Sparse CT artifact suppresion Transformer network based on multi-scale attention adaptive fusion

doi:10.11772/j.issn.1001-9081.2025070817

Journal of Computer Applications ›› 2026, Vol. 46 ›› Issue (8): 2660-2667.DOI: 10.11772/j.issn.1001-9081.2025070817

• Multimedia computing and computer simulation • Previous Articles

Sparse CT artifact suppresion Transformer network based on multi-scale attention adaptive fusion

Jielong BAI, Chenyun FANG, Zhiwei QIAO()

School of Computer and Information Technology，Shanxi University，Taiyuan Shanxi 030006，China

Received:2025-07-22 Revised:2025-09-25 Accepted:2025-09-25 Online:2025-11-05 Published:2026-08-10
Contact: Zhiwei QIAO
About author:BAI Jielong， born in 2000， M. S. candidate. His research interests include medical image reconstruction， image processing.
FANG Chenyun， born in 1997， Ph. D. candidate. Her research interests include medical image reconstruction， image processing.
Supported by:
National Natural Science Foundation of China(62071281);Local Science and Technology Development Fund Project Guided by the Central Government(YDZJSX2021A003)

基于多尺度注意力自适应融合的稀疏CT伪影抑制Transformer网络

白杰龙, 方晨韵, 乔志伟()

山西大学计算机与信息技术学院，太原 030006

通讯作者: 乔志伟
作者简介:白杰龙（2000—），男，山西吕梁人，硕士研究生，主要研究方向：医学图像重建、图像处理
方晨韵（1997—），女，安徽安庆人，博士研究生，主要研究方向：医学图像重建、图像处理
乔志伟（1977—），男，山西临汾人，教授，博士，主要研究方向：电子顺磁共振成像、图像重建算法、大规模最优化。
基金资助:
国家自然科学基金资助项目(62071281);中央引导地方科技发展资金项目(YDZJSX2021A003)

Abstract

Abstract:

Sparse Computed Tomography （CT） reconstruction can reduce patient radiation dosage and is of great significance for clinical diagnosis. In deep learning-based image reconstruction tasks， classic networks such as Uformer （U-shape Transformer）， Restormer （Restoration Transformer）， and AST （Adaptive Sparse Transformer） fail to consider multi-scale and directional information in images， neglecting the balance between local details and global structure， resulting in limited artifact suppression. To address this issue， a Multi-scale Attention Adaptive Fusion Transformer （MAAF-Transformer） network was proposed. In this network， a parallel attention fusion strategy was adopted， a Multi-Scale Channel Direction-aware Attention （MSCDA） module was combined to capture artifact features， and a Convolutional Adaptive Spatial Channel Attention （CASCA） module was used to adjust the weights dynamically. Then， effective information was filtered through a Gated Deep-convolutional Feedforward Network （GDFN）. Experimental results show that at 60 sparse angles， MAAF-Transformer achieves 0.714 1 dB higher Peak Signal-to-Noise Ratio （PSNR）， 0.33% higher Structural SIMilarity （SSIM）， and 7.76% lower Root Mean Square Error （RMSE） compared with classic Uformer， with excellent performance in terms of visual effects as well. It can be seen that MAAF-Transformer has higher sparse reconstruction accuracy and stronger artifact suppression capability.

Key words: sparse reconstruction, Computed Tomography (CT), multi-scale, attention mechanism, adaptive fusion

摘要：

稀疏计算机断层成像（CT）重建能够降低患者的辐射剂量，对临床诊断具有重要意义。在基于深度学习的图像重建任务中，经典的Uformer （U-shape Transformer）、Restormer （Restoration Transformer）和AST （Adaptive Sparse Transformer）等网络未考虑图像的多尺度与方向信息，从而忽略了局部细节和全局结构的平衡，伪影抑制效果有限。针对上述问题，提出一种多尺度注意力自适应融合Transformer（MAAF-Transformer）网络。该网络采用并行注意力融合策略，结合多尺度通道方向感知注意力（MSCDA）模块捕获伪影特征，并利用卷积自适应空间通道注意力（CASCA）模块动态调节权重，然后通过门控深度卷积前馈网络（GDFN）筛选有效信息。实验结果表明，在60个稀疏角度下，MAAF-Transformer相较于经典Uformer在峰值信噪比（PSNR）上提高了0.714 1 dB、结构相似性（SSIM）上提高了0.33%，而均方根误差（RMSE）上降低了7.76%，并在视觉效果上同样表现优异。可见，MAAF-Transformer的稀疏重建精度更高，抑制伪影能力更强。

关键词: 稀疏重建, 计算机断层成像, 多尺度, 注意力机制, 自适应融合

CLC Number:

TP391.41

Jielong BAI, Chenyun FANG, Zhiwei QIAO. Sparse CT artifact suppresion Transformer network based on multi-scale attention adaptive fusion[J]. Journal of Computer Applications, 2026, 46(8): 2660-2667.

白杰龙, 方晨韵, 乔志伟. 基于多尺度注意力自适应融合的稀疏CT伪影抑制Transformer网络[J]. 《计算机应用》唯一官方网站, 2026, 46(8): 2660-2667.

Figures/Tables 12

References 40

[1]	Rybicki F J， Otero H J， Steigner M L， et al. Initial evaluation of coronary images from 320-detector row computed tomography［J］. The International Journal of Cardiovascular Imaging， 2008， 24（5）： 535-546.
[2]	Dilmanian F A， Zhong Z， Ren B， et al. Computed tomography of X-ray index of refraction using the diffraction enhanced imaging method［J］. Physics in Medicine and Biology， 2000， 45（4）： No.933.
[3]	Wang D， Fan F， Wu Z， et al. CTformer： convolution-free Token2Token dilated Vision Transformer for low-dose CT denoising［J］. Physics in Medicine and Biology， 2023， 68（6）： No.065012.
[4]	Kramer B S， Berg C D， Aberle D R， et al. Lung cancer screening with low-dose helical CT： results from the National Lung Screening Trial （NLST）［J］. Journal of Medical Screening， 2011， 18（3）： 109-111.
[5]	Mostafapour S， Greuter M， van Snick J H， et al. Ultra-low dose CT scanning for PET/CT［J］. Medical Physics， 2024， 51（1）： 139-155.
[6]	Han Y. Hierarchical decomposed dual-domain deep learning for sparse-view CT reconstruction［J］. Physics in Medicine and Biology， 2024， 69（8）： No.085019.
[7]	Sun C， Liu Y， Yang H. Degradation-aware deep learning framework for sparse-view CT reconstruction［J］. Tomography， 2021， 7（4）： 932-949.
[8]	刘宇，张鹏程，张丽媛，等. 基于低秩与全变分联合正则化的低剂量CT图像重建［J］. 计算机应用， 2025， 45（3）： 978-987.
	Liu Yu， Zhang Pengcheng， Zhang Liyuan， et al. Low-dose CT image reconstruction based on low-rank and total variation joint regularization［J］. Journal of Computer Applications， 2025， 45（3）： 978-987.
[9]	马靓怡，乔志伟. 基于Chambolle-Pock框架的核TV多通道图像重建算法［J］. CT理论与应用研究， 2022， 31（6）： 731-747.
	Ma Jingyi， Qiao Zhiwei. Nuclear TV multi-channel image reconstruction algorithms based on Chambolle-Pock framework［J］. CT Theory and Applications， 2022， 31（6）： 731-747.
[10]	陈蒙蒙，乔志伟. 基于融合通道注意力的Uformer的CT图像稀疏重建［J］. 计算机应用， 2023， 43（9）： 2948-2954.
	Chen Mengmeng， Qiao Zhiwei. Sparse reconstruction of CT images based on Uformer with fused channel attention［J］. Journal of Computer Applications， 2023， 43（9）： 2948-2954.
[11]	Pan X， Sidky E Y， Vannier M. Why do commercial CT scanners still employ traditional， filtered back-projection for image reconstruction？［J］. Inverse Problems， 2009， 25（12）： No.123009.
[12]	Liu Y， Liang Z， Ma J， et al. Total variation-stokes strategy for sparse-view X-ray CT image reconstruction［J］. IEEE Transactions on Medical Imaging， 2014， 33（3）： 749-763.
[13]	Wang Z， Cun X， Bao J， et al. Uformer： a general U-shaped Transformer for image restoration［C］// CVPR 2022. Piscataway： IEEE， 2022： 17662-17672.
[14]	Zamir S W， Arora A， Khan S， et al. Restormer： efficient transformer for high-resolution image restoration［C］// CVPR 2022. Piscataway： IEEE， 2022： 5718-5729.
[15]	Zhou S， Chen D， Pan J， et al. Adapt or perish： adaptive sparse Transformer with attentive feature refinement for image restoration［C］// CVPR 2024. Piscataway： IEEE， 2024： 2952-2963.
[16]	Niu S， Gao Y， Bian Z， et al. Sparse-view X-Ray CT reconstruction via total generalized variation regularization［J］. Physics in Medicine and Biology， 2014， 59（12）： No.2997.
[17]	Han Y S， Jin K H， Kim K， et al. Sparse-view X-Ray spectral CT reconstruction using annihilating filter-based low rank hankel matrix approach［C］// ISBI 2016. Piscataway： IEEE， 2016： 573-576.
[18]	Zhang H， Ma J， Wang J， et al. Adaptive nonlocal means-regularized iterative image reconstruction for sparse-view CT［C］// PNSS/MIC 2014. Piscataway： IEEE， 2014： 1-7.
[19]	Lee H， Lee J， Kim H， et al. Deep-neural-network-based sinogram synthesis for sparse-view CT image reconstruction［J］. IEEE Transactions on Radiation and Plasma Medical Sciences， 2019， 3（2）： 109-119.
[20]	刘安阳，赵怀慈，蔡文龙，等. 基于主动判别机制的自适应生成对抗网络图像去模糊算法［J］. 计算机应用， 2023， 43（7）： 2288-2294.
	Liu Anyang， Zhao Huaici， Cai Wenlong， et al. Adaptive image deblurring generative adversarial network algorithm based on active discrimination mechanism［J］. Journal of Computer Applications， 2023， 43（7）： 2288-2294.
[21]	Sun C， Salimi Y， Angeliki N， et al. An efficient dual-domain deep learning network for sparse-view CT reconstruction［J］. Computer Methods and Programs in Biomedicine， 2024， 256： No.108376.
[22]	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.
[23]	Chen H， Zhang Y， Kalra M K， et al. Low-dose CT with a residual encoder-decoder convolutional neural network［J］. IEEE Transactions on Medical Imaging， 2017， 36（12）： 2524-2535.
[24]	Jiang Y， Yang C， Yang P， et al. Scatter correction of cone-beam CT using a deep residual convolution neural network［J］. Physics in Medicine and Biology， 2019， 64（14）： No.145003.
[25]	Zhong A， Li B， Luo N， et al. Image restoration for low-dose CT via transfer learning and residual network［J］. IEEE Access， 2020， 8： 112078-112091.
[26]	Xue H， Zhang Q， Zou S， et al. LCPR-Net： low-count PET image reconstruction using the domain transform and cycle-consistent generative adversarial networks［J］. Quantitative Imaging in Medicine and Surgery， 2021， 11（2）： 749-762.
[27]	Dosovitskiy A， Beyer L， Kolesnikov A， et al. An image is worth 16×16 words： Transformers for image recognition at scale［PP/OL］. V2. arXiv （2020-10-22）［2025-05-20］..
[28]	Liu Z， Lin Y， Cao Y， et al. Swin Transformer： hierarchical Vision Transformer using shifted windows［C］// ICCV 2021. Piscataway： IEEE， 2021： 9992-10002.
[29]	Yan H， Fang C， Liu P， et al. CGP-Uformer： a low-dose CT image denoising Uformer based on channel graph perception［J］. Journal of X-Ray Science and Technology， 2023， 31（6）： 1189-1205.
[30]	Huang H， Chen Z， Zou Y， et al. Channel prior convolutional attention for medical image segmentation［J］. Computers in Biology and Medicine， 2024， 178： No.108784.
[31]	Cui Y， Tha K K， Terasaka S， et al. Prognostic imaging biomarkers in glioblastoma： development and independent validation on the basis of multiregion and quantitative analysis of MR images［J］. Radiology， 2016， 278（2）： 546-553.
[32]	Bakas S， Akbari H， Sotiras A， et al. Advancing the cancer genome atlas glioma MRI collections with expert segmentation labels and radiomic features［J］. Scientific Data， 2017， 4： No.170117.
[33]	Kong Y， Xu M， Wei X， et al. CT imaging-based radiomics signatures improve prognosis prediction in postoperative colorectal cancer［J］. Journal of X-Ray Science and Technology， 2023， 31（6）： 1281-1294.
[34]	Lai W S， Huang J B， Ahuja N， et al. Deep Laplacian pyramid networks for fast and accurate super-resolution［C］// CVPR 2017. Piscataway： IEEE， 2017： 5835-5843.
[35]	Ramanarayanan S， Murugesan B， Kalyanasundaram A， et al. MRI super-resolution using Laplacian pyramid convolutional neural networks with isotropic undecimated wavelet loss［C］// EMBC 2020. Piscataway： IEEE， 2020： 1584-1587.
[36]	Gajera B， Kapil S R， Ziaei D， et al. CT-scan denoising using a Charbonnier loss generative adversarial network［J］. IEEE Access， 2021， 9： 84093-84109.
[37]	Horé A， Ziou D. Image quality metrics： PSNR vs. SSIM［C］// ICPR 2010. Piscataway： IEEE， 2010： 2366-2369.
[38]	Bai S B， Wang J， Lü G N， et al. GIS-based logistic regression for landslide susceptibility mapping of the Zhongxian segment in the Three Gorges area， China［J］. Geomorphology， 2010， 115（1/2）： 23-31.
[39]	Zhang K， Zuo W， Chen Y， et al. Beyond a Gaussian denoiser： residual learning of deep cnn for image denoising［J］. IEEE Transactions on Image Processing， 2017， 26（7）： 3142-3155.
[40]	Ronneberger O， Fischer P， Brox T. U-net： convolutional networks for biomedical image segmentation［C］// MICCAI 2015， LNCS 9351. Cham： Springer， 2015： 234-241.

模型	15视角			30视角
模型	PSNR/dB	SSIM	RMSE	PSNR/dB	SSIM	RMSE
DnCNN^［39］	28.321 6	0.841 9	0.038 9	31.739 8	0.893 0	0.026 2
RED-CNN^［23］	29.372 6	0.866 7	0.034 3	32.765 7	0.907 7	0.023 2
U-Net^［40］	31.516 3	0.892 5	0.026 9	35.082 4	0.928 2	0.017 8
FBPConvNet^［22］	31.642 5	0.893 6	0.026 5	35.422 0	0.930 9	0.017 2
Uformer^［13］	32.022 9	0.898 0	0.025 3	35.683 7	0.934 6	0.016 6
Restormer^［14］	31.981 3	0.898 1	0.025 5	35.734 0	0.934 1	0.016 5
AST^［15］	32.340 5	0.894 4	0.024 4	35.972 6	0.935 8	0.015 9
MAAF⁃Transformer	32.9030	0.9075	0.0229	36.3056	0.938 4	0.015 5
模型	60视角			90视角
模型	PSNR/dB	SSIM	RMSE	PSNR/dB	SSIM	RMSE
DnCNN^［39］	33.849 0	0.932 9	0.020 6	35.467 9	0.950 4	0.017 1
RED-CNN^［23］	35.417 3	0.944 3	0.017 1	36.630 3	0.956 7	0.014 9
U-Net^［40］	37.946 0	0.953 8	0.012 8	39.175 6	0.962 5	0.011 1
FBPConvNet^［22］	38.565 8	0.956 9	0.011 9	39.715 2	0.964 0	0.010 5
Uformer^［13］	38.795 2	0.959 1	0.011 6	39.992 4	0.965 8	0.010 1
Restormer^［14］	39.013 5	0.959 5	0.011 3	40.256 7	0.966 7	0.009 8
AST^［15］	39.213 6	0.960 7	0.011 1	40.424 7	0.967 3	0.009 7
MAAF⁃Transformer	39.509 3	0.962 3	0.010 7	40.521 9	0.967 7	0.0095

模型	15视角			30视角
模型	PSNR/dB	SSIM	RMSE	PSNR/dB	SSIM	RMSE
DnCNN^［39］	28.321 6	0.841 9	0.038 9	31.739 8	0.893 0	0.026 2
RED-CNN^［23］	29.372 6	0.866 7	0.034 3	32.765 7	0.907 7	0.023 2
U-Net^［40］	31.516 3	0.892 5	0.026 9	35.082 4	0.928 2	0.017 8
FBPConvNet^［22］	31.642 5	0.893 6	0.026 5	35.422 0	0.930 9	0.017 2
Uformer^［13］	32.022 9	0.898 0	0.025 3	35.683 7	0.934 6	0.016 6
Restormer^［14］	31.981 3	0.898 1	0.025 5	35.734 0	0.934 1	0.016 5
AST^［15］	32.340 5	0.894 4	0.024 4	35.972 6	0.935 8	0.015 9
MAAF⁃Transformer	32.9030	0.9075	0.0229	36.3056	0.938 4	0.015 5
模型	60视角			90视角
模型	PSNR/dB	SSIM	RMSE	PSNR/dB	SSIM	RMSE
DnCNN^［39］	33.849 0	0.932 9	0.020 6	35.467 9	0.950 4	0.017 1
RED-CNN^［23］	35.417 3	0.944 3	0.017 1	36.630 3	0.956 7	0.014 9
U-Net^［40］	37.946 0	0.953 8	0.012 8	39.175 6	0.962 5	0.011 1
FBPConvNet^［22］	38.565 8	0.956 9	0.011 9	39.715 2	0.964 0	0.010 5
Uformer^［13］	38.795 2	0.959 1	0.011 6	39.992 4	0.965 8	0.010 1
Restormer^［14］	39.013 5	0.959 5	0.011 3	40.256 7	0.966 7	0.009 8
AST^［15］	39.213 6	0.960 7	0.011 1	40.424 7	0.967 3	0.009 7
MAAF⁃Transformer	39.509 3	0.962 3	0.010 7	40.521 9	0.967 7	0.0095

模型	PSNR/dB	SSIM	RMSE
No GDFN	39.201 4	0.960 9	0.011 1
No CASCA	39.242 5	0.960 8	0.011 1
No MSCDA	39.264 3	0.960 9	0.011 0
MAAF⁃Transformer	39.509 3	0.962 3	0.010 7

模型	PSNR/dB	SSIM	RMSE
No GDFN	39.201 4	0.960 9	0.011 1
No CASCA	39.242 5	0.960 8	0.011 1
No MSCDA	39.264 3	0.960 9	0.011 0
MAAF⁃Transformer	39.509 3	0.962 3	0.010 7

模块	PSNR/dB	SSIM	RMSE
SMC	39.209 4	0.960 5	0.011 1
SCM	39.423 0	0.961 8	0.010 8
MAAF⁃Transformer	39.509 3	0.962 3	0.010 7

Sparse CT artifact suppresion Transformer network based on multi-scale attention adaptive fusion

基于多尺度注意力自适应融合的稀疏CT伪影抑制Transformer网络

RichHTML

PDF

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 12

References 40

Related Articles 15

Recommended Articles

Metrics

[1]	Guanghui LI, Licheng QU. Dynamic dilated convolution and hierarchical local-global attention model based on improved Lite-Mono architecture [J]. Journal of Computer Applications, 2026, 46(8): 2567-2576.
[2]	Lin HUANG, Yingshun LI, Weiyan TONG, Shuyuan ZHANG, Zihan WANG. Remote sensing image object detection algorithm based on improved real-time detection Transformer [J]. Journal of Computer Applications, 2026, 46(8): 2652-2659.
[3]	Xiaojin GUO, Xuyang SUI, Kenan ZHOU. Channel estimation of STAR-RIS-assisted hybrid-field communication system based on deep learning [J]. Journal of Computer Applications, 2026, 46(8): 2548-2554.
[4]	Ming LIU, Dongqi SHEN, Ziyang MENG. Point cloud registration network with dual-branch multi-level feature fusion [J]. Journal of Computer Applications, 2026, 46(8): 2584-2593.
[5]	Fengchun LIU, Xinying SHAO, Chunying ZHANG, Liya WANG, Jing REN. FCMdepth： monocular depth estimation framework with multi-scale feature optimization [J]. Journal of Computer Applications, 2026, 46(8): 2603-2611.
[6]	Wei LIU, Weigang LI, Zhiqiang TIAN. Representation learning method of hierarchical rotation-invariant geometric structure for point cloud classification and segmentation [J]. Journal of Computer Applications, 2026, 46(8): 2594-2602.
[7]	Li TANG, Jianyu ZHANG, Rui YAO. Object detection algorithm based on improved YOLOv11 in foggy weather [J]. Journal of Computer Applications, 2026, 46(8): 2630-2639.
[8]	Jianqiang YAN, Beibei DONG, Boting QU, Chen PENG. Bi-directional diffusion dynamic graph convolutional network with multi-source information fusion and graph-level attention for traffic flow prediction [J]. Journal of Computer Applications, 2026, 46(7): 2327-2333.
[9]	Xiuli DU, Xing GAO, Xiaoyu ZHANG, Chengsheng PAN, Qijie ZOU. Video snapshot compressive imaging reconstruction method based on dense spatio-temporal deformable attention [J]. Journal of Computer Applications, 2026, 46(7): 2288-2296.
[10]	Guoyou ZHANG, Hongyu NIE, Lihu PAN, Rundong LEI. Point-MLPBLS： point cloud semantic segmentation network based on MLP cascaded broad learning system [J]. Journal of Computer Applications, 2026, 46(7): 2259-2266.
[11]	Wenling LIU, Yong LI, Jiahui LI, Houkang ZHANG. Unsupervised dehazing method for remote sensing images with non-uniform haze [J]. Journal of Computer Applications, 2026, 46(7): 2297-2306.
[12]	Li ZHU, Lugema MI, Chunqiang ZHU, Wanru XU, Jingkai GAO, Jinqi QU. Short-term power load forecasting model based on dynamic convolution decomposition and multi-scale graph [J]. Journal of Computer Applications, 2026, 46(7): 2334-2346.
[13]	Yunping HE, Leichun WANG, Ruirui SONG, Xiangfeng LU, Jinxiang WEI, Xiaomeng LIU. Dual-channel multimodal sentiment analysis model based on contrast invariance and reinforcement specificity [J]. Journal of Computer Applications, 2026, 46(6): 1767-1775.
[14]	Liwan YAO, Hailong LIU, Zhangfan ZENG. Frequency-domain driven and diffusion-based fusion for sonar image enhancement algorithm [J]. Journal of Computer Applications, 2026, 46(6): 1947-1955.
[15]	Chao LYU, Geyao MA. Lightweight human pose estimation network based on redundant feature suppression [J]. Journal of Computer Applications, 2026, 46(6): 1973-1980.