Skip to main content
. 2021 Mar 23;12(4):2280–2298. doi: 10.1364/BOE.415962

Fig. 4.

Fig. 4.

Bad quality predictions of patches from the training dataset. Columns (A) and (B) visualize NLM patches and contrast inverted NLM patches, column (C) shows histopathologically stained H&E patch, and columns (D) and (E) visualize computationally stained H&E patch generated by the Pix2Pix model and the cycle CGAN model, respectively. The scale bar represents 50 μm. Here the patches generated by the cycle CGAN model show a promising translation of corresponding NLM patches; however, the colors in stroma regions (marked by green arrows) and nuclei signals (marked by cyan arrows) are not well represented. The computationally stained H&E patches generated by the Pix2Pix model have a low spatial resolution, as the structures within the crypts (pointed by white arrows) are not visible.