GCC-GAN and GSN-GAN script and dataset

This repository contains the scripts used to develop the GCC-GAN and GSN-GAN models for color normalization in the work:
- Salvi M., Branciforti F., Molinari F., and Meiburger K. M. , "Generative models for color normalization in digital pathology and dermatology: advancing the learning paradigm", Expert Systems with Applications 2024, DOI: 10.1016/j.eswa.2023.123105

Abstract:
Color medical images introduce an additional confounding factor when compared to conventional grayscale medical images: color variability. This variability alludes to potential critical issues of inconsistent evaluation by clinicians and the misinterpretation or suboptimal learning process of automatic quantitative algorithms. To mitigate the potential negative ramifications of color variability, various color normalization strategies have been introduced and have proven to be a powerful tool for standardizing image appearance. Here we present a novel paradigm for the color normalization process that is based on generative adversarial networks (GANs) that can standardize images of digital pathology (stain normalization) and dermatology (color constancy), two foundational research fields in which the acquired images consistently present high color variability. Specifically, we formulate the color normalization task as an image-to-image translation problem, maintaining a pixel-to-pixel correspondence between the original and normalized images. Our approach outperforms existing state of the art methods in both the digital pathology and dermatology clinical fields. Extensive validation was conducted on public datasets to evaluate the effectiveness of color normalization results on completely external test sets. Our framework generalized well on unseen data, showing how it can be effectively included in the pipeline of automatic quantitative algorithms for reducing color variability and hence enhancing final segmentation and/or classification performance. Finally, we publicly release the source code of our models to encourage open science.