Trident generative adversarial network for low-dose CT image denoising

doi:10.11772/j.issn.1001-9081.2024121765

Journal of Computer Applications ›› 2026, Vol. 46 ›› Issue (1): 270-279.DOI: 10.11772/j.issn.1001-9081.2024121765

• Multimedia computing and computer simulation • Previous Articles Next Articles

Trident generative adversarial network for low-dose CT image denoising

Lifang WANG¹, Wenjing REN¹, Xiaodong GUO²(), Rongguo ZHANG¹, Lihua HU¹

^1.College of Computer Science and Technology，Taiyuan University of Science and Technology，Taiyuan Shanxi 030024，China
^2.School of Electronic Information Engineering，Taiyuan University of Science and Technology，Taiyuan Shanxi 030024，China

Received:2024-12-16 Revised:2025-03-17 Accepted:2025-03-18 Online:2026-01-10 Published:2026-01-10
Contact: Xiaodong GUO
About author:WANG Lifang， born in 1975， Ph. D.， associate professor. Her research interests include image and graphics processing， intelligent optimization.
REN Wenjing， born in 2000， M. S. candidate. Her research interests include computer vision， image denoising.
ZHANG Rongguo， born in 1964， Ph. D.， professor. His research interests include computer graphics and aided design， image and graphics processing， pattern recognition.
HU Lihua， born in 1982， Ph. D.， professor. Her research interests include computer vision， pattern recognition.
Supported by:
National Natural Science Foundation of China(62273248);Natural Science Foundation of Shanxi Province(202203021211206);Postgraduate Educational Reform Project of Shanxi Province(2022YJJG191);Postgraduate Innovation Project of Shanxi Province(2023KY657)

用于低剂量CT图像降噪的多路特征生成对抗网络

王丽芳¹, 任文婧¹, 郭晓东²(), 张荣国¹, 胡立华¹

^1.太原科技大学计算机科学与技术学院，太原 030024
^2.太原科技大学电子信息工程学院，太原 030024

通讯作者: 郭晓东
作者简介:王丽芳（1975—），女，山西和顺人，副教授，博士， CCF会员，主要研究方向：图形图像处理、智能优化
任文婧（2000—），女，山西长治人，硕士研究生，主要研究方向：计算机视觉、图像降噪
张荣国（1964—），男，山西太原人，教授，博士， CCF会员，主要研究方向：计算机图形学与辅助设计、图形图像处理、模式识别
胡立华（1982—），女，山西忻州人，教授，博士， CCF会员，主要研究方向：计算机视觉、模式识别。
基金资助:
国家自然科学基金资助项目(62273248);山西省自然科学基金资助项目(202203021211206);山西省研究生教育改革项目(2022YJJG191);山西省研究生科研创新项目(2023KY657)

Abstract

Abstract:

In recent years， significant progress has been made in applying Generative Adversarial Network （GAN） in Low-Dose Computed Tomography （LDCT） image denoising. However， the existing methods face challenges such as insufficient modeling capability for complex noise distribution and limited ability to preserve structural details. Therefore， a multi-path GAN for LDCT image denoising — Trident GAN was proposed. Firstly， a feature guided generator Trident Uformer was designed. In this generator， a Feature Polymerization Attention （FPA） module was added at the bottleneck layer of the U-Net structure， thereby solving the problem of low spatial resolution in a U-shaped structure. Secondly， a multi-path feature extraction submodule Trident Block was designed， and in each of the three blocks， a Local Detail Enhancement Block （LDEB） was introduced to extract detailed features， a Lightweight Channel Attention Block （LCAB） was incorporated to enhance channel features， and a Spatial Interaction Attention Block （SIAB） was utilized to capture important spatial features， respectively. Within the SIAB， a multi-level interactive attention function and evaluation mechanism were employed to design a Spatial Context Attention Mechanism （SCAM）， which addresses the limitations of single-attention mechanisms. Finally， a Multi-Feature Fusion （MFF） module was designed to realize feature aggregation at the end of the three blocks， and model both local detail information and global semantic information， and solve the problem of discontinuous details across different levels. Furthermore， the Multi-Scale Pyramid Discriminator （MSPD） was used to check the quality of the generated results at different dimensions， and guide the generation of globally consistent images. Experimental results show that Trident GAN achieves the average Peak Signal-to-Noise Ratio （PSNR） and Structural Similarity Index Measure （SSIM） of 31.519 3 dB/0.883 0 and 33.633 1 dB/0.947 8， respectively， on Mayo and Piglet datasets. Compared with High-Frequency Sensitive GAN （HFSGAN）， this method has the number of parameters reduced by 75.58% and the test time reduced by 36.36%. It can be seen that compared with the existing methods such as HFSGAN， Trident GAN improves image quality with less computational load.

Key words: Low-Dose Computed Tomography (LDCT), image denoising, attention mechanism, Transformer, Generative Adversarial Network (GAN)

摘要：

近些年，把生成对抗网络（GAN）应用于低剂量计算机断层扫描（LDCT）图像降噪取得了显著进展。然而，现有方法存在对复杂噪声分布建模能力不足以及结构细节保留能力有限等问题。因此，提出一种用于LDCT图像降噪的多路特征GAN ——Trident GAN。首先，设计特征引导生成器Trident Uformer，通过在U-Net结构的瓶颈层增加特征聚合注意力（FPA）模块解决U型结构空间分辨率较低的问题；其次，设计多路特征提取子模块Trident Block，并在3个分支中分别引入局部细节增强模块（LDEB）提取细节特征，轻量通道注意力模块（LCAB）增强通道特征，以及空间交互注意力模块（SIAB）获得重要空间特征；在SIAB中采用多级交互式注意力函数和评估机制设计空间上下文注意力机制（SCAM），解决单一注意力受限的问题；最后，设计多特征融合（MFF）模块来在三分支末端进行特征聚合，并对局部细节信息和全局语义信息进行建模，解决不同层次之间细节不连续的问题。此外，利用多尺度金字塔判别器（MSPD）在不同维度下检查生成结果的质量，指导具有全局一致性图像的生成。实验结果表明，在Mayo和Piglet数据集上，Trident GAN的平均峰值信噪比（PSNR）和结构相似性（SSIM）分别达到了31.519 3 dB/0.883 0和33.633 1 dB/0.947 8，与高频敏感GAN （HFSGAN）相比，参数量降低75.58%，测试时间缩短36.36%。可见，与HFSGAN等方法相比， Trident GAN可在较少的计算负荷下提高了图像质量。

关键词: 低剂量计算机断层扫描, 图像降噪, 注意力机制, Transformer, 生成对抗网络

CLC Number:

TP391.41

Lifang WANG, Wenjing REN, Xiaodong GUO, Rongguo ZHANG, Lihua HU. Trident generative adversarial network for low-dose CT image denoising[J]. Journal of Computer Applications, 2026, 46(1): 270-279.

王丽芳, 任文婧, 郭晓东, 张荣国, 胡立华. 用于低剂量CT图像降噪的多路特征生成对抗网络[J]. 《计算机应用》唯一官方网站, 2026, 46(1): 270-279.

Figures/Tables 17

References 33

[1]	XIA W， SHAN H， WANG G， et al. Physics-/model-based and data-driven methods for low-dose CT： a survey ［J］. IEEE Signal Processing Magazine， 2023， 40（2）： 89-100.
[2]	李云飞，李淑婷，张帅，等.深度学习在肿瘤影像分类中的研究进展［J］.中华肿瘤防治杂志， 2024， 31（12）： 719-724.
	LI Y F， LI S T， ZHANG S， et al. Research progress of deep learning in tumor images classification ［J］. Chinese Journal of Cancer Prevention and Treatment， 2024， 31（12）： 719-724.
[3]	RONNEBERGER O， FISCHER P， BROX T. U-net： convolutional networks for biomedical image segmentation ［C］// Proceedings of the 2015 International Conference on Medical Image Computing and Computer-Assisted Intervention， LNCS 9351. Cham： Springer， 2015： 234-241.
[4]	ZHANG J， NIU Y， SHANGGUAN Z， et al. A novel denoising method for CT images based on U-net and multi-attention ［J］. Computers in Biology and Medicine， 2023， 152： No.106387.
[5]	GAO Q， LI Z L， ZHANG J， et al. CoreDiff： contextual error-modulated generalized diffusion model for low-dose CT denoising and generalization ［J］. IEEE Transactions on Medical Imaging， 2024， 43（2）： 745-759.
[6]	XIA W， LYU Q， WANG G. Low-dose CT using denoising diffusion probabilistic model for 20× speedup ［EB/OL］. ［2024-11-29］. .
[7]	GOODFELLOW I J， POUGET-ABADIE J， MIRZA M， et al. Generative adversarial networks ［J］. Communications of the ACM， 2020， 63（11）： 139-144.
[8]	WOLTERINK J M， LEINER T， VIERGEVER M A， et al. Generative adversarial networks for noise reduction in low-dose CT ［J］. IEEE Transactions on Medical Imaging， 2017， 36（12）： 2536-2545.
[9]	CHI J， WU C， YU X， et al. Single low-dose CT image denoising using a generative adversarial network with modified U-Net generator and multi-level discriminator ［J］. IEEE Access， 2020， 8： 133470-133487.
[10]	YANG C， SHEN Y， XU Y， et al. Improving GANs with a dynamic discriminator ［C］// Proceedings of the 36th International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2022： 15093-15104.
[11]	YANG L， SHANGUAN H， ZHANG X， et al. High-frequency sensitive generative adversarial network for low-dose CT image denoising ［J］. IEEE Access， 2020， 8： 930-943.
[12]	MAO X， LI Q， XIE H， et al. Least squares generative adversarial networks ［C］// Proceedings of the 2017 IEEE International Conference on Computer Vision. Piscataway： IEEE， 2017： 2813-2821.
[13]	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 A Y， ZHAO H C， CAI W L， et al. Adaptive image deblurring generative adversarial network algorithm based on active discrimination mechanism ［J］. Journal of Computer Applications， 2023， 43（7）： 2288-2294.
[14]	DALMAZ O， YURT M， ÇUKUR T. ResViT： residual vision transformers for multimodal medical image synthesis ［J］. IEEE Transactions on Medical Imaging， 2022， 41（10）： 2598-2614.
[15]	ZHAO J， LI D， KASSAM Z， et al. Tripartite-GAN： synthesizing liver contrast-enhanced MRI to improve tumor detection ［J］. Medical Image Analysis， 2020， 63： No.101667.
[16]	DOSOVITSKIY A， BEYER L， KOLESNIKOV A， et al. An image is worth 16×16 words： Transformers for image recognition at scale ［EB/OL］. ［2024-11-29］. .
[17]	CHEN Z， NIU C， GAO Q， et al. LIT-Former： linking in-plane and through-place Transformers for simultaneous CT image denoising and deblurring ［J］. IEEE Transactions on Medical Imaging， 2024， 43（5）： 1880-1894.
[18]	LI H， YANG X， YANG S， et al. Transformer with double enhancement for low-dose CT denoising ［J］. IEEE Journal of Biomedical and Health Informatics， 2023， 27（10）： 4660-4671.
[19]	PENG J， LI X， ZHANG X. MDnT： a multi-scale denoising Transformer beyond real noise image denoising ［C］// Proceedings of the 7th International Conference on Electronic Technology and Information Science. Piscataway： IEEE， 2022： 1-5.
[20]	WANG J， ZHANG B， WANG Y， et al. CrossU-Net： dual-modality cross-attention U-Net for segmentation of precancerous lesions in gastric cancer ［J］. Computerized Medical Image and Graphics， 2024， 112： No.102339.
[21]	WANG P， ZHU H， ZHANG H， et al. LRB-T： local reasoning back-projection transformer for the removal of bad weather effects in images ［J］. Neural Computing and Applications， 2024， 36（2）： 773-789.
[22]	AMIR S W， ARORA A， KHAN S， et al. Restormer： efficient Transformer for high-resolution image restoration ［C］// Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2022： 5718-5729.
[23]	CHEN C F R， FAN Q， PANDA R. CrossViT： cross-attention multi-scale Vision Transformer for image classification ［C］// Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision. Piscataway： IEEE， 2021： 347-356.
[24]	American Association of Physicists in Medicine. Low dose CT grand challenge ［EB/OL］. ［2024-04-20］. .
[25]	Piglet dataset ［DB/OL］. ［2024-04-26］. .
[26]	LEDIG C， THEIS L， HUSZÁR F， et al. Photo-realistic single image super-resolution using a generative adversarial network ［C］// Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2017： 105-114.
[27]	DABOV K， FOI A， KATKOVNIK V， et al. Image denoising by sparse 3-D transform-domain collaborative filtering ［J］. IEEE Transactions on Image Processing， 2007， 16（8）： 2080-2095.
[28]	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.
[29]	ISOLA P， ZHU J Y， ZHOU T， et al. Image-to-image translation with condition adversarial networks ［C］// Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2017： 5967-5976.
[30]	GENG M， MENG X， YU J， et al. Content-noise complementary learning for medical image denoising ［J］. IEEE Transactions on Medical Imaging， 2022， 41（2）： 407-419.
[31]	WANG D， WU Z， YU H. TED-Net： convolution-free T2T vision transformer-based encoder-decoder dilation network for low-dose CT denoising ［C］// Proceedings of the 2021 International Workshop on Machine Learning in Medical Imaging， LNCS 12966. Cham： Springer， 2021： 416-425.
[32]	SONG Y， HE Z， QIAN H， et al. Vision Transformers for single image dehazing ［J］. IEEE Transactions on Image Processing， 2023， 32： 1927-1941.
[33]	上官宏，任慧莹，张雄，等.基于双编码器双解码器GAN的低剂量CT降噪模型［J］.计算机应用， 2025， 45（2）： 624-632.
	SHANGGUAN H， REN H Y， ZHANG X， et al. Low-dose CT denoising model based on dual encoder-decoder generative adversarial network ［J］. Journal of Computer Applications， 2025， 45（2）： 624-632.

方法	PSNR/dB	SSIM
LDCT	26.789 1±1.721 8	0.824 4±0.042 6
BM3D	27.741 2±1.524 8	0.834 5±0.026 5
RED-CNN	27.783 2±1.548 8	0.844 2±0.022 6
pix2pix	28.325 3±1.339 4	0.823 1±0.033 9
HFSGAN	30.081 8±1.167 5	0.852 1±0.018 4
CNCL	29.065 8±2.353 4	0.860 9±0.022 4
TED-Net	31.087 1±1.754 4	0.878 3±0.039 9
DehazeFormer	31.128 1±1.803 3	0.880 5±0.040 0
DualED-GAN	31.473 6±1.749 1	0.876 6±0.041 2
Trident GAN	31.5193±1.9214	0.8830±0.0422

方法	PSNR/dB	SSIM
LDCT	26.789 1±1.721 8	0.824 4±0.042 6
BM3D	27.741 2±1.524 8	0.834 5±0.026 5
RED-CNN	27.783 2±1.548 8	0.844 2±0.022 6
pix2pix	28.325 3±1.339 4	0.823 1±0.033 9
HFSGAN	30.081 8±1.167 5	0.852 1±0.018 4
CNCL	29.065 8±2.353 4	0.860 9±0.022 4
TED-Net	31.087 1±1.754 4	0.878 3±0.039 9
DehazeFormer	31.128 1±1.803 3	0.880 5±0.040 0
DualED-GAN	31.473 6±1.749 1	0.876 6±0.041 2
Trident GAN	31.5193±1.9214	0.8830±0.0422

方法	胸部LDCT图像		腹部LDCT图像
方法	PSNR/dB	SSIM	PSNR/dB	SSIM
LDCT	15.354	0.483	21.327	0.483
BM3D	16.001	0.499	25.792	0.621
RED-CNN	17.338	0.600	25.994	0.616
pix2pix	18.122	0.613	23.131	0.462
HFSGAN	18.483	0.638	25.352	0.561
CNCL	18.425	0.634	23.620	0.550
TED-Net	18.788	0.652	25.872	0.584
DehazeFormer	19.006	0.667	26.024	0.616
DualED-GAN	18.740	0.612	26.056	0.599
Trident GAN	19.615	0.675	26.490	0.621

方法	胸部LDCT图像		腹部LDCT图像
方法	PSNR/dB	SSIM	PSNR/dB	SSIM
LDCT	15.354	0.483	21.327	0.483
BM3D	16.001	0.499	25.792	0.621
RED-CNN	17.338	0.600	25.994	0.616
pix2pix	18.122	0.613	23.131	0.462
HFSGAN	18.483	0.638	25.352	0.561
CNCL	18.425	0.634	23.620	0.550
TED-Net	18.788	0.652	25.872	0.584
DehazeFormer	19.006	0.667	26.024	0.616
DualED-GAN	18.740	0.612	26.056	0.599
Trident GAN	19.615	0.675	26.490	0.621

方法	PSNR/dB	SSIM
LDCT	32.437 6±3.032 2	0.913 9±0.035 4
BM3D	32.740 1±3.002 1	0.934 6±0.022 4
RED-CNN	32.127 2±2.893 9	0.918 6±0.022 0
pix2pix	33.033 6±2.704 7	0.911 7±0.033 4
HFSGAN	33.240 6±2.404 0	0.923 6±0.031 2
CNCL	32.924 0±2.322 1	0.927 0±0.027 3
TED-Net	32.827 5±2.545 7	0.932 9±0.025 0
DehazeFormer	33.395 3±2.896 8	0.937 7±0.024 4
DualED-GAN	33.471 8±2.748 3	0.936 3±0.025 6
Trident GAN	33.6331±2.6214	0.9478±0.0322

Trident generative adversarial network for low-dose CT image denoising

用于低剂量CT图像降噪的多路特征生成对抗网络

RichHTML

PDF

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 17

References 33

Related Articles 15

Recommended Articles

Metrics

方法	参数量/10⁶	单幅测试时间/s
BM3D	—	1.207
RED-CNN	1.84	0.104
pix2pix	57.19	0.067
HFSGAN	108.87	0.088
CNCL	46.59	0.109
TED-Net	1.75	0.142
DehazeFormer	2.25	0.067
DualED-GAN	25.11	0.072
Trident GAN	26.59	0.056

模型	子模块				PSNR/dB	SSIM
模型	Trident Block	SCAM	MFF	MSPD	PSNR/dB	SSIM
w/o Tri1		√	√	√	29.712 8	0.879 9
w/o Tri2	√		√	√	31.052 1	0.867 0
w/o MFF	√	√		√	31.120 5	0.877 2
w/o MSPD	√	√	√		30.975 4	0.875 1
Trident GAN	√	√	√	√	31.5193	0.8830

模型	PSNR/dB	SSIM
1-FPA	31.249 3	0.876 1
Trident GAN （2-FPA）	31.5193	0.883 0
3-FPA	31.349 1	0.8850

[1]	Yingjie MA, Jingying QIN, Geng ZHAO, Jing XIAO. Deep compressive sensing network for IoT images and its chaotic encryption protection method [J]. Journal of Computer Applications, 2026, 46(1): 144-151.
[2]	Junheng WU, Xiaodong WANG, Qixue HE. Time series prediction model based on statistical distribution sensing and frequency domain dual-channel fusion [J]. Journal of Computer Applications, 2026, 46(1): 113-123.
[3]	Na FAN, Chuang LUO, Zehui ZHANG, Mengyao ZHANG, Ding MU. Semantic privacy protection mechanism of vehicle trajectory based on improved generative adversarial network [J]. Journal of Computer Applications, 2026, 46(1): 169-180.
[4]	Yanan LI, Mengyang GUO, Guojun DENG, Yunfeng CHEN, Jianji REN, Yongliang YUAN. Method for life prediction of parallel branching engine based on multi-modal fusion features [J]. Journal of Computer Applications, 2026, 46(1): 305-313.
[5]	Lijin YAO, Di ZHANG, Piyu ZHOU, Zhijian QU, Haipeng WANG. Transformer and gated recurrent unit-based de novo sequencing algorithm for phosphopeptides [J]. Journal of Computer Applications, 2026, 46(1): 297-304.
[6]	Yu SANG, Tong GONG, Chen ZHAO, Bowen YU, Siman LI. Domain-adaptive nighttime object detection method with photometric alignment [J]. Journal of Computer Applications, 2026, 46(1): 242-251.
[7]	Yinlong JIAN, Xuebin CHEN, Zhongrui JING, Qi ZHONG, Zhenbo ZHANG. Data augmentation scheme based on conditional generative adversarial network in federated learning [J]. Journal of Computer Applications, 2026, 46(1): 21-32.
[8]	Yilin DENG, Fajiang YU. Pseudo random number generator based on LSTM and separable self-attention mechanism [J]. Journal of Computer Applications, 2025, 45(9): 2893-2901.
[9]	Jinggang LYU, Shaorui PENG, Shuo GAO, Jin ZHOU. Speech enhancement network driven by complex frequency attention and multi-scale frequency enhancement [J]. Journal of Computer Applications, 2025, 45(9): 2957-2965.
[10]	Yiming LIANG, Jing FAN, Wenze CHAI. Multi-scale feature fusion sentiment classification based on bidirectional cross attention [J]. Journal of Computer Applications, 2025, 45(9): 2773-2782.
[11]	Jin LI, Liqun LIU. SAR and visible image fusion based on residual Swin Transformer [J]. Journal of Computer Applications, 2025, 45(9): 2949-2956.
[12]	Weigang LI, Jiale SHAO, Zhiqiang TIAN. Point cloud classification and segmentation network based on dual attention mechanism and multi-scale fusion [J]. Journal of Computer Applications, 2025, 45(9): 3003-3010.
[13]	Xiang WANG, Zhixiang CHEN, Guojun MAO. Multivariate time series prediction method combining local and global correlation [J]. Journal of Computer Applications, 2025, 45(9): 2806-2816.
[14]	Fang WANG, Jing HU, Rui ZHANG, Wenting FAN. Medical image segmentation network with content-guided multi-angle feature fusion [J]. Journal of Computer Applications, 2025, 45(9): 3017-3025.
[15]	Li LI, Han SONG, Peihe LIU, Hanlin CHEN. Named entity recognition for sensitive information based on data augmentation and residual networks [J]. Journal of Computer Applications, 2025, 45(9): 2790-2797.