Single color digital H&E staining with In-and-Out Net.

Area of Science:

• Digital pathology
• Computational imaging
• Histology

Background:

• Virtual staining offers an efficient alternative to traditional chemical staining methods in microscopy.
• Interpreting unstained or pseudo-colored microscopic images poses challenges for pathologists and surgeons.
• Simulating histological stains digitally can bridge the gap between virtual and conventional imaging.

Purpose of the Study:

• To introduce a novel network, In-and-Out Net, for virtual staining.
• To efficiently transform Reflectance Confocal Microscopy (RCM) images into Hematoxylin and Eosin (H&E) stained images.
• To provide a valuable tool for histological image analysis.

Main Methods:

• Developed a Generative Adversarial Network (GAN) based model named In-and-Out Net.
• Applied aluminum chloride preprocessing to enhance nuclei contrast in RCM images for skin tissues.
• Utilized virtual H&E labels with two fluorescence channels for training, eliminating the need for image registration and providing pixel-level ground truth.

Main Results:

• The In-and-Out Net model successfully transformed RCM images into H&E stained images.
• The model demonstrated state-of-the-art performance in virtual staining tasks.
• An optimal training strategy was proposed and validated through an ablation study.

Conclusions:

• In-and-Out Net provides a promising solution for virtual staining, enhancing histological image analysis.
• The method facilitates efficient tissue analysis without physical sectioning or complex infrastructure.
• Perfectly matched input and ground truth images were collected without registration, simplifying the workflow.