Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LSA: Latent Style Augmentation Towards Stain-Agnostic Cervical Cancer Screening

Jiangdong Cai, Haotian Jiang, Zhenrong Shen, Yonghao Li, Honglin Xiong, Lichi Zhang, Qian Wang

TL;DR

This paper tackles domain shifts in WSI-based cervical cancer screening caused by staining variations across scanners. It introduces Latent Style Augmentation (LSA), which combines offline WSAug for WSI-level stain coherence with an online Stain Transformer to perform latent-space style transfer during MIL training, enabling efficient, scalable cross-scanner robustness. Experiments on a multi-scanner cervical cancer dataset show that LSA improves out-of-distribution performance when training on a single scanner, across multiple patch encoders and MIL methods, outperforming patch-level augmentation baselines. The approach offers a practical pathway to stain-robust WSI classification without generating numerous augmented WSIs, with potential extensions to physics-based stain simulations for further realism.

Abstract

The deployment of computer-aided diagnosis systems for cervical cancer screening using whole slide images (WSIs) faces critical challenges due to domain shifts caused by staining variations across different scanners and imaging environments. While existing stain augmentation methods improve patch-level robustness, they fail to scale to WSIs due to two key limitations: (1) inconsistent stain patterns when extending patch operations to gigapixel slides, and (2) prohibitive computational/storage costs from offline processing of augmented WSIs.To address this, we propose Latent Style Augmentation (LSA), a framework that performs efficient, online stain augmentation directly on WSI-level latent features. We first introduce WSAug, a WSI-level stain augmentation method ensuring consistent stain across patches within a WSI. Using offline-augmented WSIs by WSAug, we design and train Stain Transformer, which can simulate targeted style in the latent space, efficiently enhancing the robustness of the WSI-level classifier. We validate our method on a multi-scanner WSI dataset for cervical cancer diagnosis. Despite being trained on data from a single scanner, our approach achieves significant performance improvements on out-of-distribution data from other scanners. Code will be available at https://github.com/caijd2000/LSA.

LSA: Latent Style Augmentation Towards Stain-Agnostic Cervical Cancer Screening

TL;DR

This paper tackles domain shifts in WSI-based cervical cancer screening caused by staining variations across scanners. It introduces Latent Style Augmentation (LSA), which combines offline WSAug for WSI-level stain coherence with an online Stain Transformer to perform latent-space style transfer during MIL training, enabling efficient, scalable cross-scanner robustness. Experiments on a multi-scanner cervical cancer dataset show that LSA improves out-of-distribution performance when training on a single scanner, across multiple patch encoders and MIL methods, outperforming patch-level augmentation baselines. The approach offers a practical pathway to stain-robust WSI classification without generating numerous augmented WSIs, with potential extensions to physics-based stain simulations for further realism.

Abstract

The deployment of computer-aided diagnosis systems for cervical cancer screening using whole slide images (WSIs) faces critical challenges due to domain shifts caused by staining variations across different scanners and imaging environments. While existing stain augmentation methods improve patch-level robustness, they fail to scale to WSIs due to two key limitations: (1) inconsistent stain patterns when extending patch operations to gigapixel slides, and (2) prohibitive computational/storage costs from offline processing of augmented WSIs.To address this, we propose Latent Style Augmentation (LSA), a framework that performs efficient, online stain augmentation directly on WSI-level latent features. We first introduce WSAug, a WSI-level stain augmentation method ensuring consistent stain across patches within a WSI. Using offline-augmented WSIs by WSAug, we design and train Stain Transformer, which can simulate targeted style in the latent space, efficiently enhancing the robustness of the WSI-level classifier. We validate our method on a multi-scanner WSI dataset for cervical cancer diagnosis. Despite being trained on data from a single scanner, our approach achieves significant performance improvements on out-of-distribution data from other scanners. Code will be available at https://github.com/caijd2000/LSA.

Paper Structure

This paper contains 14 sections, 4 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Overview
  4. WSI-level Stain Augmentation
  5. Architecture and Training of Stain Transformer
  6. Latent Style Augmentation
  7. Experiments
  8. Datasets and Experimental Setup
  9. Comparison with Other MIL Methods
  10. Comparison with Other Augmentation Methods
  11. Influence of Augmented Data Size
  12. Effectiveness of Stain Transformer
  13. Influence of Augmentation Possibility
  14. Conclusion

Figures (2)

  • Figure 1: Overview of method. (a) Training with only WSIs of the source scanner limits the classifier’s ability to handle unseen stain types. (b) Offline WSI-level data augmentation broaders the capability but cannot cover the entire spectrum of stain styles. (c) Online latent style augmentation can maximize the coverage of stain styles.
  • Figure 2: Implementation details. (a) The process of WSI-level stain augmentation (WSAug). (b) Training of Stain Transformer. (c) Latent Style Augmentation during training of MIL methods.