Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Pathological Semantics-Preserving Learning for H&E-to-IHC Virtual Staining

Fuqiang Chen, Ranran Zhang, Boyun Zheng, Yiwen Sun, Jiahui He, Wenjian Qin

TL;DR

This work tackles H&E-to-IHC virtual staining by preserving molecular-level semantics and mitigating spatial misalignment between input and label. It introduces two strategies: PALS, which uses a Focal Optical Density map to quantify protein expression in the DAB channel and enforces MLPA-based alignment, and PCLS, which employs cross-image prototypical consistency to stabilize semantic interaction across misaligned pairs. On the BCI and MIST datasets, PSPStain outperforms state-of-the-art methods in pathological relevance metrics (lower $\text{mIOD}$ and $\text{IOD}$, higher $\text{Pearson-}R$) while maintaining image quality, with ablation confirming the value of the FOD and MLPA components. The approach enables clinically meaningful IHC reconstruction without additional annotations and can extend to other DAB-stained IHC applications.

Abstract

Conventional hematoxylin-eosin (H&E) staining is limited to revealing cell morphology and distribution, whereas immunohistochemical (IHC) staining provides precise and specific visualization of protein activation at the molecular level. Virtual staining technology has emerged as a solution for highly efficient IHC examination, which directly transforms H&E-stained images to IHC-stained images. However, virtual staining is challenged by the insufficient mining of pathological semantics and the spatial misalignment of pathological semantics. To address these issues, we propose the Pathological Semantics-Preserving Learning method for Virtual Staining (PSPStain), which directly incorporates the molecular-level semantic information and enhances semantics interaction despite any spatial inconsistency. Specifically, PSPStain comprises two novel learning strategies: 1) Protein-Aware Learning Strategy (PALS) with Focal Optical Density (FOD) map maintains the coherence of protein expression level, which represents molecular-level semantic information; 2) Prototype-Consistent Learning Strategy (PCLS), which enhances cross-image semantic interaction by prototypical consistency learning. We evaluate PSPStain on two public datasets using five metrics: three clinically relevant metrics and two for image quality. Extensive experiments indicate that PSPStain outperforms current state-of-the-art H&E-to-IHC virtual staining methods and demonstrates a high pathological correlation between the staging of real and virtual stains.

Pathological Semantics-Preserving Learning for H&E-to-IHC Virtual Staining

TL;DR

This work tackles H&E-to-IHC virtual staining by preserving molecular-level semantics and mitigating spatial misalignment between input and label. It introduces two strategies: PALS, which uses a Focal Optical Density map to quantify protein expression in the DAB channel and enforces MLPA-based alignment, and PCLS, which employs cross-image prototypical consistency to stabilize semantic interaction across misaligned pairs. On the BCI and MIST datasets, PSPStain outperforms state-of-the-art methods in pathological relevance metrics (lower and , higher ) while maintaining image quality, with ablation confirming the value of the FOD and MLPA components. The approach enables clinically meaningful IHC reconstruction without additional annotations and can extend to other DAB-stained IHC applications.

Abstract

Conventional hematoxylin-eosin (H&E) staining is limited to revealing cell morphology and distribution, whereas immunohistochemical (IHC) staining provides precise and specific visualization of protein activation at the molecular level. Virtual staining technology has emerged as a solution for highly efficient IHC examination, which directly transforms H&E-stained images to IHC-stained images. However, virtual staining is challenged by the insufficient mining of pathological semantics and the spatial misalignment of pathological semantics. To address these issues, we propose the Pathological Semantics-Preserving Learning method for Virtual Staining (PSPStain), which directly incorporates the molecular-level semantic information and enhances semantics interaction despite any spatial inconsistency. Specifically, PSPStain comprises two novel learning strategies: 1) Protein-Aware Learning Strategy (PALS) with Focal Optical Density (FOD) map maintains the coherence of protein expression level, which represents molecular-level semantic information; 2) Prototype-Consistent Learning Strategy (PCLS), which enhances cross-image semantic interaction by prototypical consistency learning. We evaluate PSPStain on two public datasets using five metrics: three clinically relevant metrics and two for image quality. Extensive experiments indicate that PSPStain outperforms current state-of-the-art H&E-to-IHC virtual staining methods and demonstrates a high pathological correlation between the staging of real and virtual stains.
Paper Structure (9 sections, 11 equations, 4 figures, 2 tables)

This paper contains 9 sections, 11 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Protein-Aware Learning Strategy
  4. Prototype-Consistent Learning Strategy
  5. Loss Function of PSPStain.
  6. Experiments
  7. Conclusion
  8. Acknowledgments.
  9. Disclosure of Interests.

Figures (4)

  • Figure 1: Two key problems in H&E-to-IHC virtual staining: (a) Protein expression level varies greatly in each sample from the same-grade WSI, and only preserving grade information rather than protein expression level results in a less robust representation with semantic dispersion. (b) Spatial misalignment between the generated image and label causes the false response of patches with similar semantics.
  • Figure 2: Our proposed framework consists of two newly learning strategies for preserving protein expression consistency and enhancing semantic alignment.
  • Figure 3: Qualitative comparison with samples virtually stained by various methods.
  • Figure 4: The abscissa represents the index number of samples. The ordinate is the accumulated IOD value. (a) (b) refer to the curve on BCI and MIST datasets, respectively. (c) Strategy ablation. (d) FOD factor $\alpha$ ablation. (e) MLPA module ablation