Polarization-assisted multi-domain virtual histopathological staining of tissue specimens

Polarization imaging exhibits high sensitivity to tissue structures and is now widely used in biomedical studies and clinical practice. In particular, by integrating polarization information provided by Mueller matrix microscopy, generative adversarial networks have demonstrated effective domain-to-domain virtual staining capabilities. In this Letter, we propose a polarization-assisted multi-domain virtual histopathological staining network, Polar-starGAN, which employs tissue information provided by Mueller matrix polarization images for tissue specimen staining. The unsupervised multi-domain staining results on human gastrointestinal tissue slices confirm that Polar-starGAN can perform cross-domain transformation from polarization images acquired from hematoxylin and eosin (H&E)-stained slides to three other widely used staining methods. Through a customized unsupervised generative adversarial training strategy, the model generates virtually stained images with high structural consistency and color tones closely matching real slides. This provides an effective auxiliary tool for digital pathology and clinical histopathological diagnosis.