Deep learning-based virtual special staining of H&E stained tissue sections

Hematoxylin and eosin (H&E) staining, which is standardly applied to tissues in histopathological diagnosis, is an incredible tissue staining method that reveals the morphological features of tissues and cells and gives an idea about their biomolecular structures. However, a more accurate determination of tissue and cell structure and biomolecular components is essential for the diagnosis of diseases under the light microscope. For this reason, histochemical, immunohistochemical, immunofluorescent and genetic techniques have been applied to tissues and cells and it has been aimed to reach the closest diagnosis to the truth. However, the number of these methods is quite high and depending on the method, they can be very complex and time-consuming in terms of procurement, cost, time requirement and application. Manual preparation of each is an expensive and labor-intensive process that requires complex and difficult methods of using many chemicals, and ready-to-use kits are often costly. Here, we propose a novel virtual staining tool that transforms H&E-stained tissue images into specially stained versions in just a few minutes. Virtual staining of some of the frequently used methods in daily pathology practice with a transformation that is derived through a learning process, using H&E slide as a basis, will make the use of these methods in the histopathological diagnosis process incredibly practical, inexpensive, simple and easy to apply in a short time. As well as, we propose a novel StainKid dataset of stomach and kidney tissue samples stained with a wide collection of histological stains. The StainKid dataset can make a significant contribution to the development of computer aided diagnosis in histopathology by paving the way for new artificial intelligence-based virtual staining techniques.