HcGAN: Harmonic conditional generative adversarial network for efficiently generating high-quality IHC images from...

Area of Science:

• Digital pathology
• Computational imaging
• Artificial intelligence in medicine

Background:

• High-quality histopathology images, such as immunohistochemistry (IHC), are vital for accurate cancer diagnosis and the development of computer-aided diagnostic (CAD) systems.
• Laboratory production of IHC images is costly and time-consuming.
• Current AI-based IHC image generation methods face limitations due to cellular complexity, staining variability, and overfitting.

Purpose of the Study:

• To propose a novel technique, the harmonic conditional generative adversarial network (HcGAN), for generating high-quality IHC-stained images.
• To leverage widely available hematoxylin and eosin (H&E) images as input for generating realistic IHC images.
• To enhance the visual quality and reduce overfitting in generated IHC images.

Main Methods:

• The proposed HcGAN model utilizes generative adversarial learning with generator and discriminator networks.
• Harmonic convolution, based on discrete cosine transform filter banks, is employed in both generator and discriminator networks, replacing standard convolution.
• The model is trained using hematoxylin and eosin (H&E) images containing underlying cellular and morphological structures of various cancer tissues.

Main Results:

• The HcGAN model successfully generates high-quality IHC images that mimic real stained images, effectively highlighting positive cells.
• Qualitative and quantitative analyses demonstrate superior performance of HcGAN compared to state-of-the-art methods.
• The use of harmonic convolution improved visual quality and mitigated overfitting issues.

Conclusions:

• The HcGAN technique offers a significant advancement in generating realistic and high-quality IHC images from H&E stains.
• This method provides a cost-effective and efficient alternative to traditional laboratory-based IHC image production.
• The proposed approach has the potential to greatly benefit the field of digital pathology and computer-aided diagnosis.