Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Self-Paced Sample Selection for Barely-Supervised Medical Image Segmentation

Junming Su, Zhiqiang Shen, Peng Cao, Jinzhu Yang, Osmar R. Zaiane

TL;DR

Barely-supervised medical image segmentation suffers from severely limited pixel-level labels and noisy registration-based pseudo labels. The authors propose SPSS, a self-paced framework combining Self-Paced Uncertainty Sample Selection (SU) and Self-Paced Bidirectional Feature Contrastive Learning (SC) to improve pseudo-label quality in both image and feature spaces. SU creates a voxel-level self-paced mask to select high-confidence voxels, while SC enforces class-discriminative representations through bidirectional contrastive losses, all trained in a self-paced loop. On LA and KiTS, SPSS achieves state-of-the-art performance with substantially fewer labeled slices and is demonstrated to outperform both barely-supervised and SSL baselines; the authors also release their code.

Abstract

The existing barely-supervised medical image segmentation (BSS) methods, adopting a registration-segmentation paradigm, aim to learn from data with very few annotations to mitigate the extreme label scarcity problem. However, this paradigm poses a challenge: pseudo-labels generated by image registration come with significant noise. To address this issue, we propose a self-paced sample selection framework (SPSS) for BSS. Specifically, SPSS comprises two main components: 1) self-paced uncertainty sample selection (SU) for explicitly improving the quality of pseudo labels in the image space, and 2) self-paced bidirectional feature contrastive learning (SC) for implicitly improving the quality of pseudo labels through enhancing the separability between class semantics in the feature space. Both SU and SC are trained collaboratively in a self-paced learning manner, ensuring that SPSS can learn from high-quality pseudo labels for BSS. Extensive experiments on two public medical image segmentation datasets demonstrate the effectiveness and superiority of SPSS over the state-of-the-art. Our code is release at https://github.com/SuuuJM/SPSS.

Self-Paced Sample Selection for Barely-Supervised Medical Image Segmentation

TL;DR

Barely-supervised medical image segmentation suffers from severely limited pixel-level labels and noisy registration-based pseudo labels. The authors propose SPSS, a self-paced framework combining Self-Paced Uncertainty Sample Selection (SU) and Self-Paced Bidirectional Feature Contrastive Learning (SC) to improve pseudo-label quality in both image and feature spaces. SU creates a voxel-level self-paced mask to select high-confidence voxels, while SC enforces class-discriminative representations through bidirectional contrastive losses, all trained in a self-paced loop. On LA and KiTS, SPSS achieves state-of-the-art performance with substantially fewer labeled slices and is demonstrated to outperform both barely-supervised and SSL baselines; the authors also release their code.

Abstract

The existing barely-supervised medical image segmentation (BSS) methods, adopting a registration-segmentation paradigm, aim to learn from data with very few annotations to mitigate the extreme label scarcity problem. However, this paradigm poses a challenge: pseudo-labels generated by image registration come with significant noise. To address this issue, we propose a self-paced sample selection framework (SPSS) for BSS. Specifically, SPSS comprises two main components: 1) self-paced uncertainty sample selection (SU) for explicitly improving the quality of pseudo labels in the image space, and 2) self-paced bidirectional feature contrastive learning (SC) for implicitly improving the quality of pseudo labels through enhancing the separability between class semantics in the feature space. Both SU and SC are trained collaboratively in a self-paced learning manner, ensuring that SPSS can learn from high-quality pseudo labels for BSS. Extensive experiments on two public medical image segmentation datasets demonstrate the effectiveness and superiority of SPSS over the state-of-the-art. Our code is release at https://github.com/SuuuJM/SPSS.
Paper Structure (12 sections, 5 equations, 3 figures, 3 tables)

This paper contains 12 sections, 5 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Overview
  4. Self-Paced Uncertainty Sample Selection
  5. Self-Paced Bidirectional Feature Contrastive Learning
  6. Loss Function
  7. Experiments and Results
  8. Comparison with SOTA
  9. Results on LA
  10. Results on KiTS
  11. Ablation Study
  12. Conclusion

Figures (3)

  • Figure 1: Illustration of a) segmentation performance of PLN and SPSS, b) registration noise, and c) qualitative results of MT, PLN, and SPSS. Note that the green region in a) is the performance drop from ground truth caused by registration noise.
  • Figure 2: Overview of the proposed self-paced sample selection framework (SPSS). SPSS includes: 1) a self-paced uncertainty sample selection strategy (SU) for explicitly pseudo labels selection in the image space and 2) a self-paced bidirectional feature contrastive learning scheme (SC) for class semantics discrimination in the feature space.
  • Figure 3: Qualitative results on the LA dataset and KiTS dataset.