Laser & Optoelectronics Progress, Volume. 59, Issue 2, 0210008(2022)
LDCT Denoising Method Based on Dual Attention Mechanism and Compound Loss
Fig. 1. Dual attention module. (1) Channel attention module; (2) spatial attention module
Fig. 2. Channel attention module
Fig. 3. Spatial attention module
Fig. 4. Network structure model
Fig. 5. Preprocessing module
Fig. 6. CT images of abdomen. (a) Test fig.1; (b) test fig.2
Fig. 7. LDCT and denoising effect of different algorithms of test Fig. 1. (a) LDCT; (b) BM3D; (c) K-SVD; (d) RED-CNN; (e) WGAN-VGG; (f) CycleGAN; (g) proposed algorithm; (h) NDCT
Fig. 8. Partial enlarged view of ROI in Figs.7. (a) LDCT; (b) BM3D; (c) K-SVD; (d) RED-CNN; (e) WGAN-VGG; (f) CycleGAN; (g) proposed algorithm; (h) NDCT
Fig. 9. Noise after denoising by LDCT and different algorithms in Figs. 7. (a) LDCT; (b) BM3D; (c) K-SVD; (d) RED-CNN; (e) WGAN-VGG; (f) CycleGAN; (g) proposed algorithm
Fig. 10. LDCT and denoising effect of different algorithms of test Fig. 2.(a) LDCT; (b) BM3D; (c) K-SVD; (d) RED-CNN; (e) WGAN-VGG; (f) CycleGAN; (g) proposed algorithm; (h) NDCT
Fig. 11. Partial enlarged view of ROI in Figs.10. (a) LDCT; (b) BM3D; (c) K-SVD; (d) RED-CNN; (e) WGAN-VGG; (f) CycleGAN; (g) proposed algorithm ; (h) NDCT
Fig. 12. Noise after denoising by LDCT and different algorithms in Figs.10. (a) LDCT;(b) BM3D; (c) K-SVD; (d) RED-CNN; (e) WGAN-VGG; (f) CycleGAN; (g) proposed algorithm
Fig. 13. Performance indicators of ROI area in Fig. 7 and Fig. 10. (a) PSNR indicator; (b) SSIM indicator
|
|
Get Citation
Copy Citation Text
Zhitao Guo, Yi Su, Jinli Yuan, Linlin Zhao. LDCT Denoising Method Based on Dual Attention Mechanism and Compound Loss[J]. Laser & Optoelectronics Progress, 2022, 59(2): 0210008
Category: Image Processing
Received: Dec. 23, 2020
Accepted: Mar. 11, 2021
Published Online: Dec. 23, 2021
The Author Email: Yuan Jinli (jinli_yuan@hebut.edu.cn)