Table 1 A summary of the related work on various virtual staining and image to image translation models, showing evolution and breakthrough in GANs from 2009–2024.
From: VISGAB: Virtual staining-driven GAN benchmarking for optimizing skin tissue histology
Method | Year | Domain | Key Aspects | Limitations |
|---|---|---|---|---|
Stain normalization19 | 2009 | Histopathology | Standardizes H&E stain appearance for quantitative analysis | Sensitive to staining variability and noise |
Pix2Pix4 | 2017 | Image synthesis | Paired image-to-image translation for stain synthesis | Requires aligned datasets, often produce localized artifacts when applied to complex tissue structures |
CycleGAN5 | 2017 | Virtual staining | Unpaired stain translation using cyclicconsistency loss | Struggles with unrealistic textures or color shifts, without strong alignment constraints |
Deep learning20 | 2017 | Survey medical imaging | Highlights gaps in virtual staining research | No GANs benchmarking or metrics proposal |
Deep learning6 | 2019 | Digital staining | Pioneered virtual H&E staining from auto fluorescence | Focused on paired data, limited clinical validation |
CUTGAN7 | 2021 | Digital pathology | Contrastive learning for local details preservation | Lacks explicit global context, leading to artifacts or subtle distortions in morphological patterns |
DCLGAN8 | 2024 | Virtual staining | Dual contrastive loss for stain and texture fidelity | Localized receptive fields do not enforce coherence across the entire tissue image, moderate artifacts |
