Fig. 1. Cycle generative adversarial network model

Fig. 2. Feature converter residual structure network model. (a) ResNet network model; (b) DenseNet network model

Fig. 3. Generator network training model based on DenseNet

Fig. 4. 5-layer DenseNet network model

Fig. 5. Discriminator network training model

Fig. 6. Algorithm flowchart

Fig. 7. Validation scheme of proposed method

Fig. 8. Denoising results of an image for patient No.48. (a) LDCT; (b) SDCT; (c) ST-NLM processed LDCT; (d) Res-CycleGAN processed LDCT; (e) Dense-CycleGAN processed LDCT

Fig. 9. Automatic calcification score of a computed tomography image. (a) SDCT Agatston score; (b) ST-NLM Agatston score; (c) Res-CycleGAN Agatston score; (d) Dense-CycleGAN Agatston score

Fig. 10. Denoising results of an image for patient No.79. (a) LDCT after adding noise; (b) original image; (c) ST-NLM; (d) Res-CycleGAN; (e) Dense-CycleGAN