Pathological Section Staining Transferring with Tailored Metric-based Model Selection

As the important pathological section staining, Immunohistochemistry (IHC) staining uses labeled antibodies to highlight specific antigens, providing clearer results for malignancy identification compared with Hematoxylin and Eosin (H&E) staining. However, obtaining IHC manually is labor-intensive and costly. In this study, we attempt to automatically generate IHC staining image by style transferring from H&E, which is easy to get. Due to the unalignment between single evaluation metric and real generation quality, we propose a tailored metric-based model selection (TMMS) method for pathological section staining transferring (PSST). Our method can fuse multiple traditional metrics and multiple relevant losses to measure the quality of generated IHC and select a best-performed model. Extensive experiments and analysis demonstrate the effectiveness of our method TMMS.