To address the core challenge of dependency on paired LDCT-NDCT (Low Dose CT-Normal Dose CT) data in low-dose CT denoising, this paper proposes CTDM(CNN-Transformer Diffusion Model), an unsupervised denoising model based on diffusion priors. We establish a joint learning framework that integrates noise distribution estimation and cross-scale structure preservation, high-fidelity CT denoising is achieved without the constraints of paired data. First, we construct an anatomy-prior constrained diffusion probabilistic model, where pre-trained prior knowledge is dynamically embedded into the iterative denoising process via maximum a posteriori (MAP) optimization, enabling reliable noise suppression under unsupervised conditions. Second, we design a CNN-Transformer dual-stream hybrid network that combines parallelized local-global feature extraction with adaptive fusion mechanisms, achieving efficient denoising within a single-stage training framework. Compared to traditional cascaded models, our approach significantly reduces GPU memory consumption and inference time by eliminating multi-stage serial processing. Experimental results demonstrate that CTDM outperforms state-of-the-art unsupervised methods and even surpasses supervised models like CTformer in denoising fidelity.
1) 设计单阶段扩散模型实现无监督LDCT去噪,通过预训练模型提供的图像先验知识构建最大后验概率(Maximum A Posteriori,MAP)问题,并迭代求解去噪结果。相较于需级联多阶段优化的传统扩散模型,该设计不仅摆脱了对配对LDCT-NDCT数据的依赖,更通过单阶段逆向扩散过程实现了去噪加速。
HobbsJ B, GoldsteinN, LindK E,et al.Physician knowledge of radiation exposure and risk in medical imaging[J].J Am Coll Radiol, 2018, 15(1): 34-43.
[2]
ChenH, ZhangY, KalraM K,et al.Low-dose CT with a residual encoder-decoder convolutional neural network[J].IEEE Trans Med Imaging,2017,36(12): 2524-2535.
[3]
MingJ, YiB S, ZhangY G,et al.2020.Low-dose CT image denoising using classification densely connected residual network[J].KSII Transactions on Internet and Information Systems (TIIS), 2020,6: 2480-2496.
[4]
LiangT, JinY, LiY, et al.EDCNN: edge enhancement-based densely connected network with compound loss for low-dose CT denoising[C]//2020 15th IEEE International Conference on Signal Processing (ICSP).Beijing,China: IEEE,2020: 193-198.
[5]
LiM, HsuW, XieX,et al.SACNN: Self-attention convolutional neural network for low-dose CT denoising with self-supervised perceptual loss network[J].IEEE Trans Med Imaging, 2020, 39(7): 2289-2301.
[6]
WangX, GirshickR, GuptaA,et al.Non-local neural networks[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Salt Lake City, UT, USA: IEEE, 2018: 7794-7803.
[7]
LuthraA, SulakheH, MittalT,et al.Eformer:Edge enhancement based transformer for medical image denoising[EB/OL].[2025-05-01].
[8]
WangD, FanF, WuZ, et al.CTformer: Convolution-free Token2Token dilated vision transformer for low-dose CT denoising[J].Phys Med Biol, 2023, 68(6): 065012.
[9]
LiH, YangX, YangS,et al.Transformer with double enhancement for low-dose CT denoising[J].IEEE J Biomed Health Inform,2023,27(10): 4660-4671.
[10]
ZhangZ, YuL, LiangX,et al.TransCT: Dual-path transformer for low dose computed tomography [M] // Medical image computing and computer assisted intervention-MICCAI 2021.Cham: Springer International Publishing, 2021: 55-64.
[11]
ZhangJ, ShangguanZ, GongW,et al.A novel denoising method for low-dose CT images based on transformer and CNN[J].Comput Biol Med, 2023, 163: 107162.
[12]
LiuX, LiangX, DengL,et al.Learning low-dose CT degradation from unpaired data with flow-based model [J].Med Phys, 2022, 49(12): 7516-7530.
[13]
YuanN, ZhouJ, QiJ.Half2Half: Deep neural network based CT image denoising without independent reference data[J].Phys Med Biol,2020, 65(21): 215020.
[14]
NiuC, LiM, FanF,et al.Noise suppression with similarity-based self-supervised deep learning[J].IEEE Trans Med Imaging, 2023, 42(6): 1590-1602.
BaoF, NieS, XueK,et al.One transformer fits all distributions in multi-modal diffusion at scale[C]//International Conference on Machine Learning.[S.l.]: PMLR, 2023: 1692.
[31]
ZhaoZ, BaiH, ZhuY,et al.DDFM: Denoising diffusion model for multi-modality image fusion[C]//2023 IEEE/CVF International Conference on Computer Vision (ICCV).Paris: IEEE, 2023: 8048-8059.
[32]
YuC, SongY, WangQ,et al.Leveraging denoising diffusion probabilistic model to improve the multi-thickness CT segmentation[J].Neurocomputing,2024,610: 128573.
[33]
LiuZ, LinY, CaoY,et al.Swin transformer: Hierarchical vision transformer using shifted windows[C]//2021 IEEE/CVF International Conference on Computer Vision (ICCV).Montreal,QC,Canada: IEEE, 2021: 9992-10002.
[34]
LiH, SongQ, GuiD,et al.Reconstruction-assisted feature encoding network for histologic subtype classification of non-small cell lung cancer[J].IEEE J Biomed Health Inform, 2022, 26(9): 4563-4574.
[35]
WangP, ZhengW, ChenT,et al.Anti-oversmoothing in deep vision transformers via the Fourier domain analysis: From theory to practice[EB/OL].[2025-05-08].
[36]
MoenT R, ChenB, HolmesIII D R,et al.Low dose CT image and projection data[J].Med Phys, 2021, 48(2): 902-911.
[37]
McColloughC H, BartleyA C, CarterR E,et al.Low-dose CT for the detection and classification of metastatic liver lesions: Results of the 2016 Low Dose CT Grand Challenge[J].Med Phys, 2017, 44(10): e339-e352.
[38]
LuhmanT, LuhmanE.Improving diffusion model efficiency through patching[EB/OL].[2025-05-08].
[39]
DabovK, FoiA, KatkovnikV,et al.Image denoising by sparse 3-D transform-domain collaborative filtering [J].IEEE Trans Image Process, 2007, 16(8): 2080-2095.
[40]
BuadesA, CollB, MorelJ M.A non-local algorithm for image denoising[C] //2005 IEEE ComputerSociety Conference on Computer Vision and Pattern Recognition (CVPR’05).San Diego: IEEE,2005: 60.
[41]
GengM, MengX, YuJ,et al.Content-noise complementary learning for medical image denoising[J].IEEE Trans Med Imaging, 2022, 41(2): 407-419.
[42]
GuoY F, PeiX, WangD H, et al. HyperSegUNet: A hyperparameter self-learning medical image segmentation network based on hypernetworks[J].Journal of Sichuan Normal University(Natural Science), 2024, 47(1): 127-135.
WangZ Q, ZengX H. Contrast-enhanced relational memory network for medical image report generation [J].Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition),2024, 36(3): 503-512.