文摘Reconstructing from multi-view images is a longstanding problem in 3D vision,where neural radiance fields(NeRFs)have shown great potential and get realistic rendered images of novel views.Currently,most NeRF methods either require accurate camera poses or a large number of input images,or even both.Reconstructing NeRF from few-view images without poses is challenging and highly ill-posed.To address this problem,we propose CAD-NeRF,a method reconstructed from less than 10 images without any known poses.Specifically,we build a mini library of several CAD models from ShapeNet and render them from many random views.Given sparse-view input images,we run a model and pose retrieval from the library,to get a model with similar shapes,serving as the density supervision and pose initializations.Here we propose a multi-view pose retrieval method to avoid pose conflicts among views,which is a new and unseen problem in uncalibrated NeRF methods.Then,the geometry of the object is trained by the CAD guidance.The deformation of the density field and camera poses are optimized jointly.Then texture and density are trained and fine-tuned as well.All training phases are in self-supervised manners.Comprehensive evaluations of synthetic and real images show that CAD-NeRF successfully learns accurate densities with a large deformation from retrieved CAD models,showing the generalization abilities.
