Virtual Stain Transfer in Histology via Cascaded Deep Neural Networks

Pathological diagnosis relies on the visual inspection of histologically stained thin tissue specimens, where different types stains are applied to bring contrast and highlight various desired histological features. However, destructive histochemical staining procedures usually irreversible, making it very difficult obtain multiple same section. Here, we demonstrate a virtual stain transfer framework via cascaded deep neural network (C-DNN) digitally transform hematoxylin eosin (H&E) images into other stains. Unlike single structure that only takes one type as input output another type, C-DNN first uses autofluorescence microscopy H&E then performs from domain in manner. This training phase allows model directly exploit histochemically image data both target special interest. advantage alleviates challenge paired acquisition improves quality color accuracy stain. We validated superior performance this approach using kidney needle core biopsy sections successfully transferred PAS (periodic acid-Schiff) method provides high-quality existing, slides creates new opportunities digital pathology by performing highly accurate stain-to-stain transformations.