Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Prototype-Guided Coarse Annotations Refining Approach for Whole Slide Images

Bingjian Yao, Weiping Lin, Yan He, Zheng Wang, Liangsheng Wang

TL;DR

This work addresses the high cost and variability of obtaining fine-grained annotations for whole-slide images by introducing a prototype-guided refinement framework. It jointly models intra-slide local semantics and inter-slide contextual patterns to build a hierarchy of prototypes (local and global) that guide a pseudo-labeling process, followed by dynamic data sampling and re-finetuning to train a patch classifier. Across Camelyon16, PAIP2019, and PAIP2020, the method outperforms state-of-the-art approaches in accuracy and efficiency, with robust generalization to unseen data. The use of hierarchical prototypes and adaptive training offers a practical, scalable solution for refining coarse annotations in clinical histopathology tasks.

Abstract

The fine-grained annotations in whole slide images (WSIs) show the boundaries of various pathological regions. However, generating such detailed annotation is often costly, whereas the coarse annotations are relatively simpler to produce. Existing methods for refining coarse annotations often rely on extensive training samples or clean datasets, and fail to capture both intra-slide and inter-slide latent sematic patterns, limiting their precision. In this paper, we propose a prototype-guided approach. Specifically, we introduce a local-to-global approach to construct non-redundant representative prototypes by jointly modeling intra-slide local semantics and inter-slide contextual relationships. Then a prototype-guided pseudo-labeling module is proposed for refining coarse annotations. Finally, we employ dynamic data sampling and re-finetuning strategy to train a patch classifier. Extensive experiments on three publicly available WSI datasets, covering lymph, liver, and colorectal cancers, demonstrate that our method significantly outperforms existing state-of-the-art (SOTA) methods. The code will be available.

A Prototype-Guided Coarse Annotations Refining Approach for Whole Slide Images

TL;DR

This work addresses the high cost and variability of obtaining fine-grained annotations for whole-slide images by introducing a prototype-guided refinement framework. It jointly models intra-slide local semantics and inter-slide contextual patterns to build a hierarchy of prototypes (local and global) that guide a pseudo-labeling process, followed by dynamic data sampling and re-finetuning to train a patch classifier. Across Camelyon16, PAIP2019, and PAIP2020, the method outperforms state-of-the-art approaches in accuracy and efficiency, with robust generalization to unseen data. The use of hierarchical prototypes and adaptive training offers a practical, scalable solution for refining coarse annotations in clinical histopathology tasks.

Abstract

The fine-grained annotations in whole slide images (WSIs) show the boundaries of various pathological regions. However, generating such detailed annotation is often costly, whereas the coarse annotations are relatively simpler to produce. Existing methods for refining coarse annotations often rely on extensive training samples or clean datasets, and fail to capture both intra-slide and inter-slide latent sematic patterns, limiting their precision. In this paper, we propose a prototype-guided approach. Specifically, we introduce a local-to-global approach to construct non-redundant representative prototypes by jointly modeling intra-slide local semantics and inter-slide contextual relationships. Then a prototype-guided pseudo-labeling module is proposed for refining coarse annotations. Finally, we employ dynamic data sampling and re-finetuning strategy to train a patch classifier. Extensive experiments on three publicly available WSI datasets, covering lymph, liver, and colorectal cancers, demonstrate that our method significantly outperforms existing state-of-the-art (SOTA) methods. The code will be available.

Paper Structure

This paper contains 11 sections, 4 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Problem Formulation
  4. Overview
  5. Local-to-Global Prototype Extraction
  6. Prototype-Guided Pseudo-Labeling
  7. Dynamic Data Sampling and Re-Finetuning
  8. Results
  9. Experimental Setup
  10. Experimental Results
  11. Conclusion

Figures (3)

  • Figure 1: Overview of our proposed prototype-guided coarse annotations refinement framework. Local-to-global prototypes are extracted from WSI patches and utilized by the pseudo-label generation module to refine coarse annotations. Dynamic data sampling and re-finetuning module is utilized to train the patch classifier.
  • Figure 2: Examples of coarse annotations and refinement. (a,b), (c,d) and (e,f) are from Camelyon16, PAIP2019, and PAIP2020, respectively.
  • Figure 3: (a) Visualization of external validation.(b) Distribution of local prototypes and corresponding global prototypes.The left column represents the ground truth, while the right two columns use different colors to indicate local and global prototypes, corresponding to various tissues or pathological regions.