Virtual staining for digital pathology based on Mueller matrix polarimetry

Abstract

In recent years, Mueller matrix polarimetry has demonstrated significant advantages in assisting clinical pathology diagnosis. However, to address the challenge of providing clinicians with intuitive understandings of the structural information for the samples in polarization imaging, it is often necessary to directly transform polarization images into standard pathological stained images for pathologists through virtual staining techniques. In this study, we propose a polarimetric virtual staining method based on Cycle-Consistent Generative Adversarial Networks (CycleGAN) that employs unpaired Mueller matrix polarimetric images and bright-field images to generate standard hematoxylin and eosin (H&E) stained tissue images. In comparison to existing techniques that are primarily based on paired image model training, the proposed method simplifies the process of data acquisition and preprocessing. This preliminary demonstration offers insight into the potential of polarization-assisted digital pathology in clinical applications.