Table 6 Computed person correlation coefficients among qualitative evaluations data of virtual stains, real H&E images and HSFI, indicating direction of correlation, followed by the corresponding inference relationship. HSFI , HSFI
From: VISGAB: Virtual staining-driven GAN benchmarking for optimizing skin tissue histology
Qualitative metric | Pearson r | Direction | Inference |
|---|---|---|---|
Stain Specificity | |||
H&E consistency | 0.982 | Strong positive | HSFI increases with improved staining consistency, reflecting close alignment with stain fidelity |
Melanin differentiation | 0.981 | Strong positive | Strong correlation confirms HSFI’s sensitivity to pigment differentiation, crucial for skin histopathology |
Diagnostic trustworthiness | |||
Nuclear atypia | 0.995 | Strong positive | Near-perfect correlation; HSFI strongly reflects nuclear-level diagnostic trustworthiness |
Tissue architecture | 0.882 | Very positive | HSFI correlates tightly with architectural integrity, supporting its structural awareness, but requires further weighting coefficient optimization |
Mitotic figure accuracy | 0.987 | Strong positive | Indicates HSFI is sensitive to cellular-level morphological fidelity |
Artifact severity | |||
No blurring (None rate) | 0.974 | Strong positive | Lower blur correlates with higher HSFI, confirming its structural sharpness sensitivity |
Overstaining (mean) | −0.994 | Strong negative | Strong negative correlation shows HSFI penalizes overstaining consistent with human perception |
Hallucinations (mean) | −0.991 | Strong negative | Strong inverse relationship; HSFI decreases with artifact prevalence, capturing pathological distortions |
Inter-rater agreement (Fleiss’ κ) | 0.973 | Strong positive | High correlation confirms HSFI aligns with expert consensus robustness |
Turing test success | 0.992 | Strong positive | HSFI matches human perceptual success rates, confirming clinical interpretability |
