Opto-Electronic Engineering, Volume. 52, Issue 6, 240306(2025)
Low-dose nano-CT reconstruction image denoising based on Swin Transformer and convolutional neural network
Figures & Tables(8)
Fig. 2. Denoising results of reconstructed slices of Huangpeng bentonite particles obtained by different methods
Fig. 3. Reconstructed slices denoising results of Huangpeng bentonite particles
Fig. 4. (a) Conventional nano-CT reconstruction result; (b) Low-dose nano-CT reconstruction result; (c) EDCNN denoising result; (d) SwinIR denoising result; (e) SwinCBD denoising result; (f) Profile map
Table 1. Quantitative comparison of different image denoising algorithms
View tableView in ArticleTable 1. Quantitative comparison of different image denoising algorithms
Method DAT SNR CNR SSIM NIQE BRISQUE GT 2 h 40 min 24.14 4.06 / / / LDCT 5 min 13.94 2.33 0.74 20.98 59.07 BM3D / 20.14 3.93 0.83 8.41 51.33 DnCNN / 11.19 2.63 0.77 8.91 54.29 Noise2Noise / 10.59 2.96 0.78 4.39 38.55 EDCNN / 12.97 2.55 0.75 17.05 57.30 SwinIR / 24.94 4.44 0.81 6.68 48.39 SwinCBD / 30.85 6.55 0.85 6.13 42.03
Table 2. Quantitative comparison of reconstructed slices before and after treatment with different projection numbers
View tableView in ArticleTable 2. Quantitative comparison of reconstructed slices before and after treatment with different projection numbers
评价指标 状态 GT 181 projections 91 projections 61 projections SNR 去噪前 24.14 13.94 13.88 10.57 去噪后 / 30.85 19.63 20.64 CNR 去噪前 4.06 2.33 1.33 0.86 去噪后 / 6.55 3.68 1.56
Table 3. Quantitative image quality metrics for NCM811 particle
View tableView in ArticleTable 3. Quantitative image quality metrics for NCM811 particle
模型 SNR CNR SSIM NIQE BRISQUE GT 14.94 5.25 / / / LDCT 10.18 2.21 0.47 5.71 54.75 EDCNN 10.03 2.39 0.36 5.64 52.96 SwinIR 21.06 8.87 0.66 5.02 46.78 SwinCBD 22.05 9.42 0.69 4.99 44.67
Table 4. Comparison of ablation modeling results
View tableView in ArticleTable 4. Comparison of ablation modeling results
模型 SNR CNR NIQE BRISQUE GT 19.63 3.68 / / LDCT 11.76 1.71 36.69 58.12 Ablation model 20.06 4.18 11.68 53.92 SwinCBD 24.34 4.66 8.62 45.54
Tools
Get Citation
Copy Citation Text
Ao Wang, Zhen Peng, Jun Wang, Fen Tao, Ling Zhang, Guohao Du, Yi Liu, Biao Deng. Low-dose nano-CT reconstruction image denoising based on Swin Transformer and convolutional neural network[J]. Opto-Electronic Engineering, 2025, 52(6): 240306
Paper Information
Category: Article
Received: Dec. 25, 2024
Accepted: Apr. 28, 2025
Published Online: Sep. 3, 2025
The Author Email: Yi Liu (刘一), Biao Deng (邓彪)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20