Virtually stained H&E images and nuclei segmentation combining neural networks and spectral phasor analysis

H&E stained sections are the gold standard for disease diagnosis but, unfortunately, the staining process is time-consuming and expensive. In an effort to overcome these problems, here, we propose a virtual staining algorithm, able to predict an Hematoxylin/Eosin (H&E) image, usually exploited during clinical evaluations, starting from the autofluorescence signal of entire liver tissue sections acquired by a confocal microscope. The color and texture contents of the generated virtually stained images have been analyzed through the phasor-based approach to detect tumorous tissue and to segment relevant biological structures (accuracy>90% compared to the expert manual analysis).