Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-target stain normalization for histology slides

Desislav Ivanov, Carlo Alberto Barbano, Marco Grangetto

TL;DR

Stain variability across histology slides can impair downstream AI tasks. This paper proposes a parameter-free multi-target stain normalization method that extends the Macenko framework by aggregating stain information from multiple reference images. Through comparative analysis of Stochastic, Concat, Avg-pre, and Avg-post formulations, Concat and especially Avg-post are shown to improve segmentation performance and generalization to external data. The approach is simple to integrate into existing pipelines and holds practical potential for enhancing robustness in computational pathology tasks like automatic nuclei segmentation.

Abstract

Traditional staining normalization approaches, e.g. Macenko, typically rely on the choice of a single representative reference image, which may not adequately account for the diverse staining patterns of datasets collected in practical scenarios. In this study, we introduce a novel approach that leverages multiple reference images to enhance robustness against stain variation. Our method is parameter-free and can be adopted in existing computational pathology pipelines with no significant changes. We evaluate the effectiveness of our method through experiments using a deep-learning pipeline for automatic nuclei segmentation on colorectal images. Our results show that by leveraging multiple reference images, better results can be achieved when generalizing to external data, where the staining can widely differ from the training set.

Multi-target stain normalization for histology slides

TL;DR

Stain variability across histology slides can impair downstream AI tasks. This paper proposes a parameter-free multi-target stain normalization method that extends the Macenko framework by aggregating stain information from multiple reference images. Through comparative analysis of Stochastic, Concat, Avg-pre, and Avg-post formulations, Concat and especially Avg-post are shown to improve segmentation performance and generalization to external data. The approach is simple to integrate into existing pipelines and holds practical potential for enhancing robustness in computational pathology tasks like automatic nuclei segmentation.

Abstract

Traditional staining normalization approaches, e.g. Macenko, typically rely on the choice of a single representative reference image, which may not adequately account for the diverse staining patterns of datasets collected in practical scenarios. In this study, we introduce a novel approach that leverages multiple reference images to enhance robustness against stain variation. Our method is parameter-free and can be adopted in existing computational pathology pipelines with no significant changes. We evaluate the effectiveness of our method through experiments using a deep-learning pipeline for automatic nuclei segmentation on colorectal images. Our results show that by leveraging multiple reference images, better results can be achieved when generalizing to external data, where the staining can widely differ from the training set.
Paper Structure (13 sections, 4 equations, 3 figures, 1 table)

This paper contains 13 sections, 4 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Experimental data
  4. Proposed method
  5. Background
  6. Multi-target normalization
  7. Results
  8. Experimental setup
  9. Evaluation on same dataset
  10. Generalization to external data
  11. Software Implementation
  12. Related Works
  13. Conclusions

Figures (3)

  • Figure 1: Multi-target normalization is more robust to stain variation than traditional approaches that consider only one target image. In the figure, a given input image is normalized with different targets (top row) chosen from different sources, using the method in macenko. Multi-target (bottom left) employs all of the reference images.
  • Figure 2: Comparison of the mIoU (y-axis) between different normalization methods based on the size of the subset of reference images (x-axis). Training and testing on the whole Lizard dataset (a) Generalization to external data w/o normalization during training (b) Generalization to external data w/ Macenko normalization during training (c).
  • Figure 3: Visual comparison of the different multi-target approaches.