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

Area of Science:

• Digital pathology
• Histological image analysis
• Medical imaging

Background:

• Traditional histological staining is time-consuming and requires extensive infrastructure.
• Digital staining offers an efficient, low-infrastructure alternative for generating stained images.
• Interpreting non-traditional microscopic images (grayscale, pseudo-color) is challenging for clinicians.

Purpose of the Study:

• To develop a novel network for digital staining, specifically transforming Reflectance Confocal Microscopy (RCM) images into Hematoxylin and Eosin (H&E) stained images.
• To address the challenge of interpreting non-traditional microscopic images for pathologists and surgeons.

Main Methods:

• Developed In-and-Out Net, a Generative Adversarial Network (GAN) model.
• Utilized aluminum chloride preprocessing to enhance nuclei contrast in RCM images of skin tissue.
• Trained the model with digital H&E labels from two fluorescence channels, ensuring pixel-level ground truth without image registration.

Main Results:

• The In-and-Out Net model efficiently transformed RCM images into H&E stained images.
• Achieved state-of-the-art performance in digital staining tasks, validated through comparative analysis and ablation studies.
• Successfully generated perfectly matched input and ground truth images without the need for registration.

Conclusions:

• In-and-Out Net provides a valuable tool for digital staining, enhancing histological image analysis.
• The proposed method streamlines tissue analysis by generating realistic H&E stains from RCM images.
• This advancement facilitates quicker and more accessible tissue interpretation in digital pathology.