Fig. 5: Visual and quantitative comparison of virtual staining performance for RegGAN, Pix2Pix, and CycleGAN models.

Each example shows the PARS input image, ground truth chemical stain, and outputs from all three models, along with distributions of quantitative metrics (DISTS and MS-SSIM). The comparison covers all stains used in this study and includes representative disease cases where possible: a H& E-stained human kidney with ccRCC, (b) H& E-stained human skin with nodular melanoma, (c) PAS-stained human skin with fungal infection, (d) Masson’s trichrome-stained human kidney with ccRCC, and (e) JMS-stained healthy mouse kidney.