Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Boundary-aware Contrastive Learning for Semi-supervised Nuclei Instance Segmentation

Ye Zhang, Ziyue Wang, Yifeng Wang, Hao Bian, Linghan Cai, Hengrui Li, Lingbo Zhang, Yongbing Zhang

TL;DR

This work tackles semi-supervised nuclei instance segmentation in pathology, where boundary noise arises from subtle nucleus-tissue contrast and diverse nuclear morphology. It introduces Boundary-aware Contrastive Learning for Semi-supervised nuclei segmentation (BASS), comprising a low-resolution denoising (LRD) head to smooth boundaries and a cross-RoI contrastive learning (CRC) module to enhance boundary feature discrimination within a teacher-student framework. LRD provides coarse boundary denoising with a weighted loss to downweight uncertain boundary regions, while CRC performs cross-RoI region-based contrastive learning by inner/outer contour and foreground/background partitioning to refine boundary features. Experiments on CryoNuSeg and DigestPath demonstrate state-of-the-art performance, including robustness with limited labeled data, validating the effectiveness of boundary-denoising and boundary-focused contrastive learning for pathological image segmentation.

Abstract

Semi-supervised segmentation methods have demonstrated promising results in natural scenarios, providing a solution to reduce dependency on manual annotation. However, these methods face significant challenges when directly applied to pathological images due to the subtle color differences between nuclei and tissues, as well as the significant morphological variations among nuclei. Consequently, the generated pseudo-labels often contain much noise, especially at the nuclei boundaries. To address the above problem, this paper proposes a boundary-aware contrastive learning network to denoise the boundary noise in a semi-supervised nuclei segmentation task. The model has two key designs: a low-resolution denoising (LRD) module and a cross-RoI contrastive learning (CRC) module. The LRD improves the smoothness of the nuclei boundary by pseudo-labels denoising, and the CRC enhances the discrimination between foreground and background by boundary feature contrastive learning. We conduct extensive experiments to demonstrate the superiority of our proposed method over existing semi-supervised instance segmentation methods.

Boundary-aware Contrastive Learning for Semi-supervised Nuclei Instance Segmentation

TL;DR

This work tackles semi-supervised nuclei instance segmentation in pathology, where boundary noise arises from subtle nucleus-tissue contrast and diverse nuclear morphology. It introduces Boundary-aware Contrastive Learning for Semi-supervised nuclei segmentation (BASS), comprising a low-resolution denoising (LRD) head to smooth boundaries and a cross-RoI contrastive learning (CRC) module to enhance boundary feature discrimination within a teacher-student framework. LRD provides coarse boundary denoising with a weighted loss to downweight uncertain boundary regions, while CRC performs cross-RoI region-based contrastive learning by inner/outer contour and foreground/background partitioning to refine boundary features. Experiments on CryoNuSeg and DigestPath demonstrate state-of-the-art performance, including robustness with limited labeled data, validating the effectiveness of boundary-denoising and boundary-focused contrastive learning for pathological image segmentation.

Abstract

Semi-supervised segmentation methods have demonstrated promising results in natural scenarios, providing a solution to reduce dependency on manual annotation. However, these methods face significant challenges when directly applied to pathological images due to the subtle color differences between nuclei and tissues, as well as the significant morphological variations among nuclei. Consequently, the generated pseudo-labels often contain much noise, especially at the nuclei boundaries. To address the above problem, this paper proposes a boundary-aware contrastive learning network to denoise the boundary noise in a semi-supervised nuclei segmentation task. The model has two key designs: a low-resolution denoising (LRD) module and a cross-RoI contrastive learning (CRC) module. The LRD improves the smoothness of the nuclei boundary by pseudo-labels denoising, and the CRC enhances the discrimination between foreground and background by boundary feature contrastive learning. We conduct extensive experiments to demonstrate the superiority of our proposed method over existing semi-supervised instance segmentation methods.
Paper Structure (15 sections, 7 equations, 3 figures, 6 tables)

This paper contains 15 sections, 7 equations, 3 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Framework Overview
  4. Low-resolution Denoising Module
  5. Cross-RoI Contrastive Learning
  6. Experiments
  7. Datasets
  8. Implementation Details and Evaluation Metrics
  9. Comparison with the State of the Art Methods
  10. Ablation Studies
  11. Conclusions
  12. Supplement Materials
  13. Data Split
  14. Segmentation head visualization experiments
  15. Comparison experiments

Figures (3)

  • Figure 1: The framework of our semi-supervised nuclei segmentation method. (a) The training flow of our BASS. First, the teacher model generates pseudo-labels, and then the student model is used to train the nuclei segmentation network. (b) and (c) is the proposed low-resolution denoising module and cross-RoI contrastive learning module.
  • Figure 2: The semi-supervised instance segmentation visualization comparisons.
  • Figure 3: The feature map of prediction. (a) represents naive mask head (NMH); (b) represents cross-RoI contrastive learning head (CRC).