Auxiliary Cyclegan-Guided Task-Aware Domain Translation from Duplex to Monoplex IHC Images

Generative models enable the translation from a source image domain where readily trained models are available to an unseen target domain. While Cycle Generative Adversarial Networks (CycleGANs) are commonly used for unpaired translation, the associated cycle consistency constrain assumes an invertible mapping between the two domains. However, this assumption does not hold true for the translation between images stained with chromogenic monoplex and duplex immunohistochemistry (IHC) assays. To address this, we introduce of a novel training design which leverages a set of immunofluorescence (IF) images as an auxiliary and unpaired image domain to guide the direct translation from duplex to monoplex IHC. The results, both quantitative and qualitative, on a downstream segmentation task demonstrate the advantages of the proposed method over multiple baseline methods.