模型与数据双驱动的联合有限角CT重建与金属伪影校正方法

石保顺; 程诗展; 姜轲; 傅昭然

doi:10.11999/JEIT240703

模型与数据双驱动的联合有限角CT重建与金属伪影校正方法

doi: 10.11999/JEIT240703

1.
燕山大学信息科学与工程学院秦皇岛 066004
2.
秦皇岛市第一医院口腔科秦皇岛 066001

基金项目: 国家自然科学基金(62371414)，河北省自然科学基金(F2023203043)，河北省重点实验室资助课题(202250701010046)

详细信息

作者简介:
石保顺：男，副教授，研究方向为医学图像处理、深度字典网络、计算机视觉

程诗展：男，硕士生，研究方向为有限角度CT重建、医学图像处理

姜轲：女，博士生，研究方向为少角度CT图像重建、深度字典网络

傅昭然：女，主任医师，研究方向为金属伪影校正、医学图像处理

通讯作者:
石保顺　shibaoshun@ysu.edu.cn

中图分类号: TN911.73; TP391
计量
- 文章访问数: 48
- HTML全文浏览量: 21
- PDF下载量: 9
- 被引次数: 0
出版历程
- 收稿日期: 2024-08-12
- 修回日期: 2025-04-14
- 网络出版日期: 2025-05-09

Model and Data Dual-driven Joint Limited-Angle CT Reconstruction and Metal Artifact Reduction Method

1.
School of Information Science and Engineering, Yanshan University, Qinhuangdao 066004, China
2.
Department of Stomatology, Qinhuangdao First Hospital, Qinhuangdao 066001, China

Funds: The National Natural Science Foundation of China (62371414), The Natural Science Foundation of Hebei Province (F2023203043), The Key Laboratory of Hebei Province Funding Project (202250701010046)

摘要

摘要: 有限角度计算机断层扫描(LACT)旨在通过减少扫描角度的范围来减少辐射剂量。由于投影数据是不完备的且未考虑联合有限角度和金属伪影校正(LAMAR)任务，传统方法重建的CT图像往往存在伪影，特别是当患者携带金属植入物时，伪影将进一步加重，影响后期医疗诊断及下游任务的精度。为解决这一问题，该文利用双域知识和深度展开技术，融合Transformer的非局部特性捕获能力和卷积神经网络(CNN)的局部特征提取能力，提出了能够联合解决LAMAR和LACT任务的模型与数据双驱动双域重建网络，记为MD³Net。该文首先构建了双域优化模型，使用邻近梯度下降算法对优化模型进行求解，并将其展开成模型驱动的CT重建网络。其次，设计了任务选择(TS)模块，通过判断初始估计CT图像中有无金属以利用同一模型同时处理有金属和无金属的重建任务。在数据驱动网络中，构建了融合Transformer和CNN的双分支的迹感知投影域邻近子网络和结合通道注意力、空间注意力的图像域邻近子网络，进而提升网络表示能力。实验结果表明，与现有方法相比，所提算法在联合LACT和LAMAR任务上重建效果更好。
- 有限角度CT /
- 金属伪影校正 /
- 模型与数据双驱动 /
- 双域网络
Abstract: Objective Computed Tomography (CT) is widely used in medical diagnostics due to its non-destructive, non-contact imaging capabilities. To lower cancer risk from radiation exposure, clinical practice often limits scanning angles, referred to as Limited-Angle CT (LACT). Incomplete projection data in LACT leads to wedge-shaped artifacts in reconstructions using Filtered Back-Projection (FBP) algorithms. These artifacts worsen in the presence of metallic implants. Although LACT reconstruction without metal and full-angle CT Metal Artifact Reduction (MAR) have been extensively studied, the joint task of Limited-Angle and Metal Artifact Reduction (LAMAR) has received limited attention. This study proposes a model- and data-driven CT network that integrates a Task Selection (TS) module to apply appropriate gradient descent steps for different tasks. This enables simultaneous processing of LACT and LAMAR. The network also incorporates dual-domain information interaction during alternating iterations to reconstruct high-quality CT images. Methods First, a dual-domain reconstruction model integrating both model- and data-driven model is constructed to address the joint task of LACT reconstruction and LAMAR. The model comprises four components: an image-domain data fidelity term, a projection-domain data fidelity term, an image-domain regularization term, and a projection-domain regularization term. These terms are solved using an alternating iteration strategy. The image- and projection-domain subproblems are addressed using the proximal gradient descent algorithm, with the iterative process unrolled into a Deep Neural Network (DNN). Each stage of the deep unrolling network includes three components: a TS module, a projection-domain subnetwork, and an image-domain subnetwork. The TS module dynamically determines gradient descent step sizes for the LACT and LAMAR tasks by comparing image-domain FBP reconstruction results with predefined thresholds. The projection-domain subnetwork is shared by both tasks. Finally, the data-driven proximal network comprises the projection-domain and image-domain subnetworks. The projection-domain subnetwork includes an encoder, a dual-branch structure, and a decoder. The encoder has two stages, each consisting of a convolutional layer followed by an activation function; the decoder mirrors this architecture. A Transformer-based long-range branch incorporates non-metal trace information into a self-attention mechanism to guide correction of metal trace data using contextual information from non-metal regions. A short-range branch, composed of six residual blocks, extracts local features. The outputs of the two branches are fused using a weighted strategy before being passed to the decoder. The image-domain subnetwork is implemented as an attention-based U-Net. Channel and spatial attention mechanisms are applied before each of the four downsampling operations in the U-Net encoder. This design allows the decoder to more effectively leverage encoded information for high-quality CT image reconstruction without increasing the number of network parameters. Results and Discussions Experimental results on both LACT reconstruction and LAMAR tasks show that the proposed method outperforms existing CT reconstruction algorithms in both qualitative and quantitative evaluations. Quantitative comparisons (Table 1) indicate that the proposed method achieves higher average Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), and lower Root Mean Square Error (RMSE) for both tasks across three angular ranges. Specifically, average PSNR improvements for the LAMAR and LACT tasks reach 2.78 dB, 2.88 dB, and 2.32 dB, respectively, compared with the best-performing baseline methods. Qualitative comparisons (Fig 4 and Fig 5) show that reconstructing CT images and projection data through alternating iterations, combined with dual-domain information interaction, enables the network to effectively suppress composite artifacts and improve the reconstruction of soft tissue regions and fine structural details. These results consistently exceed those of existing approaches. Visual assessment of reconstruction performance on the clinical dataset for the LAMAR task (Fig 6) further demonstrates the method’s effectiveness in reducing metal artifacts around implants. The reconstructed images exhibit clearer structural boundaries and improved tissue visibility, indicating strong generalization to previously unseen clinical data. Conclusions To address the combined task of LACT reconstruction and LAMAR, this study proposes a dual-domain, model- and data-driven reconstruction framework. The optimization problem is solved using an alternating iteration strategy and unfolded into a model-driven CT reconstruction network, with each subnetwork trained in a data-driven manner. In the projection-domain network, a TS module identifies the presence of metallic implants in the initial CT estimates, allowing a single model to simultaneously handle cases with and without metal. A trace-aware projection-domain proximal subnetwork, integrating Transformer and convolutional neural network architectures, is designed to capture both local and non-local contextual features for restoring metal-corrupted regions. In the image-domain network, a U-Net architecture enhanced with channel and spatial attention mechanisms is used to maximize spatial feature utilization and improve reconstruction quality. Experimental results on the AAPM and DeepLesion datasets confirm that the proposed method consistently outperforms existing algorithms under various limited-angle conditions and in the presence of metal artifacts. Further evaluation on the SpineWeb dataset demonstrates the network’s generalization capability across clinical scenarios.
- Limited-Angle Computed Tomography (LACT) /
- Metal Artifact Reduction (MAR) /
- Model and data dual-driven /
- Dual-domain network

HTML全文

图 1 不同情况下重建图像的对比

下载: 全尺寸图片幻灯片

图 2 MD³Net网络架构图

下载: 全尺寸图片幻灯片

图 3 预处理模块

下载: 全尺寸图片幻灯片

图 4 不同方法在不同有限角度下LACT重建任务的重建结果

下载: 全尺寸图片幻灯片

图 5 不同方法在不同有限角度下LAMAR任务的重建结果

下载: 全尺寸图片幻灯片

图 6 SpineWeb真实数据集LAMAR重建结果的视觉比较

下载: 全尺寸图片幻灯片

表 1 使用PSNR(dB), SSIM和RMSE对3种有限角度设置下的LACT及LAMAR任务进行定量比较

方法	$ [0,{90^ \circ }] $			$ [0,{120^ \circ }] $			$ [0,{150^ \circ }] $
方法	LACT	LAMAR	Average	LACT	LAMAR	Average	LACT	LAMAR	Average
FBP^[2]	17.27 0.5317 0.0688	14.82 0.3739 0.0927	16.05 0.4528 0.0808	19.51 0.5881 0.0532	17.16 0.3448 0.0709	18.34 0.4665 0.0621	22.97 0.6453 0.0358	19.68 0.2939 0.0533	21.33 0.4969 0.0446
FBPConvNet^[16]	25.64 0.8431 0.0268	29.11 0.7877 0.0180	27.38 0.8154 0.0224	27.11 0.8822 0.0231	32.50 0.8575 0.0121	29.81 0.8699 0.0176	32.59 0.9141 0.0126	34.94 0.9224 0.0093	33.77 0.9183 0.0110
DDNet^[17]	26.29 0.7397 0.0247	29.33 0.8688 0.0174	27.81 0.8043 0.0211	29.42 0.8301 0.0175	31.13 0.8745 0.0141	30.28 0.8523 0.0158	34.23 0.8878 0.0101	33.28 0.9131 0.0111	33.76 0.9005 0.0106
EPNet^[10]	27.85 0.8359 0.0211	28.63 0.8368 0.0189	28.24 0.8364 0.0200	29.59 0.8818 0.0172	30.14 0.8635 0.0160	29.87 0.8727 0.0166	34.31 0.9312 0.0105	33.43 0.9156 0.0111	33.87 0.9234 0.0108
DuDoTrans^[8]	26.65 0.8542 0.0236	25.48 0.7717 0.0272	26.07 0.8130 0.0254	29.09 0.8896 0.0181	28.17 0.8492 0.0199	28.63 0.8694 0.0190	32.83 0.9268 0.0119	31.34 0.8866 0.0139	32.09 0.9067 0.0129
FreeSeed^[5]	26.05 0.8606 0.0254	29.55 0.8051 0.0170	27.80 0.8329 0.0212	28.58 0.8861 0.0192	32.03 0.8674 0.0128	30.31 0.8768 0.0160	33.78 0.9284 0.0108	34.88 0.9044 0.0092	34.33 0.9164 0.0100
MD³Net	30.39 0.9142 0.0162	31.45 0.8404 0.0139	30.92 0.8773 0.0151	32.25 0.9309 0.0137	34.12 0.8625 0.0100	33.19 0.8967 0.0119	35.20 0.9507 0.0099	38.09 0.8934 0.0064	36.65 0.9221 0.0082

下载: 导出CSV

表 2 本文方法各个模块3个角度下对联合LACT和LAMAR任务的平均PSNR(dB)/SSIM

方法	X-Net	S-Net	S-Net w/o L	X-Net w/o A	$ [0,{90^ \circ }] $	$ [0,{120^ \circ }] $	$ [0,{150^ \circ }] $	Average
方法	X-Net	S-Net	S-Net w/o L	X-Net w/o A	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM
M1	√	×	×	×	30.78/0.8698	32.83/0.8668	35.56/0.9183	33.09/0.8850
M2	×	√	×	×	26.67/0.6737	27.95/0.6934	30.18/0.7616	28.27/0.7096
M3	√	×	√	×	30.66/0.8721	32.97/0.8927	36.26/0.9120	33.30/0.8923
M4	×	√	×	√	30.83/0.8639	33.11/0.8878	36.49/0.9171	33.48/0.8896
M5	√	√	×	×	30.92/0.8773	33.19/0.8967	36.65/0.9221	33.59/0.8987

下载: 导出CSV

表 3 MD³Net关于阈值h的性能分析

阈值h	LACT	LAMAR	Average
阈值h	PSNR(dB)/ SSIM	PSNR(dB)/ SSIM	PSNR(dB)/ SSIM
1	30.70/0.9133	30.64/0.7999	30.67/0.8566
1.5	30.39/0.9142	31.45/0.8404	30.92/0.8773
2	30.67/0.9124	30.99/0.7894	30.83/0.8509
1.5(可学习的)	30.87/0.9115	30.93/0.8111	30.90/0.8613

下载: 导出CSV

参考文献(33)

[1]	DE BASEA M B, THIERRY-CHEF I, HARBRON R, et al. Risk of hematological malignancies from CT radiation exposure in children, adolescents and young adults[J]. Nature Medicine, 2023, 29(12): 3111–3119. doi: 10.1038/s41591-023-02620-0.
[2]	PAN Xiaochuan, SIDKY E Y, and VANNIER M. Why do commercial CT scanners still employ traditional, filtered back-projection for image reconstruction?[J]. Inverse Problems, 2009, 25(12): 123009. doi: 10.1088/0266-5611/25/12/123009.
[3]	YOO S, YANG Xiaogang, WOLFMAN M, et al. Sinogram image completion for limited angle tomography with generative adversarial networks[C]. 2019 IEEE International Conference on Image Processing (ICIP), Taipei, China, 2019: 1252–1256. doi: 10.1109/ICIP.2019.8804416.
[4]	ZHANG Yanbo and YU Hengyong. Convolutional neural network based metal artifact reduction in x-ray computed tomography[J]. IEEE Transactions on Medical Imaging, 2018, 37(6): 1370–1381. doi: 10.1109/TMI.2018.2823083.
[5]	MA Chenglong, LI Zilong, ZHANG Junping, et al. FreeSeed: Frequency-band-aware and self-guided network for sparse-view CT reconstruction[C]. 26th International Conference on Medical Image Computing and Computer Assisted Intervention, Vancouver, Canada, 2023: 250–259. doi: 10.1007/978-3-031-43999-5_24.
[6]	SHI Baoshun, ZHANG Shaolei, and FU Zhaoran. Artifact region-aware transformer: Global context helps CT metal artifact reduction[J]. IEEE Signal Processing Letters, 2024, 31: 1249–1253. doi: 10.1109/LSP.2024.3393355.
[7]	于佳弘, 张昆鹏, 靳爽, 等. 弦图插值结合UNIT网络图像转换的CT金属伪影校正[J]. 南方医科大学学报, 2023, 43(7): 1214–1223. doi: 10.12122/j.issn.1673-4254.2023.07.18. YU Jiahong, ZHANG Kunpeng, JIN Shuang, et al. Sinogram interpolation combined with unsupervised image-to-image translation network for CT metal artifact correction[J]. Journal of Southern Medical University, 2023, 43(7): 1214–1223. doi: 10.12122/j.issn.1673-4254.2023.07.18.
[8]	WANG Ce, SHANG Kun, ZHANG Haimiao, et al. DuDoTrans: Dual-domain transformer for sparse-view CT reconstruction[C]. 5th International Workshop on Machine Learning for Medical Image Reconstruction, Singapore, 2022: 84–94. doi: 10.1007/978-3-031-17247-2_9.
[9]	LIN Weian, LIAO Haofu, PENG Cheng, et al. DuDoNet: Dual domain network for CT metal artifact reduction[C]. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Long Beach, USA, 2019: 10504–10513. doi: 10.1109/CVPR.2019.01076.
[10]	WANG Ce, ZHANG Haimiao, LI Qian, et al. Improving generalizability in limited-angle CT reconstruction with sinogram extrapolation[C]. 24th International Conference on Medical Image Computing and Computer Assisted Intervention, Strasbourg, France, 2021: 86–96. doi: 10.1007/978-3-030-87231-1_9.
[11]	WAGN Hong, LI Yuexiang, MENG Deyu, et al. Adaptive convolutional dictionary network for CT metal artifact reduction[C]. Thirty-First International Joint Conference on Artificial Intelligence, Vienna, Austria, 2022: 1401–1407. doi: 10.24963/ijcai.2022/195.
[12]	袁伟, 席雅睿, 谭川东, 等. 基于Swin-Transformer迭代展开的有限角CT图像重建用于PTCT成像[J]. 光学学报, 2024, 44(8): 0834001. doi: 10.3788/AOS231823. YUAN Wei, XI Yarui, TAN Chuandong, et al. Limited-angle CT image reconstruction based on Swin-Transformer iterative unfolding for PTCT imaging[J]. Acta Optica Sinica, 2024, 44(8): 0834001. doi: 10.3788/AOS231823.
[13]	赵云松, 李宏伟, 张朋. 有限角CT成像研究进展[J]. 中国体视学与图像分析, 2022, 27(4): 436–446. doi: 10.13505/j.1007-1482.2022.27.04.010. ZHAO Yunsong, LI Hongwei, and ZHANG Peng. Development of limited-angle CT imaging[J]. Chinese Journal of Stereology and Image Analysis, 2022, 27(4): 436–446. doi: 10.13505/j.1007-1482.2022.27.04.010.
[14]	YAZDANPANAH A P and REGENTOVA E E. Sparse-view CT reconstruction using curvelet and TV-based regularization[C]. 13th International Conference on Image Analysis and Recognition, Póvoa de Varzim, Portugal, 2016: 672–677. doi: 10.1007/978-3-319-41501-7_75.
[15]	XU Moran, HU Dianlin, LUO Fulin, et al. Limited-angle X-ray CT reconstruction using image gradient l₀-norm with dictionary learning[J]. IEEE Transactions on Radiation and Plasma Medical Sciences, 2021, 5(1): 78–87. doi: 10.1109/TRPMS.2020.2991887.
[16]	JIN K H, MCCANN M T, FROUSTEY E, et al. Deep convolutional neural network for inverse problems in imaging[J]. IEEE Transactions on Image Processing, 2017, 26(9): 4509–4522. doi: 10.1109/TIP.2017.2713099.
[17]	ZHANG Zhicheng, LIANG Xiaokun, DONG Xu, et al. A sparse-view CT reconstruction method based on combination of DenseNet and deconvolution[J]. IEEE Transactions on Medical Imaging, 2018, 37(6): 1407–1417. doi: 10.1109/TMI.2018.2823338.
[18]	GAO Yuan, ZHANG Liu, YU Wenxue, et al. Cas-PRED: A limited-angle cone beam CT reconstruction research based on prior information[J]. IEEE Transactions on Instrumentation and Measurement, 2024, 73: 4504117. doi: 10.1109/TIM.2023.3335513.
[19]	HU Dianlin, ZHANG Yikun, QUAN Guotao, et al. CROSS: Cross-domain residual-optimization-based structure strengthening reconstruction for limited-angle CT[J]. IEEE Transactions on Radiation and Plasma Medical Sciences, 2023, 7(5): 521–531. doi: 10.1109/TRPMS.2023.3242662.
[20]	CHENG Weiting, HE Jichun, LIU Yi, et al. CAIR: Combining integrated attention with iterative optimization learning for sparse-view CT reconstruction[J]. Computers in Biology and Medicine, 2023, 163: 107161. doi: 10.1016/j.compbiomed.2023.107161.
[21]	VERBURG J M and SECO J. CT metal artifact reduction method correcting for beam hardening and missing projections[J]. Physics in Medicine & Biology, 2012, 57(9): 2803–2818. doi: 10.1088/0031-9155/57/9/2803.
[22]	KALENDER W A, HEBEL R, and EBERSBERGER J. Reduction of CT artifacts caused by metallic implants[J]. Radiology, 1987, 164(2): 576–577. doi: 10.1148/radiology.164.2.3602406.
[23]	MEYER E, RAUPACH R, LELL M, et al. Normalized metal artifact reduction (NMAR) in computed tomography[J]. Medical Physics, 2010, 37(10): 5482–5493. doi: 10.1118/1.3484090.
[24]	ZHANG Haimiao, DONG Bin, and LIU Baodong. A reweighted joint spatial-radon domain CT image reconstruction model for metal artifact reduction[J]. SIAM Journal on Imaging Sciences, 2018, 11(1): 707–733. doi: 10.1137/17M1140212.
[25]	WANG Hong, LI Yuexiang, ZHANG Haimiao, et al. InDuDoNet: An interpretable dual domain network for CT metal artifact reduction[C]. 24th International Conference on Medical Image Computing and Computer Assisted Intervention, Strasbourg, France, 2021: 107–118. doi: 10.1007/978-3-030-87231-1_11.
[26]	WANG Hong, LI Yuexiang, ZHANG Haimiao, et al. InDuDoNet+: A deep unfolding dual domain network for metal artifact reduction in CT images[J]. Medical Image Analysis, 2023, 85: 102729. doi: 10.1016/j.media.2022.102729.
[27]	SHI Baoshun, JIANG Ke, ZHANG Shaolei, et al. Mud-Net: Multi-domain deep unrolling network for simultaneous sparse-view and metal artifact reduction in computed tomography[J]. Machine Learning: Science and Technology, 2024, 5(1): 015010. doi: 10.1088/2632-2153/ad1b8e.
[28]	SHI Baoshun, ZHANG Shaolei, JIANG Ke, et al. Coupling model-and data-driven networks for CT metal artifact reduction[J]. IEEE Transactions on Computational Imaging, 2024, 10: 415–428. doi: 10.1109/TCI.2024.3369408.
[29]	RONNEBERGER O, FISCHER P, and BROX T. U-Net: Convolutional networks for biomedical image segmentation[C]. 18th International Conference on Medical Image Computing and Computer-Assisted Intervention, Munich, Germany, 2015: 234–241. doi: 10.1007/978-3-319-24574-4_28.
[30]	LIU Xiangbin, LIU Ying, FU Weina, et al. RETRACTED ARTICLE: SCTV-UNet: A COVID-19 CT segmentation network based on attention mechanism[J]. Soft Computing, 2024, 28(2): 473. doi: 10.1007/s00500-023-07991-7.
[31]	MCCOLLOUGH C. TU-FG-207A-04: Overview of the low dose CT grand challenge[J]. Medical Physics, 2016, 43(6Part35): 3759–3760. doi: 10.1118/1.4957556.
[32]	YAN Ke, WANG Xiaosong, LU Le, et al. Deep lesion graphs in the wild: Relationship learning and organization of significant radiology image findings in a diverse large-scale lesion database[C]. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 9261–9270. doi: 10.1109/CVPR.2018.00965.
[33]	GLOCKER B, ZIKIC D, KONUKOGLU E, et al. Vertebrae localization in pathological spine CT via dense classification from sparse annotations[C]. 16th International Conference on Medical Image Computing and Computer-Assisted Intervention, Nagoya, Japan, 2013: 262–270. doi: 10.1007/978-3-642-40763-5_33.