Table 1. Summary of current-generation neural network architectures for OCT denoising.
| Architecture | Num. repeated images for ground truth | Short description | Ref. |
|---|---|---|---|
| U-net | 128 | U-net | [34] |
| SM-GAN | 100 | Speckle-modulated GAN | [11] |
| DRUNET | 75 | Dilated-Residual U-Net | [35,36] |
| DnCNN | 50 | Feed-forward CNN with a perceptually-sensitive loss function | [10] |
| SiameseGAN | 40 | GAN with a siamese twin network | [27] |
| GCDS | 40 | Gated convolution–deconvolution structure | [33] |
| DeSpecNet | 20 | Residue-learning-base deconvolution | [30] |
| Caps-cGAN | 20 | Capsule conditional GAN | [24] |
| cGAN | 20 | Edge-sensitive cGAN | [25] |
| BRUNet | 9 | Branch Residual U-shape Network | [37] |
| CNN-MSE, CNN-WGAN | 6 | MSE and Wasserstein GANs | [28] |
| SRResNet | 1 | Super-resolution residual network | [40] |