Table 1.
Overview of GAN-based architectures for medical image augmentation, including hybrid status of architectures (if applicable), indicating used combinations of VAEs, GANs, and DMs.
| Reference | Architecture | Hybrid Status | Dataset | Modality | 3D | Eval. Metrics |
|---|---|---|---|---|---|---|
| Classification | ||||||
| [42] | DCGAN, ACGAN | Private | CT | Sens., Spec. | ||
| [41] | DCGAN, WGAN | BraTS2016 | MR | Acc. | ||
| [49] | PGGAN, MUNIT | BraTS2016 | MR | ✓ | Acc., Sens., Spec., | |
| [50] | AE-GAN | Hybrid (V + G) | BraTS2018, ADNI | MR | ✓ | MMD, MS-SSIM |
| [51] | ICW-GAN | OpenfMRI, HCP | MR | ✓ | Acc., Prec., F1 | |
| NeuroSpin, IBC | Recall | |||||
| [52] | ACGAN | IEEE CCX | X-ray | Acc., Sens., Spec. | ||
| Prec., Recall, F1 | ||||||
| [53] | PGGAN | BraTS2016 | MR | Acc., Sens., Spec. | ||
| [54] | ANT-GAN | BraTS2018 | MR | Acc. | ||
| [55] | MG-CGAN | LIDC-IDRI | CT | Acc., F1 | ||
| [56] | FC-GAN | Hybrid (V + G) | ADHD, ABIDE | MR | Acc., Sens., Spec., AUC | |
| [57] | TGAN | Private | Ultrasound | Acc., Sens., Spec. | ||
| [58] | AAE | Private | MR | Prec., Recall, F1 | ||
| [59] | DCGAN, InfillingGAN | DDSM | CT | LPIPS, Recall | ||
| [60] | SAGAN | COVID-CT, SARS-COV2 | CT | Acc. | ||
| [61] | StyleGAN | Private | MR | - | ||
| [62] | DCGAN | PPMI | MR | Acc., Spec., Sens. | ||
| [63] | TMP-GAN | CBIS-DDMS, Private | CT | Prec., Recall, F1, AUC | ||
| [64] | VAE-GAN | Hybrid (V + G) | Private | MR | Acc., Sens., Spec. | |
| [65] | CounterSynth | UK Biobank, OASIS | MR | ✓ | Acc., MSE, SSIM, MAE | |
| Segmentation | ||||||
| [43] | CGAN | DRIVE | Fundus photography | KLD, F1 | ||
| [66] | DCGAN | SCR | X-ray | Dice, Hausdorff | ||
| [67] | CB-GAN | BraTS2015 | MR | Dice, Prec., Sens. | ||
| [68] | Pix2Pix | BraTS2015, ADNI | MR | ✓ | Dice | |
| [12] | CycleGAN | NIHPCT | CT | Dice | ||
| [69] | CM-GAN | Private | MR | KLD, Dice | ||
| hausdorff | ||||||
| [70] | CGAN | COVID-CT | CT | FID, PSNR, SSIM, RMSE | ||
| [71] | Red-GAN | BraTS2015, ISIC | MR | Dice | ||
| [44] | Pix2Pix, SPADE, CycleGAN | Private | MR | Dice | ||
| [72] | StyleGAN | LIDC-IDRI | CT | Dice, Pres., Sens. | ||
| [73] | DCGAN, GatedConv | Private | X-ray | MAE, PSNR, SSIM, FID, AUC | ||
| Cross-modal translation | ||||||
| [74] | CycleGAN | Private | MR ↔ CT | ✓ | Dice | |
| [75] | CycleGAN | Private | MR → CT | MAE, PSNR | ||
| [76] | Pix2Pix | ADNI, Private | MR → CT | ✓ | MAE, PSNR, Dice | |
| [77] | MedGAN | Private | PET → CT | SSIM, PSNR, MSE | ||
| VIF, UQI, LPIPS | ||||||
| [47] | pGAN, CGAN | BraTS2015, MIDAS, IXI | T1 ⟷ T2 | SSIM, PSNR | ||
| [69] | CM-GAN | Private | MR | KLD, Dice | ||
| hausdorff | ||||||
| [46] | mustGAN | IXI, ISLES | T1 ↔ T2 ↔ PD | SSIM, PSNR | ||
| [78] | CAE-ACGAN | Hybrid (V + G) | Private | CT → MR | ✓ | PSNR, SSIM, MAE |
| [79] | GLA-GAN | ADNI | MR → PET | SSIM, PSNR, MAE | ||
| Acc., F1 | ||||||
| Other | ||||||
| [80] | VAE-CGAN | Hybrid (V + G) | ACDC | MR | ✓ | - |
Note: V = variational autoencoders, G = generative adversarial networks.