Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ReStainGAN: Leveraging IHC to IF Stain Domain Translation for in-silico Data Generation

Dominik Winter, Nicolas Triltsch, Philipp Plewa, Marco Rosati, Thomas Padel, Ross Hill, Markus Schick, Nicolas Brieu

TL;DR

ReStainGAN introduces a CycleGAN-based framework to disentangle IHC stain components by leveraging an auxiliary IF domain, enabling restaining operations that generate in-silico nuclear-marker IHC images while preserving morphology. The method learns bijective mappings between IHC and IF domains via generators $\mathcal{G}_{AB}$ and $\mathcal{G}_{BA}$ and manipulates stain channels with a restaining function $\kappa_{\alpha}$ controlled by $\alpha_{hh},\alpha_{dh},\alpha_{hd},\alpha_{dd}$, imposing $[\alpha_{hd},\alpha_{dd}]=0$. Evaluations on nucleus segmentation show significant gains when training StarDist on the synthetic data compared to baselines, e.g., F1 up to $0.848$ vs $0.604$–$0.697$. This work suggests a scalable path to expand annotated datasets across staining modalities, reducing labeling costs in computational pathology and enabling broader downstream analyses such as epithelium segmentation.

Abstract

The creation of in-silico datasets can expand the utility of existing annotations to new domains with different staining patterns in computational pathology. As such, it has the potential to significantly lower the cost associated with building large and pixel precise datasets needed to train supervised deep learning models. We propose a novel approach for the generation of in-silico immunohistochemistry (IHC) images by disentangling morphology specific IHC stains into separate image channels in immunofluorescence (IF) images. The proposed approach qualitatively and quantitatively outperforms baseline methods as proven by training nucleus segmentation models on the created in-silico datasets.

ReStainGAN: Leveraging IHC to IF Stain Domain Translation for in-silico Data Generation

TL;DR

ReStainGAN introduces a CycleGAN-based framework to disentangle IHC stain components by leveraging an auxiliary IF domain, enabling restaining operations that generate in-silico nuclear-marker IHC images while preserving morphology. The method learns bijective mappings between IHC and IF domains via generators and and manipulates stain channels with a restaining function controlled by , imposing . Evaluations on nucleus segmentation show significant gains when training StarDist on the synthetic data compared to baselines, e.g., F1 up to vs –. This work suggests a scalable path to expand annotated datasets across staining modalities, reducing labeling costs in computational pathology and enabling broader downstream analyses such as epithelium segmentation.

Abstract

The creation of in-silico datasets can expand the utility of existing annotations to new domains with different staining patterns in computational pathology. As such, it has the potential to significantly lower the cost associated with building large and pixel precise datasets needed to train supervised deep learning models. We propose a novel approach for the generation of in-silico immunohistochemistry (IHC) images by disentangling morphology specific IHC stains into separate image channels in immunofluorescence (IF) images. The proposed approach qualitatively and quantitatively outperforms baseline methods as proven by training nucleus segmentation models on the created in-silico datasets.
Paper Structure (4 sections, 2 equations, 1 figure, 1 table)

This paper contains 4 sections, 2 equations, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. Results
  4. Discussion

Figures (1)

  • Figure 1: a$)$ ReStainGAN disentangles nuclear and cell membrane representations in IHC cell membrane marker images. Manipulating these representations with $\alpha_{hh}$, $\alpha_{dh}$ while setting $[\alpha_{hd}, \alpha_{dd}] = 0$ yields various in-silico IHC nuclear marker stained images. b$)$ The StarDist nucleus segmentation model performs best trained on in-silico data created with the proposed ReStainGAN (right).