Preservation of Image Content in Stain-to-stain Translation for Digital Pathology

In digital pathology, unsupervised domain adaptation of differently
stained whole-slide images (WSIs) through image-to-image translation has be-
come increasingly important for various applications such as stain augmentation
or for the stain-independent application of deep learning models. In previous
work, different variants of generative adversarial networks (GANs) were pro-
posed to translate a real WSI obtained in the staining domain A into a fake WSI
in the target staining domain B. However, GANs perform unpaired image-to-
image translation and do not enforce consistency with respect to image content,
whichlimitstheirapplicabilityindigitalpathologysettings.Inthispaper,wefirst
investigate the tissue inconsistency problem in such a stain-to-stain translation
scenario using a quantitative evaluation of the distortion between real and fake
imagesindifferentdomains.Then,weinvestigatetwopossiblesolutions,namely
(1) stain colorization inspired by natural image colorization, and (2) a modified
Cycle-GAN, where an intensity invariant loss is proposed to balance the tissue
consistencyacrossstainingdomains.Ourresultshighlightthesuperiorityofthese
methodscomparedtoconventionalunpairedstaintranslationsolutionsfortypical
stainingprotocolsin digital pathology.