E-Staining DermaRepo: H&E whole slide image staining dataset.

Area of Science:

• Digital pathology
• Medical imaging
• Artificial intelligence in healthcare

Background:

• Automated disease diagnosis relies on large annotated datasets for AI and machine learning frameworks.
• Timely and accurate diagnosis is crucial for enhancing patient outcomes.
• Current methods often require extensive manual annotation and processing of medical images.

Purpose of the Study:

• To develop a comprehensive dataset of whole slide images for skin biopsy analysis.
• To virtually replicate the Hematoxylin and Eosin (H&E) staining process.
• To enhance the efficiency and accuracy of computer-aided diagnostic pipelines.

Main Methods:

• A repository of unstained skin biopsy images was created using brightfield microscopy.
• Chemically and virtually stained image samples were generated for comparison.
• A Dual Contrastive Generative Adversarial Network (GAN) was trained to synthesize virtually stained images.

Main Results:

• The trained GAN achieved a Fréchet Inception Distance (FID) score of 80.47 between virtually and chemically stained images, indicating high content correlation.
• FID scores of 342.01 and 320.40 were observed between unstained and virtually stained, and unstained and chemically stained images, respectively.
• The results demonstrate the model's capability to generate realistic virtual stains.

Conclusions:

• The developed dataset and virtual staining method can significantly improve the efficiency of disease diagnosis pipelines.
• Virtual staining offers a promising approach to reduce the reliance on manual staining procedures.
• This work contributes to advancing AI-driven diagnostic tools in digital pathology.