Table 4 Computed person correlation coefficients among quantitative evaluations data of virtual stains, real H&E images and HSFI, indicating direction of correlation, followed by the corresponding inference relationship.
From: VISGAB: Virtual staining-driven GAN benchmarking for optimizing skin tissue histology
Comparison | Pearson r | Direction | Inference |
|---|---|---|---|
HSFI vs. SSIM | + 0.982 | Strong positive | HSFI strongly aligns with structural similarity, confirming that higher HSFI corresponds to higher structural fidelity |
HSFI vs. PSNR | + 0.967 | Strong positive | HSFI rises almost linearly with PSNR, indicating sensitivity to reconstruction quality |
HSFI vs. FID | −0.956 | Strong negative | Strong inverse relationship: lower FID (better realism) corresponds to higher HSFI |
HSFI vs. KID | −0.931 | Strong negative | HSFI decreases as KID increases, reflecting alignment with perceptual domain similarity |
HSFI vs. LPIPS | −0.942 | Strong negative | Higher perceptual dissimilarity (LPIPS) results in lower HSFI des, indicating a desirable correlation |
