Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Stain Style Transfer using Transitive Adversarial Networks

Shaojin Cai, Yuyang Xue3 Qinquan Gao, Min Du, Gang Chen, Hejun Zhang, Tong Tong

TL;DR

This paper tackles color variation in digitized pathology slides by introducing Transitive Adversarial Networks (TAN) for stain style transfer in an unpaired setting. It proposes a novel generator, Trans-Net, and a PatchGAN discriminator within a CycleGAN-inspired framework, combining adversarial and cycle-consistency losses $L_{adv}$ and $L_{cyc}$ to achieve faithful color transfer while preserving tissue structure. Across a MITOS-ATYPIA14-based dataset, TAN with Trans-Net outperforms StainGAN and other generators in PSNR/SSIM and significantly reduces processing time, demonstrating practical viability for cross-center stain normalization. The work highlights architectural choices—deeper, information-propagating downsampling/upsampling and fewer convolutional layers—that yield higher-quality color/texture transfer and faster inference, with potential downstream benefits for pathology diagnosis and CAD systems.

Abstract

Digitized pathological diagnosis has been in increasing demand recently. It is well known that color information is critical to the automatic and visual analysis of pathological slides. However, the color variations due to various factors not only have negative impact on pathologist's diagnosis, but also will reduce the robustness of the algorithms. The factors that cause the color differences are not only in the process of making the slices, but also in the process of digitization. Different strategies have been proposed to alleviate the color variations. Most of such techniques rely on collecting color statistics to perform color matching across images and highly dependent on a reference template slide. Since the pathological slides between hospitals are usually unpaired, these methods do not yield good matching results. In this work, we propose a novel network that we refer to as Transitive Adversarial Networks (TAN) to transfer the color information among slides from different hospitals or centers. It is not necessary for an expert to pick a representative reference slide in the proposed TAN method. We compare the proposed method with the state-of-the-art methods quantitatively and qualitatively. Compared with the state-of-the-art methods, our method yields an improvement of 0.87dB in terms of PSNR, demonstrating the effectiveness of the proposed TAN method in stain style transfer.

Stain Style Transfer using Transitive Adversarial Networks

TL;DR

This paper tackles color variation in digitized pathology slides by introducing Transitive Adversarial Networks (TAN) for stain style transfer in an unpaired setting. It proposes a novel generator, Trans-Net, and a PatchGAN discriminator within a CycleGAN-inspired framework, combining adversarial and cycle-consistency losses and to achieve faithful color transfer while preserving tissue structure. Across a MITOS-ATYPIA14-based dataset, TAN with Trans-Net outperforms StainGAN and other generators in PSNR/SSIM and significantly reduces processing time, demonstrating practical viability for cross-center stain normalization. The work highlights architectural choices—deeper, information-propagating downsampling/upsampling and fewer convolutional layers—that yield higher-quality color/texture transfer and faster inference, with potential downstream benefits for pathology diagnosis and CAD systems.

Abstract

Digitized pathological diagnosis has been in increasing demand recently. It is well known that color information is critical to the automatic and visual analysis of pathological slides. However, the color variations due to various factors not only have negative impact on pathologist's diagnosis, but also will reduce the robustness of the algorithms. The factors that cause the color differences are not only in the process of making the slices, but also in the process of digitization. Different strategies have been proposed to alleviate the color variations. Most of such techniques rely on collecting color statistics to perform color matching across images and highly dependent on a reference template slide. Since the pathological slides between hospitals are usually unpaired, these methods do not yield good matching results. In this work, we propose a novel network that we refer to as Transitive Adversarial Networks (TAN) to transfer the color information among slides from different hospitals or centers. It is not necessary for an expert to pick a representative reference slide in the proposed TAN method. We compare the proposed method with the state-of-the-art methods quantitatively and qualitatively. Compared with the state-of-the-art methods, our method yields an improvement of 0.87dB in terms of PSNR, demonstrating the effectiveness of the proposed TAN method in stain style transfer.

Paper Structure

This paper contains 15 sections, 4 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. The Framework
  4. Network Architectures
  5. Generator
  6. Discriminator
  7. Experiments and Results
  8. Dateset and Details
  9. Dataset
  10. Training details
  11. Evaluation Metrics
  12. Results with different levels of downsampling
  13. Comparisons of results using different generators
  14. Comparison with state-of-the-art method
  15. Discussion and Conclusion

Figures (4)

  • Figure 1: The proposed framework for stain style transfer. $x$ and $y$ are unpaired images randomly sampled from their respective domains.
  • Figure 2: The network of our proposed generator that we refer to as Trans-Net.
  • Figure 3: Visual comparisons of results using different generators.
  • Figure 4: Visual comparison between the result of our proposed method and that of StainGAN.