Objective To evaluate the performance of reconstruction networks with different sparse views under the condition of keeping the same number of model parameters.Methods The number of network channels and network layers were adjusted to make the parameter quantity of each network similar when keeping the structure of each image-domain network and dual-domain network unchanged.The reconstruction performance of each network at different sparsity levels was compared.The AAPM Low-Dose CT Grand Challenge datasets were used in the experiment,including 10 976 images for training,979 images for validation,and 4 256 images for testing.The performance of each model was evaluated visually in combination with objective metrics such as peak signal-to-noise ratio,structural similarity and root mean square error.Results Before adjusting the model parameters,the hybrid domain network Tensor-Net obtained the best visual evaluation and objective evaluation metrics.After parament adjustment,with a similar number of parameters,Tensor-Net outperformed the other models at various projection angles in image anatomical detail recovery,but its structural similarity was slightly lower than that of RED-CNN.The parameters of the hybrid domain model Dual-FBPConvNet were all worse than those of FBPConvNet.Conclusion The hybrid domain model is advantageous in sparse-view CT reconstruction,but it faces more serious overfitting problems.Using a larger image domain model can achieve results similar to those of hybrid domain model.

OBJECTIVES: We propose a sparse-view cone-beam CT reconstruction algorithm based on bidirectional flow field guided projection completion (BBC-Recon) to solve the ill-posed inverse problem in sparse-view cone-beam CT imaging. METHODS: The BBC-Recon method consists of two main modules: the projection completion module and the image restoration module. Based on flow field estimation, the projection completion module, through the designed bidirectional and multi-scale correlators, fully calculates the correlation information and redundant information among projections to precisely guide the generation of bidirectional flow fields and missing frames, thus achieving high-precision completion of missing projections and obtaining pseudo complete projections. The image restoration module reconstructs the obtained pseudo complete projections and then refines the image to remove the residual artifacts and further improve the image quality. RESULTS: The experimental results on the public datasets of Mayo Clinic and Guilin Medical University showed that in the case of a 4-fold sparse angle, compared with the suboptimal method, the BBC-Recon method increased the PSNR index by 1.80% and the SSIM index by 0.29%, and reduced the RMSE index by 4.12%; In the case of an 8-fold sparse angle, the BBC-Recon method increased the PSNR index by 1.43% and the SSIM index by 1.49%, and reduced the RMSE index by 0.77%. CONCLUSIONS The BBC-Recon algorithm fully exploits the correlation information between projections to allow effective removal of streak artifacts while preserving image structure information, and demonstrates significant advantages in maintaining inter-slice consistency.
Algorithms ; Cone-Beam Computed Tomography/methods* ; Image Processing, Computer-Assisted/methods* ; Humans

OBJECTIVES: We propose a segmented backprojection tensor degradation feature encoding (SBP-MAC) model for motion artifact correction in dental cone beam computed tomography (CBCT) to improve the quality of the reconstructed images. METHODS: The proposed motion artifact correction model consists of a generator and a degradation encoder. The segmented limited-angle reconstructed sub-images are stacked into the tensors and used as the model input. A degradation encoder is used to extract spatially varying motion information in the tensor, and the generator's skip connection features are adaptively modulated to guide the model for correcting artifacts caused by different motion waveforms. The artifact consistency loss function was designed to simplify the learning task of the generator. RESULTS: The proposed model could effectively remove motion artifacts and improve the quality of the reconstructed images. For simulated data, the proposed model increased the peak signal-to-noise ratio by 8.28%, increased the structural similarity index measurement by 2.29%, and decreased the root mean square error by 23.84%. For real clinical data, the proposed model achieved the highest expert score of 4.4221 (against a 5-point scale), which was significantly higher than those of all the other comparison methods. CONCLUSIONS The SBP-MAC model can effectively extract spatially varying motion information in the tensors and achieve adaptive artifact correction from the tensor domain to the image domain to improve the quality of reconstructed dental CBCT images.
Cone-Beam Computed Tomography/methods* ; Artifacts ; Humans ; Motion ; Image Processing, Computer-Assisted/methods* ; Signal-To-Noise Ratio ; Algorithms

OBJECTIVES: We propose a low-dose CT image restoration method based on central guidance and alternating optimization (FedGP). METHODS: The FedGP framework revolutionizes the traditional federated learning model by adopting a structure without a fixed central server, where each institution alternatively serves as the central server. This method uses an institution-modulated CT image restoration network as the core of client-side local training. Through a federated learning approach of central guidance and alternating optimization, the central server leverages local labeled data to guide client-side network training to enhance the generalization capability of the CT imaging model across multiple institutions. RESULTS: In the low-dose and sparse-view CT image restoration tasks, the FedGP method showed significant advantages in both visual and quantitative evaluation and achieved the highest PSNR (40.25 and 38.84), the highest SSIM (0.95 and 0.92), and the lowest RMSE (2.39 and 2.56). Ablation study of FedGP demonstrated that compared with FedGP(w/o GP) without central guidance, the FedGP method better adapted to data heterogeneity across institutions, thus ensuring robustness and generalization capability of the model in different imaging conditions. CONCLUSIONS FedGP provides a more flexible FL framework to solve the problem of CT imaging heterogeneity and well adapts to multi-institutional data characteristics to improve generalization ability of the model under diverse imaging geometric configurations.
Tomography, X-Ray Computed/methods* ; Humans ; Radiation Dosage ; Image Processing, Computer-Assisted/methods* ; Algorithms

Objective To investigate the low-dose CT image denoising and generalization performance of the existing mainstream deep learning based denoising networks.Methods The public AAPM Mayo challenge dataset was used to train the denoising network using 3 image-domain methods(REDCNN,WGAN-VGG,CTformer)and 2 projection-image dual-domain methods(VVBP-UNet,CLEAR),separately.The denoising networks were evaluated quantitatively for peak signal-to-noise ratio(PSNR),structural similarity index,root mean square error,number of network parameters and floating point operations,and their generalization performance was analyzed on the AbdomenCT-1K Dataset.Results Image-domain denoising networks effectively suppressed low-dose CT image noise,with REDCNN demonstrating the best denoising performance and achieving a PSNR of 42.0988 dB.The dual-domain denoising networks were better at preserving tiny tissue structures while removing image noise,with VVBP-UNet performing the best and increasing PSNR to 42.150 9 dB.Conclusion The projection-image dual-domain method exhibits superior denoising and generalization performances than the image-domain method,despite requiring a relatively large amount of network parameters and computations.When computing resources are sufficient,the denoising results obtained by dual-domain method better fulfill the requirements for clinical diagnosis.

Objective To investigate the low-dose CT image denoising and generalization performance of the existing mainstream deep learning based denoising networks.Methods The public AAPM Mayo challenge dataset was used to train the denoising network using 3 image-domain methods(REDCNN,WGAN-VGG,CTformer)and 2 projection-image dual-domain methods(VVBP-UNet,CLEAR),separately.The denoising networks were evaluated quantitatively for peak signal-to-noise ratio(PSNR),structural similarity index,root mean square error,number of network parameters and floating point operations,and their generalization performance was analyzed on the AbdomenCT-1K Dataset.Results Image-domain denoising networks effectively suppressed low-dose CT image noise,with REDCNN demonstrating the best denoising performance and achieving a PSNR of 42.0988 dB.The dual-domain denoising networks were better at preserving tiny tissue structures while removing image noise,with VVBP-UNet performing the best and increasing PSNR to 42.150 9 dB.Conclusion The projection-image dual-domain method exhibits superior denoising and generalization performances than the image-domain method,despite requiring a relatively large amount of network parameters and computations.When computing resources are sufficient,the denoising results obtained by dual-domain method better fulfill the requirements for clinical diagnosis.

Objective To evaluate the performance of reconstruction networks with different sparse views under the condition of keeping the same number of model parameters.Methods The number of network channels and network layers were adjusted to make the parameter quantity of each network similar when keeping the structure of each image-domain network and dual-domain network unchanged.The reconstruction performance of each network at different sparsity levels was compared.The AAPM Low-Dose CT Grand Challenge datasets were used in the experiment,including 10 976 images for training,979 images for validation,and 4 256 images for testing.The performance of each model was evaluated visually in combination with objective metrics such as peak signal-to-noise ratio,structural similarity and root mean square error.Results Before adjusting the model parameters,the hybrid domain network Tensor-Net obtained the best visual evaluation and objective evaluation metrics.After parament adjustment,with a similar number of parameters,Tensor-Net outperformed the other models at various projection angles in image anatomical detail recovery,but its structural similarity was slightly lower than that of RED-CNN.The parameters of the hybrid domain model Dual-FBPConvNet were all worse than those of FBPConvNet.Conclusion The hybrid domain model is advantageous in sparse-view CT reconstruction,but it faces more serious overfitting problems.Using a larger image domain model can achieve results similar to those of hybrid domain model.

Objective To propose a low-dose CT reconstruction algorithm across different scanners based on federated feature learning(FedCT)to improve the generalization of deep learning models for multiple CT scanners and protect data privacy.Methods In the proposed FedCT framework,each client is assigned an inverse Radon transform-based reconstruction model to serve as a local network model that participates in federated learning.A projection-domain specific learning strategy is adopted to preserve the geometry specificity in the local projection domain.Federated feature learning is introduced in the model,which utilizes conditional parameters to mark the local data and feed the conditional parameters into the network for encoding to enhance the generalization of the model in the image domain.Results In the cross-client,multi-scanner,and multi-protocol low-dose CT reconstruction experiments,FedCT achieved the highest PSNR(+2.8048,+2.7301,and +2.7263 compared to the second best federated learning method),the highest SSIM(+0.0009,+0.0165,and +0.0131 in the same comparison),and the lowest RMSE(-0.6687,-1.5956,and-0.9962).In the ablation experiment,compared with the general federated learning strategy,the model with projection-specific learning strategy showed an average improvement by 1.18 on Q1 of the PSNR and an average decrease by 1.36 on Q3 of the RMSE on the test set.The introduction of federated feature learning in FedCT further improved the Q1 of the PSNR on the test set by 3.56 and reduced the Q3 of the RMSE by 1.80.Conclusion FedCT provides an effective solution for collaborative construction of CT reconstruction models,which can enhance model generalization and further improve the reconstruction performance on global data while protecting data privacy.

Objective To propose a CT truncated data reconstruction model(DDTrans)based on projection and image dual-domain Transformer coupled feature learning for reducing truncation artifacts and image structure distortion caused by insufficient field of view(FOV)in CT scanning.Methods Transformer was adopted to build projection domain and image domain restoration models,and the long-range dependency modeling capability of the Transformer attention module was used to capture global structural features to restore the projection data information and enhance the reconstructed images.We constructed a differentiable Radon back-projection operator layer between the projection domain and image domain networks to enable end-to-end training of DDTrans.Projection consistency loss was introduced to constrain the image forward-projection results to further improve the accuracy of image reconstruction.Results The experimental results with Mayo simulation data showed that for both partial truncation and interior scanning data,the proposed DDTrans method showed better performance than the comparison algorithms in removing truncation artifacts at the edges and restoring the external information of the FOV.Conclusion The DDTrans method can effectively remove CT truncation artifacts to ensure accurate reconstruction of the data within the FOV and achieve approximate reconstruction of data outside the FOV.

Objective We propose a dual-domain cone beam computed tomography(CBCT)reconstruction framework DualCBR-Net based on improved differentiable domain transform for cone-angle artifact correction.Methods The proposed CBCT dual-domain reconstruction framework DualCBR-Net consists of 3 individual modules:projection preprocessing,differentiable domain transform,and image post-processing.The projection preprocessing module first extends the original projection data in the row direction to ensure full coverage of the scanned object by X-ray.The differentiable domain transform introduces the FDK reconstruction and forward projection operators to complete the forward and gradient backpropagation processes,where the geometric parameters correspond to the extended data dimension to provide crucial prior information in the forward pass of the network and ensure the accuracy in the gradient backpropagation,thus enabling precise learning of cone-beam region data.The image post-processing module further fine-tunes the domain-transformed image to remove residual artifacts and noises.Results The results of validation experiments conducted on Mayo's public chest dataset showed that the proposed DualCBR-Net framework was superior to other comparison methods in terms of artifact removal and structural detail preservation.Compared with the latest methods,the DualCBR-Net framework improved the PSNR and SSIM by 0.6479 and 0.0074,respectively.Conclusion The proposed DualCBR-Net framework for cone-angle artifact correction allows effective joint training of the CBCT dual-domain network and is especially effective for large cone-angle region.